Structural Heart: anatomy, repair & mechanics

Try a repair

The procedural workspace follows Inspect anatomy → Plan repair → Test loading → Compare. Apply or discard an unfinished operation, anatomy brush or attachment placement before changing stage or workspace. Current, applied and stale states distinguish the edited plan from an accepted calculation. Compare includes the untouched baseline; loading experiments remain explicit model assumptions.

Open Structural Heart from the main navigation. The initial synthetic Barlow case preserves an untouched baseline with bileaflet prolapse. Choose the P2 flail example to study focal chord rupture, or the reference case for comparison. Primary anatomy dimensions stay visible; supporting tissue/chord fields and Inspect structure expand when needed. Select a leaflet region and papillary anchor in the viewport or their numeric selectors. Preview a neochord and Apply edit; the worker automatically reassesses the case. Recalculate is also available. Shorten or lengthen it and compare again.

Use Duplicate for an independent alternative. Add a generic annuloplasty ring, or create a P2 resection followed by boundary approximation. Choose a loading experiment and recalculate every plan under the same pressures. Synchronized cameras and phase controls keep comparisons aligned. On mobile, Compare shows one canvas at a time with Original and Current plan buttons. Open editor controls jumps to anatomy, repair and loading; Return to valve brings the canvas back into view.

Save retains one complete case in this browser; Export creates a portable JSON file. New case replaces the open unsaved workspace. Import and browser reload discard cached result validity and require deterministic recalculation. The Repair table uses readable operation names, expandable settings and Edit/Remove actions; the sidebar links to that same history rather than duplicating it. Undo/Redo beside the viewport restores applied plan edits; an active trim draft instead uses stroke Undo/Redo. History replay is checked before a change is accepted, so removing or revising an earlier operation cannot silently leave invalid dependent attachments.

Starting case: Barlow disease

The Barlow example represents diffuse redundant, thickened mitral tissue, an enlarged annulus and elongated intact chords with prolapse of both leaflets. Barlow disease does not require chord rupture. The separate P2 flail example disables P2 native chord support; it is not the default Barlow mechanism.

Preset inputs are annular diameter 42 mm, leaflet-length parameter 22 mm, synthetic thickness 2 mm, excess tissue 30% and native chord rest-length increment 10 mm. Flail is Off. Other valves and prescribed loading retain reference values. These are illustrative settings chosen for this reduced model, not measured patient anatomy, diagnostic thresholds or a prescription for repair. The length parameter still passes through the existing excess-tissue and anterior-leaflet geometry factors.

Pressure and intact elongated chords permit representative systolic bileaflet displacement toward the atrium. An original mitral display term adds redundant mid-leaflet billowing: (excess/100) × effective length × 0.55 × sin²(πu) × sin²(πv) × clamp((θ−0.9)/0.6, 0, 1). It vanishes at attachments, free edges and shared scallop seams, so solved free-edge and chord constraints are preserved. This is regional-hinge geometry with an illustrative surface shape, not finite-element reconstruction of local myxomatous tissue bending. Annular disjunction, tissue histology, dynamic annular curling and arrhythmic risk are not inferred. The displayed opening and gap remain geometric surrogates. The separate suspected categories below use original guarded rules; they do not establish clinical severity.

Geometry and identities

The model uses indexed XYZ surfaces in a common left-posterior-superior (LPS) frame, with mm length, N force and mmHg pressure. Original parameterized surfaces represent annuli, six mitral scallop regions, three tricuspid regions, semilunar cusps, two simplified papillary groups per AV valve, chords, roots, sinuses, sinotubular junctions and ventricular context. This is deliberately simplified regional anatomy, not a complete atlas.

Mitral regions within the same leaflet share boundary coordinates and vertex identities. Each region has a 10×10 surface grid. Annular diameter, shape, height, saddle, leaflet length/excess, papillary depth/spacing, chord elongation, material coefficients and aortic orientation alter geometry or constraints. Papillary displacement preserves native chord rest lengths; explicit elongation changes them.

The semilunar cusp height uses L × (offset + cos θ) × u along its local positive z axis, plus annular and curvature terms, before valve orientation is applied. L is effective leaflet length and u runs from attachment to free edge. Offset is 0.22 for aortic and 0.18 for pulmonic cusps; the pulmonary reference length is 14 mm and hinge stiffness is 1, compared with the generic reference stiffness of 5. The lower pulmonary stiffness permits closure under the smaller prescribed pulmonary pressure difference. These are original geometry/compliance choices, not measured cusp proportions, clinical calibration or material laws. AV leaflets retain their separate ventricular-side construction.

P2 flail removes its native chord constraints. Restriction changes hinge stiffness and resting angle (tethering); commissural restriction changes stiffness, resting angle and motion bounds rather than welding a pathological commissure. Perforation and resection remove retained mesh cells. Vegetation adds a structural mass to the view only: infection, embolic risk and lesion inertia are not modeled.

Mitral orientation

A1 faces P1 at the anterolateral end, A2 faces P2 centrally, and A3 faces P3 at the posteromedial end. Both leaflets are numbered in the same lateral-to-medial direction. The anterior leaflet faces the aortic root; papillary support, selection labels and repair attachments use the same geometry. The annular saddle is highest anteriorly and posteriorly. The mesh remains an idealized educational shape, not a patient reconstruction.

What each edit changes

A ring blends the annular diameter and aspect ratio toward device dimensions by 1/(1+relative compliance), and changes annular attachment height while papillary anchors remain fixed. The ring is generic; no manufacturer device behavior is asserted.

Neochords add tension-only constraints at explicit leaflet coordinates and papillary anchors. Existing chords may be detached, relocated or assigned a new rest length. Removed native attachments remain identifiable but disabled. A neochord attached to removed or perforated tissue is rejected.

P2 triangular removal is quantized to retained 10×10 mesh cells and changes pressure-loaded area and effective hinge stiffness. Boundary approximation contracts the retained cut margins geometrically, with zero displacement at adjacent scallop seams. This is a geometric closure constraint, not a sutural-force or watertight reconstructed-tissue solver. One active resection is supported; revise it rather than stacking cuts.

Thinning retains the mesh and reduces effective thickness; bending stiffness scales with remaining thickness cubed. A patch restores removed mesh cells in the chosen region with a separate displayed surface and chosen synthetic thickness/relative stiffness. It shares boundary vertices with retained tissue; area-weighted thickness-cubed stiffness enters the regional hinge model. Its geometry follows the reference surface, not an inferred patient patch shape. Patch interfaces, suture stresses, tissue failure and healing are not solved. Anatomy edits establish the plan geometry before explicit repair constraints are replayed; later operations are revalidated when an earlier edit changes.

Clips, sutures and pledgets

A mitral clip bridges anterior and posterior regions; a tricuspid clip bridges distinct leaflets. Sutures connect two retained mesh vertices, including opposite sides of a regional cut. The selected anchors use the 0.1 mesh grid. Missing tissue, wrong-valve devices/attachments and reserved identity collisions are rejected. One clip or suture per valve is supported; revise it instead of stacking unmodeled contacts.

A generic clip or suture retains the original single-pair tension-only energy ½k·max(0, separation−target gap)². Tension is k·max(0, separation−target gap), in N. Named G4 profiles instead distribute that same total synthetic stiffness over nine paired material samples, each using k/9 and the same target gap. Three samples span the arm width and three span its full length from the chosen distal grasp toward the attachment. Local annular-ellipse arc length converts width to surface coordinates; opposing tangents align the two footprints. These coordinate choices and finite springs are original approximations, not manufacturer mechanics.

All mesh cells covered by a named footprint must retain tissue. A footprint extending outside one selected region or across missing tissue is rejected. This is a coarse geometry check, not proof of leaflet insertion, capture, adequate tissue or device eligibility. Regional rotations respond to pressure and the sampled constraints; local tissue folding/contact is not solved or forcibly closed. Residual gaps may remain large, and a larger profile is not assumed better. Same-region sutures have limited effect because both bites share one hinge.

Gap reports the largest paired-sample separation. Tension reports the largest individual spring force, not total implant load or tissue stress. Named devices are drawn as two ribbons using the published arm dimensions plus an original 3 mm center marker. They are not engineering CAD or a delivery/deployment simulation. Capture success, safety, durability, device sizing and clinically predictive regurgitation are not modeled; the separate flow summary remains an assumed-flow geometric estimate.

Pledgets add two pads with explicit width and thickness at suture bites. They show a finite support footprint; this model does not predict contact stress or increase suture holding strength. Published pledget contact work in myocardium does not validate these leaflet mechanics.

MitraClip and TriClip G4 profiles

Choose the mitral or tricuspid valve, then a G4 size profile or the generic bridge. Expand grasp positions and constraint settings for precise placement or synthetic gap/stiffness changes. These are schematic representations using published Abbott arm dimensions. The choices do not establish suitability, successful capture, residual regurgitation or a recommended device. The simulator does not reproduce a delivery system or the complete implant.

Arm length is the published maximum arm/coaptation length; width is the arm width. Neither is a measured tissue-grasp depth, achievable coaptation area or a patient-specific safety limit. G4 profiles are identified explicitly; no claim about the latest device generation is made.

Mouse tools, repair previews and view controls

Select (V), Pan (H) and Rotate (R) are separate navigation modes. Select picks an editable structure without turning an ordinary drag into camera rotation. Pan translates the view; Rotate changes its orientation. Holding Space or dragging with the middle mouse button temporarily pans without producing a repair gesture. Wheel or pinch zooms; arrows rotate, Shift plus arrows pans, and +/− changes zoom while the canvas is focused. View retains numeric zoom, yaw, tilt and roll. Center anatomy clears only the pan offset; Reset view restores orientation, zoom and pan. Comparison cameras stay synchronized. The Original pane permits selection and navigation but never starts a repair, even when a repair tool is active. Pan is stored as fractions of the shorter viewport dimension, bounded to ±1.5 in each direction; absent values in older cases mean zero. It changes the projection and picking coordinates, not anatomy, loading, pressure, operation history or the mechanical solve. Play uses elapsed-time animation and interpolates adjacent solved hinge angles, including the cycle boundary, to display smooth motion at the prescribed heart rate. Both views use the same display phase. This adds display positions, not new mechanical solutions. Numeric pressures and metrics, the phase slider and the paused state use the nearest solved frame (24 per cycle by default). Playback suspends its clock when the page is hidden and is unavailable while unsolved. Geometry-editing tools pause playback.

Draw creates surface-anchored markings; Measure displays endpoint distance in mm. Markings follow the same surface points through the cycle and are saved with the synthetic case. Undo marking removes the latest drawing. They are annotations, not tissue edits.

Trim brush builds a visible draft on one leaflet region. Paint adds cells to the proposed cut; Erase removes cells from that draft. Stroke Undo/Redo corrects the selection, Apply commits it and Cancel discards it. Reopen an applied brush operation from the Repair table to revise its cells. Fine, Wide and Broad set the brush footprint. Annular attachments and regional seams are protected; invalid cuts or cuts through retained repair attachments are rejected. Cuts remain discrete 10×10 regional mesh edits, not arbitrary smooth surgical cuts or greater anatomical precision.

Direct repair tools collect retained surface attachments and create a preview of the existing validated operation, rather than freely deforming the tissue or assuming a successful result. Neochord clicks a retained mitral leaflet attachment and previews the existing chosen papillary anchor; clicking a papillary anchor before or after placement changes that anchor. Unsupported attachment locations are rejected rather than silently relocated. Clip clicks two opposing leaflet points on the mitral or tricuspid valve; Suture clicks two retained tissue points on the same valve, with drag placement also available. The first point and prospective second point are marked on the canvas. An invalid second point leaves the first point available for correction. Ring clicks a mitral annulus or leaflet to preview the current generic ring settings. These tools retain their existing valve, attachment, footprint and dependency limits.

Clip and suture attachment picks remain on the original 0.1 regional grid; neochords retain their existing bounded regional coordinates and separate papillary anchors. Edge trim previews bounded free-edge removal. Adjust chord drags a placed neochord or its highlighted handle vertically to preview an 8–60 mm rest length. Native chords remain editable through the numeric Repair controls, without a direct-drag handle. Apply records one accepted operation or revision and reassesses force and geometry through the same solver; no pointer motion itself becomes a solved clinical result. The context controls beside the viewport and the Repair tab share the procedure parameters, with numeric inputs retained for precision and keyboard use.

A clip can also be dragged directly between opposing retained leaflets. Gold attachment handles reposition a neochord, clip or suture preview on supported retained tissue. Matching attachment buttons support arrow-key nudging, and Enter applies the valid preview. Invalid drops and interrupted drags restore the preceding preview; Escape cancels unapplied work. Right-click tissue can seed a repair at that location. Right-click applied repair hardware opens Edit or Remove: accepting a revision replaces that operation rather than appending a duplicate, and removal uses the same dependency-safe, reversible history. These tools edit the procedural geometry; they do not change original scan pixels or establish mechanical success.

Escape cancels unapplied work and returns to Select. Interrupted drags do not commit an operation; an interrupted chord adjustment discards its preview. Undo cancels an unapplied point or repair preview before traversing the accepted plan; trim-draft Undo instead removes the most recent paint or erase stroke. Redo is unavailable while a point or repair preview is pending, and an accepted new edit clears the redo branch. Switching cases or plans clears temporary draft and editor histories. Canvas shortcuts leave typing fields, content-editable text and composition input to their native behavior.

Valve options at the point of work

Right-click retained valve anatomy to open its repair and view menu. With the canvas focused, Shift plus F10 or the Context Menu key opens options for the selected visible structure. The three-dot button beside the viewport zoom controls also opens the menu, including on a touch screen. Empty canvas and nonvalve context retain the normal browser context menu. Escape, Tab or a click outside closes the menu; arrow keys, Home and End move among its actions, and keyboard/action dismissal returns focus to the canvas. Outside clicks keep the clicked control available for native focus.

The menu exposes the existing mitral neochord and ring tools, AV edge-to-edge clips, and supported suture and trim tools on the other valves. Choosing a repair pauses motion and selects its existing placement mode. Retained-tissue validation, Preview, Apply, solver reassessment and operation history remain unchanged; opening a menu or choosing a tool never applies a repair. Open repair controls starts with a supported suture procedure on the clicked valve, avoiding stale parameters from another valve. Original-pane menus contain view actions only. Select, Rotate, Pan, Focus, Hide, Show all valves, Reset view and baseline comparison share the existing view state. Hidden valves have an explicit Show all valves button, and hiding changes only display visibility.

Synthetic ECG and R-wave-gated playback

The first complete, current valid solution starts playing automatically. A user Pause, phase scrub or editing action suppresses that initial automatic start; recalculation does not silently restart a paused cycle. Stale, failed, partial or nonfinite solutions and pending repair drafts cannot play. Both comparison views and the ECG cursor share the same elapsed-time clock and prescribed heart rate. Scrubbing pauses at the chosen solved frame and moves the ECG cursor to exactly the same phase. A hidden page suspends the clock rather than counting hidden time.

The visible strip samples the existing synthetic sinus lead-II voltage primitive. Its R component is shifted from the source phase 0.17 to structural phase 0, the beginning of the existing prescribed ventricular pressure pulse; every R-to-R interval is 60/heart-rate seconds. This is an original display convention for the regular prescribed-pressure cycle, not a measured electromechanical delay. ASE/EAE guidance supports a clear R wave for ECG-gated 3D acquisition, but does not validate this teaching alignment or any simulated repair. No patient recording, arrhythmia acquisition, ECG import or independent ECG timer is supplied.

Display publication is bounded to 30 frames per second while the elapsed clock remains refresh-rate independent. Adjacent solved hinge angles are interpolated only for display; numeric readouts still use the nearest solved frame. Mechanical validity checks are memoized by anatomy, loading, solver settings, echo entries, plan and result references; camera and phase changes cannot make a stale result current. Canvas backing dimensions change only when size or device pixel ratio changes. These are rendering optimizations, not new physiological solutions.

Chamber context and axes

Atrial and ventricular envelopes can be shown or hidden independently. The mesh, structural axes and labels are controlled in View. Reference axes identify LV/RV long axes, the mitral short axes and aortic–mitral orientation in the shared LPS frame. These are synthetic landmarks, not registered patient axes.

Enter width, depth, long-axis height and volume for each chamber. With Fit volume enabled, dimensions are uniformly scaled by the cube root of target volume divided by ellipsoid volume; aspect ratios are preserved. With it disabled, entered dimensions are used and volume is π/6 × width × depth × height, converted from mm³ to mL. Chamber envelopes remain reference context, not solved dynamic chamber volumes, material meshes or exact anatomical reconstructions. These manual context-envelope entries do not themselves modify valve loading or papillary geometry. The separate Echo & ventricular function inputs use the explicitly documented sensitivity mappings below.

Whole-heart display and supporting structures

The outer heart, septal context, LVOT/RVOT and left/right atrial appendages are original schematic surfaces positioned from the current synthetic annuli and chamber envelopes. They are not patient anatomy or registered to a scan. Display controls can hide context structures while preserving the valve workspace, including independent LVOT and RVOT visibility. The outflow tracts connect schematic ventricular regions to their own valve locations; they are not measured luminal casts or computational flow domains.

Fibrous context identifies the aorto-mitral continuity, left/right trigones and central septal attachment. Annular guides do not imply complete solid fibrous rings: mitral fibrous tissue may be discontinuous and most of the tricuspid annular support is fibroadipose rather than a continuous fibrous ring. Pulmonary support is muscular and remains separate from the trigone/curtain system.

The outer envelope is a closed radial support surface around current chamber points with a 2 mm display margin. LAA is drawn as a narrow-neck lobulated finger and RAA as a broader triangular shape, using the current atrial dimensions. These offsets, shape factors and margins are original display choices. Appendage outlines and the closed outer envelope show approximate relationships only. They do not reproduce individual lobes, wall thickness, coronary anatomy, electrical conduction, trabeculation or a full imaging segmentation. Their mesh dimensions, curves and offsets are original display heuristics; they add no tissue coupling, mechanics or patient-specific operative prediction.

Choose a regurgitation mechanism

Mitral Carpentier type I describes normal leaflet motion with a coaptation defect, such as annular enlargement or perforation. Type II is excessive motion, including prolapse or flail. Type IIIa restricts leaflet motion during opening and closure; type IIIb restricts systolic closure through ventricular/papillary tethering. These describe mechanisms, not severity or a specific repair outcome.

The aortic repair-oriented classification uses type I for normal cusp motion with annular/root enlargement or a tissue defect, type II for cusp prolapse and type III for restricted cusp motion. Aortic type III is not a mitral type IIIb category. Multiple mechanisms may coexist clinically.

Preview mechanism creates a reversible anatomy operation. Applying it resets flail, restriction, fusion, excess, elongation, perforation and vegetation before the selected example is assigned; dimensions not specified by that example remain unchanged. Existing repair history is revalidated. These examples do not classify images, assign severity or establish which procedure will work.

Mitral I sets diameter to at least 42 mm, length 20 mm and thickness 1 mm. II sets diameter to at least 38 mm, length 22 mm, thickness 1.5 mm, excess 25% and elongation 8 mm with intact chords. IIIa uses length 19 mm, thickness 1.8 mm, stiffness 8, restriction 40 and fusion 35. IIIb uses diameter at least 38 mm, length 20 mm, thickness 1 mm, papillary depth 43 mm and spread 24 mm. Other mitral examples use depth 32 mm and spread 17 mm; stiffness is 5 except IIIa. These are original tuning values, not phenotype diagnostic criteria.

Aortic I uses diameter at least 32 mm, length 14 mm, thickness 1 mm and stiffness 5. II uses length 16 mm, thickness 1 mm, stiffness 4 and excess 15%; the noncoronary regional hinge can reach 1.95 radians before the fusion adjustment, while other semilunar hinges stop at π/2. III uses length 13 mm, thickness 1.8 mm, stiffness 8, restriction 40 and fusion 25. The reduced aortic I example changes annular geometry; it does not reproduce every root-dilatation or perforation subtype.

Additional synthetic energy terms distinguish motion phases: mitral IIIa and aortic III target hinge angles 1.04 and 0.95 radians with stiffness weight 8 in both phases. AV systolic tethering adds ½ × stiffness × 10 × tethering severity × retained-area fraction × max(0,θ−1)² only when ventricular pressure exceeds atrial pressure. This unilateral penalty resists motion beyond 1 radian; it never pulls a leaflet already below that angle toward better closure. IIIb adds severity 0.5, bounded to 1. These original potentials are not measured tissue laws or clinical restrictions in degrees.

Enter echo measurements

Optional entries include left/right atrial volume, LV and RV end-diastolic/end-systolic volumes, LVEF, LV diastolic/systolic diameter, septal/posterior wall thickness, left atrial dimension, LVOT/aortic root diameter, RV basal/RVOT diameter, LVOT/RVOT modeled segment lengths and RV end-diastolic/end-systolic area. Volumes use mL, linear dimensions mm and RV areas cm². These are user-entered measurements, not values extracted automatically from a scan.

For paired LV volumes, EF = 100 × (EDV−ESV)/EDV; paired volumes take priority over a separately entered LVEF, and a difference above 5 percentage points is flagged. For paired RV areas, FAC = 100 × (EDA−ESA)/EDA. FAC remains separate from the selected qualitative RV function. End-systolic values cannot exceed their paired end-diastolic values; absent entries remain unspecified.

The resolved chamber envelope records its input source, effective width/depth/length and represented volume. Echo volumes take priority over the chamber editor mode. A supplied diameter fixes one axis; the remaining axes fit the supplied volume while preserving their fallback proportions. Without echo volumes, Fit volume retains the chamber volume target; dimensions-only mode uses the entered width, depth and length and reports their ellipsoid volume as a geometric estimate. One volume or several linear dimensions cannot uniquely reconstruct a chamber. RV area inputs constrain a schematic width–length ellipse and do not become a measured RV volume or FAC-derived stroke volume.

Paired ED/ES values prescribe an original display interpolation: sin²(π × phase / 0.42) during the first 42% of the common cycle, returning to the diastolic reference thereafter. Maximum contraction is at phase 0.21; atrial volumes remain reference values. This synchronized geometry is not a solved myocardial contraction model. Derived axes are bounded to 1–300 mm, and any target-volume mismatch caused by those display bounds is reported instead of silently discarding the target. Closed triangulated cavity envelopes approximate the analytic ellipsoid volume within 0.9% in software checks. That tolerance verifies geometry generation, not patient anatomy.

LVOT/RVOT diameter and modeled segment length change the corresponding tube and curved cavity-to-annulus display corridor. Missing values use the associated annular diameter and an original 18 mm modeled segment length. The overlapping corridor and annular/cutaway display collars are excluded from numeric cavity volume. Enabling LV mesh also enables LVOT; enabling general ventricular context enables both outflows. Anatomy overlay can then hide or show either outflow independently, including in the associated mitral/aortic or tricuspid/pulmonic close-up. Length edits do not change the separate reduced LVOT-clearance screen’s existing 18 mm sampling extent. Aortic root diameter independently scales the sinus envelope. Septal/posterior thickness adds an illustrative outer wall, mirroring the entered side if only one is supplied. These are display mappings, not reconstructed wall distribution, obstruction calculations or material properties; segmented image surfaces stay separate and prescribed pressure loads are retained.

LV segment map and ventricular function

The polar map follows the AHA 17-segment names: anterior at the top, septum to the left, inferior at the bottom and lateral wall to the right. The outer and middle rings contain six basal and six mid-ventricular segments; the inner ring has four apical segments, with the apical cap at the center.

Click a segment to set normal, hypokinetic, akinetic or dyskinetic motion, shown in green, yellow, red or blue. Scores are 1, 2, 3 and 4. The wall-motion score index averages segments 1–16; segment 17 is retained as a location marker/annotation but excluded from this index because routine echocardiographic wall-motion assessment does not score the apical cap. The colors are display choices, not scar maps or predicted coronary territories.

An entered LVEF, RV function category and regional motion do not establish contractility or myocardial viability. Any reduced-model changes to load or papillary geometry are illustrative sensitivity experiments, not measured stress, strain or individualized ventricular mechanics.

How echo inputs affect the reduced model

EF, FAC and wall motion do not determine ventricular pressure. All LV, RV, aortic, pulmonary and atrial pressures honor the entered Loading values. Function inputs affect the separate flow-budget and tethering assumptions. The LV contextual function factor is clamp(0.45 + 0.55 × EF/60, 0.45, 1.1), or 1 − 0.35 × mean regional loss when EF is absent. Regional loss values are 0, 0.35, 0.7 and 1 for normal, hypokinetic, akinetic and dyskinetic motion, averaging segments 1–16. Only the regional-motion fallback without EF/paired volumes uses 5 × this factor as its assumed ejection budget; EF and paired-volume budgets use the volume relationships below. The factor never scales pressure or establishes myocardial contractility.

Qualitative RV function supplies contextual factors 1, 0.85, 0.65 and 0.45 for normal, mild, moderate and severe dysfunction. When no category is entered, FAC gives clamp(0.45 + 0.55 × FAC/35, 0.45, 1.1). This factor informs RV tethering and an optional 5 × factor fallback budget, not RV pressure or an area-derived stroke volume. If the entered loading itself provides no pressure-open portion, the assessment is unavailable rather than inventing flow.

Mitral papillary displacement combines local motion loss, LV enlargement and EF loss. Dilation is clamp((LVEDD−50)/35, 0, 1), or clamp((LVEDV−120)/200, 0, 1) if LVEDD is absent. EF loss is clamp((60−EF)/60, 0, 1). Severity = clamp(0.6 × local loss + 0.25 × dilation + 0.15 × EF loss, 0, 1). The anterolateral teaching group uses AHA segments 1, 6, 7 and 12; the posteromedial group uses 4, 5, 10 and 11. This coarse map is not an individual coronary distribution or imaging-derived papillary position.

At severity s, anterolateral and posteromedial anchors receive local offsets (+4s,+3s,−10s) and (−4s,+3s,−10s) mm. A1/P1 use lateral severity, A3/P3 medial severity and A2/P2 their average; posterior tethering is multiplied by 1.1 and bounded to 1. Offsets move rendered papillary groups and actual chord endpoints together while native chord rest lengths are preserved. Dysfunction does not automatically rupture chords or create flail.

RV tethering severity is clamp(0.75 × max(0,1−RV factor) + 0.25 × clamp((RV basal diameter−42)/40, 0, 1), 0, 1). Two tricuspid anchor groups move (0,−3s,−8s) and (0,+3s,−8s) mm, with the same regional tethering severity. All coefficients, thresholds and mappings in this section are original bounded sensitivity assumptions, not measured patient myocardial mechanics.

Local DICOM images

Open DICOM reference and select a source series. A browser worker parses CT, MR and standard ultrasound Part 10 files. It accepts uncompressed explicit or implicit VR little-endian 8/16-bit monochrome and palette pixels, and unsigned 8-bit RGB, YBR_FULL and YBR_FULL_422 colour. RGB and YBR_FULL can be interleaved or planar; packed YBR_FULL_422 must be interleaved. Standard palette lookup tables map indices to display colours; segmented palettes need conversion. RLE lossless and 8-bit JPEG Baseline images are decoded locally; JPEG decoding requires browser bitmap and offscreen-canvas support. Shared/per-frame spatial and temporal metadata are read when present. Rescale slope/intercept is applied to pixel values; MONOCHROME1 changes the display, not segmentation intensities. Frames are never assigned invented spatial positions.

The reference view supports frame navigation, source-colour or grayscale display, grayscale windowing and two-point measurements. Source colour is preserved by default in both the reference viewer and the segmentation editor. Colour pixels also have a luminance copy (0.299 R + 0.587 G + 0.114 B, in display values 0–255) for intensity-based editing; colour windowing does not change the original RGB pixels. Grayscale window controls are disabled while source colour is displayed. In-plane distance uses row/column Pixel Spacing or a spatial ultrasound calibration region containing both points. Ultrasound region axes must both be centimetres; their deltas are converted to millimetres. Doppler time/velocity axes and conflicting region scales are rejected. A measurement can populate a synthetic anatomy draft for annular diameter, leaflet length or papillary depth, then requires Preview and Apply. It remains a manually interpreted dimension: oblique or foreshortened planes can misrepresent the intended measurement. Missing calibration disables applying it.

Image-derived reconstruction requires at least three calibrated, parallel slices from one physical stack and one identified phase, or an internally consistent stack with phase unknown. A known frame of reference, consistent pixel spacing, dimensions and orientation, and increasing physical slice positions are required. Duplicate planes, gaps above 5 mm and shifted/tilted grids are rejected; resample them in a suitable imaging tool. A cine sequence is never treated as slice depth. Unknown phase is not proof of ECG gating.

Patient-based CT/MR stacks are supported. Standard ultrasound images may be viewed as references; an ordinary 2D cine does not provide a reconstructable 3D valve. JPEG 2000, JPEG-LS, other unsupported compression, big-endian files, unsupported vendor-private 3D/4D echo, SEG and Surface Segmentation objects need conversion before import. A colour Doppler overlay is displayed as acquired; it is not separated into velocity, blood flow or tissue, and its colours cannot establish a tissue segmentation. SlicerHeart documents vendor-specific decoding/export needs; compatible source pixels and physical geometry must be available before this workflow can use them.

Import is bounded to 256 files, 128 MB total, 64 MB per file and 16 million voxels. Parsing is cancellable. Source pixels and identifiers are not uploaded, broadcast to classrooms, saved in browser storage or included in the synthetic case export. Close & clear images releases the reference workspace. Use appropriately de-identified files; local processing is not a de-identification service.

Seed, review and correct tissue

Automatic draft starts On. Select a source slice and click the intended valve tissue. The selected point defines a local crop and intensity seed. This is bounded intensity region growth, not anatomical recognition, a trained classifier or validated valve segmentation. Similar-intensity blood, myocardium, calcium or artifact can be included; thin leaflet tissue may be missed. Review the marked tissue on every slice.

The draft includes six-face-connected voxels within the chosen intensity tolerance of the seed value. Tolerance is in the rescaled image units for monochrome sources, or display luminance (0–255) for colour sources, not a universal tissue threshold. Colour Doppler can enter an intensity-based draft and must be reviewed manually. Changing display mode or grayscale contrast affects visibility only. The source pixels remain unchanged. Replace with draft repeats growth and replaces the manual mask; it does not preserve corrections.

With Automatic draft Off, select a region and paint or erase tissue on its slices, then Build surface from mask. The visible brush footprint uses source pixels, not mm; it is clipped to the selected region. Continuous strokes sample the path between pointer events. Undo mask and Redo mask keep up to 20 completed edits. A new region starts a fresh mask history; a new edit clears redo. Escape, lost pointer capture, changing slice or leaving the focused canvas cancels an unfinished stroke. Manual edits can still create disconnected regions or holes, which need review. Mask edits do not update the 3D surface until rebuilt.

The region radius is 3–25 mm, bounded to at most 97×97 pixels and 64 planes. The crop keeps the actual image spacing, orientation and plane positions. These are performance limits, not minimum image-resolution requirements for clinical accuracy. Worker reconstruction can be cancelled and has a time limit; smaller crops or simpler masks may be needed.

Philips 3D ultrasound volumes and orthogonal inspection

A bounded experimental reader recognizes the Philips private DIN_3D_ECHO layout used by the openly available EchoNet 3D echo sample. It verifies the known private creator blocks, XOR-masked dimensions, frame-offset table and matching 32-byte frame headers. Only the selected acquisition phase is inflated, with a hard decompressed-size limit. This supports the inspected byte layout; it is not a general vendor-volume decoder.

Scan conversion uses the published EchoNet coordinate mapping x = radius × cos(theta) × cos(phi), y = radius × sin(theta), z = radius × cos(theta) × sin(phi), and trilinear interpolation in the radial/angular source grid. The Cartesian preview has at most 128 samples on its longest dimension. Voxel size therefore depends on the field of view; thin leaflet details may be lost. Empty space outside the acquired frustum is zero. Resampling cannot restore unobserved tissue.

These volume coordinates are local ultrasound millimetres, explicitly separate from patient LPS. Phase selection replaces the volume; time frames are never stacked into artificial depth. The three orthogonal acquisition-plane views preserve physical aspect ratio and use real slice depths, including interpolation for nonuniform depths in standard calibrated stacks. Their crosshair is a navigation aid. Colour sources use luminance for these inspection planes. The source acquisition phase and coordinate frame are retained in reconstruction exports.

Integrated 3D/4D image anatomy project

Choose 3D/4D image anatomy in Structural Heart. Import a supported acquisition or try the clearly labelled synthetic three-phase volume. The image project retains immutable source bytes, acquisition geometry and timing separately from anatomy labels, registered surfaces, repair variants and optional computational approximations. The procedural-model workspace remains available independently.

The image editor presents Source → Orient → Trace → Repair → Review. Orient provides linked source and arbitrary oblique planes; Trace retains manual labels and landmarks; Repair edits an independent geometric variant; Review compares the accepted frame and provides geometry reports. Stage changes do not establish annotation review or clinical validity.

The working flow is source inspection → annulus and leaflet annotation → review and correction across acquired frames → surface construction → an independent geometric repair variant → same-frame comparison → Save imaging project. Ordinary procedural Save/Export does not preserve this separate image project. Neither workflow establishes clinical accuracy or patient-specific predictive mechanics.

Oblique source-plane inspection

Three linked planes can be aligned to the source or the annotated annulus and adjusted around their shared world-coordinate crosshair with yaw, pitch and roll. Scalar source intensity uses trilinear sampling; label previews use the nearest voxel. These are resliced acquired samples, not newly detected anatomy. Physical units and coordinate registration come from the accepted acquisition.

Oblique planes support inspection and crosshair navigation; annotation and mask painting remain in the orthogonal editing views. Outside-source samples stay empty. Rotation or a smoother-looking slice cannot restore missing spatial detail or establish a correct boundary. The guided plane and review workflow is informed by official Philips 3DQ/MVN descriptions; this independent educational implementation does not reproduce proprietary QLAB algorithms or claim equivalent accuracy.

Registered geometry measurements and reports

Geometry measurements use the accepted annulus and actual retained triangle faces in known millimetres. Annulus perimeter closes the ordered 3D point polyline. Projected area uses its simple closed polygon in an orthogonal least-squares plane; a crossing projected contour has unknown area. Major and minor spans are extents along geometric principal axes, not anatomical anterior-posterior or commissural axes. Saddle height is the peak-to-peak signed distance from the fitted plane.

Anterior, posterior and total retained leaflet mesh areas sum referenced triangle areas. Unused vertices contribute nothing. Thick or closed label surfaces include both faces and boundaries, so these are mesh areas, not single anatomical leaflet-face areas or projected valve openings. Missing labels, unbuilt or stale surfaces, unknown units and unavailable frames stay unknown; they are not zero. Boundary-clipped, inferred, ambiguous, propagated and unreviewed geometry stays identified.

The acquired-frame table measures original accepted anatomy independently at each source frame, retaining the supplied timing or explicitly unknown timing. No tracking or interpolation fills missing measurements. A selected repair variant is compared only with its own immutable baseline at the same frame. A stale or wrong-frame variant remains unavailable; original anatomy is not silently substituted. Deltas describe geometric edits, not matched-load mechanical outcomes, clinical severity, stress, coaptation or repair success.

Local CSV and JSON reports include source frame indices, acquisition phase and timing, units, source hashes, content dependency stamps, review/provenance, methods and same-frame deltas. Export rechecks the current source and geometry dependencies; edits during measurement or export require a fresh report. Reports contain no source image pixels, but identifying text in a project is not thereby de-identified. An acquisition exceeding four million retained triangles requires a smaller or simpler labelled region for reporting.

Volume capabilities, Philips, GE and Slicer interchange

Capability is established by the actual geometry. A raster image, a 2D cine and a movie of a rendered 3D heart do not supply sampled spatial depth. Standard DICOM stacks need coherent calibrated planes. The recognized Philips private frustum layout supplies a selected acquired volume phase in ultrasound-local millimetres; the original radial/angular layout and Cartesian conversion descriptor remain traceable.

The vendor-neutral bridge accepts attached scalar NRRD version 4/5, three orthogonal spatial axes, and optionally a fourth Slicer list axis for acquired frames. Supported encodings are raw and gzip; samples are signed/unsigned 8/16-bit integers or float32, with explicit endian for multibyte types. Detached files, vectors, RGB volumes, shear and unsupported compression require conversion. Original axis ordering, source affine, display-to-source mapping and sequence index labels are retained.

Unknown units remain unknown. Physical editing and measurements require documented millimetre calibration or an explicit recorded user declaration of source units. Generic metric local coordinates remain UNREGISTERED when patient registration is unavailable. Patient RAS/LAS coordinates are converted explicitly to LPS. A working crop/resampling retains its actual world positions; its visual smoothness does not add source information.

The GE E95/E90 panel now connects to an explicit local Windows Image3dAPI converter, or accepts its converted scalar sequence through Import volume / sequence. It requires a compatible, legitimately installed GE cardiac reader. Native Windows execution and the actual E95/E90 exports remain unverified; reader availability is not dataset validation. SlicerHeart also documents older Kretz and other export paths, which do not establish compatibility with modern cardiac TEE data.

GE Vivid E95 / E90 local volume import

In Source, expand Import GE E95 / E90, download the converter and follow its README on a Windows 64-bit workstation with a legitimately installed GEHC_CARD_US.Image3dFileLoader reader. Run ge_convert.py capabilities, then ge_convert.py serve; enter the printed 127.0.0.1 address and session token in the panel. A local --type-library argument is available when registry type-library discovery is unavailable. The download contains no GE reader DLL or vendor SDK. Actual native E95/E90 datasets and Windows/COM execution have not been verified here; injected synthetic checks establish software behavior only.

The selected original goes only to that explicitly paired loopback service, never a hosted image endpoint. The browser checks its SHA-256 against converted GE provenance, checks the output checksum and receipt, then validates actual NRRD geometry and timing before accepting a project. The saved image project preserves the converted source and original-source fingerprint; keep the original GE export separately. Import does not de-identify image content. Tokens are ephemeral and excluded from URLs and saved projects.

The Cartesian grid cap requests resampled scalar intensities from the reader, not raw ultrasound beams or RF data. All acquired volume frames retain their original order; limits fail instead of silently dropping phases. Bounds are 128 MiB input, 64 MiB output, 256 MiB decoded intensities, 1024 frames and a 16–240 grid cap (160 default). Native reader allocations are outside those host-buffer guarantees. Transducer-relative, right-handed metre geometry becomes millimetres using the returned dimensions and first-to-last direction vectors; patient LPS/RAS orientation is not assigned.

Strictly increasing acquired timestamps remain known time; missing or contradictory timing retains raw values and ordinal frame indices with elapsed time unknown. Optional bounded probe, ECG and display lookup-table metadata remain provenance. No per-frame cardiac phase is inferred from ECG; an intensity-to-colour lookup table is not colour Doppler velocity, flow calibration or RGB source voxels.

Reader, format, authentication, resource and integrity failures retain the last accepted project. Cancel and newer operations prevent late results from replacing it; known local jobs are deleted and temporary data have bounded expiry. A native COM call may finish before cancellation takes effect. Before calling an export compatible, compare first/middle/last phases, frame count/order, timing, orientation and calibrated distances with its native review system. Import alone validates neither segmentation nor repair predictions.

Manual registered anatomy and source review

Anterior and posterior leaflets use separate binary masks. Paint, erase, filled contours and optional seeded intensity growth operate on the displayed physical grid. Region growing is assistance, not anatomical recognition. Mask overlap between the two leaflet labels is rejected or reassigned explicitly; uncertain boundaries can be marked with ambiguous annotations.

Place an ordered annulus contour across the linked acquisition planes. Papillary, visible chordal, commissural and named scallop/other landmarks retain their source-frame world coordinates. Manual completion and unseen structures must be marked inferred; visible image structure alone does not establish complete chordal anatomy. Labels have independent selection, visibility and locking. Completed edits support Undo/Redo; source voxels are unchanged.

The editing grid is bounded to 97×97×64 samples. Whole-volume previews may be resampled; smaller valve-focused regions can retain finer native spacing. Source planes, working-grid positions and source transforms remain associated. Thin leaflet boundaries can be lost at coarse sampling. Surface extraction represents the binary-mask boundary, including artificial closure where a mask reaches the region edge; it is not automatically a leaflet midsurface or a mechanical mesh.

Review is an explicit user assessment against the acquired source. Marking a frame reviewed does not validate its anatomy. Changing physical annotations increments their revision and makes dependent surfaces, measurements, propagation proposals and computational mappings stale. Display visibility and locking do not alter physical anatomy.

Acquired phases, keyframes and propagation proposals

Phase navigation displays actual decoded source frames. The Philips sample has acquired frame identities but no verified acquisition timestamps; display frame rate is not promoted to ECG timing. NRRD indices produce elapsed milliseconds only when an explicit numeric time axis, recognized temporal unit and valid increasing values establish it. DICOM cine timing follows the declared Frame Increment Pointer and supported Frame Time/Frame Time Vector metadata; available trigger information remains distinct.

Playback uses known timing where available, or an explicitly labelled display rate. Lazy decoding and a bounded frame cache may limit playback speed. Interpolated display views never become additional acquired frames. Browser-hidden state pauses playback.

Propagation begins from a reviewed keyframe and a selected target range. A world-coordinate copy, or a proposal between two keyframes using nearest-keyframe masks with interpolated centroids/landmarks, supplies an editable starting point. It is an original geometric convenience, not image registration or reliable tissue tracking. Preview the yellow overlay on each actual target volume, accept or reject it, then correct and review each accepted target. Accepted propagation remains identifiable and unreviewed; Undo restores the previous accepted annotations. Changing an endpoint or target revision makes an outstanding proposal stale.

Repairs on actual image-derived surfaces

Create a named variant after building current registered surfaces. Its baseline is an independent snapshot of the acquired frame and retained geometry. A ring moves the actual annotated annulus and nearby leaflet vertices toward its chosen plane and ellipse. A constructed neochord uses a papillary world point and a retained triangle/barycentric leaflet attachment; requested length changes that local geometric configuration.

Trimming removes selected retained faces, with an explicit open-boundary report. Approximation uses supported retained boundary attachments and bounded local displacement; it does not silently cap, weld or invent missing tissue. Stable face identities prevent removed attachments being silently reassigned. Source anatomy is preserved and the original baseline can be overlaid at the same acquired phase.

Checks reject invalid connectivity, collapsed/nonmanifold triangles, new disconnected regions, unsupported attachments and detected within-surface self-intersections. The intersection search has a bounded candidate budget and may require a smaller/simpler surface. Cross-structure penetration, tissue-force equilibrium, general contact and material failure are not solved. A geometric edit is not a converged mechanical result or evidence of repair success.

The procedural simulator requires a separate explicit approximation. It can fit reduced annular and leaflet parameters from the reviewed contour and retained surface, while keeping thickness, stiffness, chordal completeness and prescribed loads as stated assumptions. Its generated regional mesh is distinct from the registered source mesh; image-derived devices are not transferred as mechanically valid constraints. Store the mapping and review it before opening. Computational comparisons require matching stated loads; altered-load reassessment remains separate.

Portable imaging projects and exact source relinking

Save imaging project creates a versioned binary .cstudio-image container with a bounded JSON manifest and SHA-256-checked binary chunks. Original immutable DICOM/NRRD sources are embedded by default, alongside materialized volume caches, working grids, masks, registered meshes, labels, landmarks, review/provenance, variants, operation histories, measurements, computational assumptions and view settings. Large images do not enter localStorage.

If original sources are intentionally excluded, the project records their hashes and lengths and reopens as unresolved until the selected files match exactly. A thumbnail, object URL or cache alone cannot satisfy relinking. Package import validates chunk ranges, sizes, types, hashes, source associations and replayable repair history; cached calculation validity is not trusted.

Import and export remain local. Source originals may contain patient identifiers and acquisition metadata; importing does not establish de-identification. Excluding originals does not make voxel caches or derived geometry anonymous. The optional filename setting controls displayed filenames, not metadata inside an embedded original. Save before clearing the project from the page.

Verification, resource bounds and SlicerHeart scope

New software checks use asymmetric anisotropic oblique synthetic volumes and distinct authored phases, plus the public recognized Philips sample. They test axis/sample ordering, source/world registration, masks and history, propagation decisions and stale states, actual surface repair, independent variants, binary save/reopen, corrupted inputs, cancellation and exact source relinking. The synthetic mitral-like fixture is a verification exercise, not clinical anatomy.

Expensive decoding, working-grid generation, surface building, propagation and repair checking run in cancellable browser workers. Encoded sources are bounded to 64 MB each and 128 MB per import; one decoded phase is limited to 16 million voxels. Working anatomy and a small decoded cache have separate bounds. A portable package has a 512 MB cap and binary allocation limits. Superseded work cannot replace a newer accepted project.

Native WebGL 2 renders a bounded 3D volume texture, while source planes and registered surfaces remain available when graphics support is unavailable. The current browser editor is a distinct implementation, not a claim of complete SlicerHeart parity. Advanced SlicerHeart modules, vendor readers and their platform/software dependencies must be reported separately and verified through their actual engine. Authorized Philips and GE TEE exports, independent segmentations and clinical/mechanical validation remain external needs.

Optional local SlicerHeart companion

The companion connects only after an explicit local address and session token are entered. The browser client accepts the IPv4 loopback address, keeps credentials in memory, omits cookies and follows no redirects. It reports the running Slicer version and actual loaded module identifiers; an installed module, a successful module opening or a passing protocol test does not establish a validated native workflow.

Explicit handoffs send attached raw/gzip NRRD to that local session. Generated working volumes retain their accepted affine geometry; label masks become an unreviewed segmentation. Local or unregistered ultrasound coordinates stay identified as such. Annulus handoff requires patient LPS registration, known millimetres, a successfully imported matching working volume and the native valve module. Its one-frame working volume uses native frame index zero while the browser retains the originating source phase.

Native jobs report queued, running, succeeded, failed or cancelled status. Only queued jobs can be cancelled by the companion. Stopping a browser wait or disconnecting does not stop a running native operation; review and save the Slicer session locally. No native mechanics result, manufacturer device model or successful repair claim is substituted into the browser procedural solver. Native extension availability and end-to-end validation remain separate requirements.

Estimate a procedural valve from limited imaging

Estimate procedural valve fits the existing editable mitral model to three image landmarks: two annular hinges and one leaflet tip. Suggestions are deterministic intensity-informed starting points, not an anatomical detector; they can be repositioned and explicitly marked reviewed. Spatial DICOM calibration supplies scale when all landmarks share a calibrated region. Otherwise an entered annular diameter supplies scale, or the disclosed 34 mm synthetic reference prior is used.

Annular diameter and leaflet span are bounded to the existing anatomy ranges. A reviewed, calibrated tissue surface can supply physical bounds for the diameter and annular aspect-ratio guess, but a tissue bounding box is not a traced annulus or a leaflet segmentation. The displayed exploration ranges express original sensitivity choices, not statistical confidence intervals or validated reconstruction error. Saddle height, unseen leaflet shape, chords, papillary anatomy, thickness and material stiffness retain stated shape assumptions or model priors unless adjusted.

Use estimate in procedural simulator creates an image-informed parameter case with an untouched estimated baseline and a fresh repair plan. The actual procedural valve mesh is rebuilt from the fitted parameters, so existing clips, rings, neochords, trimming, comparisons and altered-load reassessment can be explored. The fitted model is a geometric hypothesis, not the original segmented mesh, and the pressure/hinge solver is not calibrated to that individual. Do not interpret its residual grades, stresses or repair behavior as patient predictions.

The source modality, frame, acquisition phase, coordinate frame, landmark review state, parameter origins, exploration ranges and assumptions travel with Save and Export. Source pixels and patient identifiers are excluded. Reimport validates the estimate and clears cached results for recalculation. Unknown anatomy stays labelled inferred; applying a repair does not improve the evidence behind the starting reconstruction.

LV and LA mesh context

LV mesh and LA mesh independently display transparent chamber shells around the active and reference mitral valves. The chamber necks share the actual ordered mitral annular boundary; an anterior cutaway keeps leaflet and chordal editing visible. The shell body follows the resolved supplied chamber envelope, with an explicitly procedural annular collar rather than a newly enlarged half-ellipsoid. The LVOT is shown with the LV mesh. These are anatomical display estimates, not segmented chamber walls; collar/cutaway and outflow additions are not included in the numeric cavity volume.

The shells cannot be selected as repair tissue. Their visibility persists with the case and does not change loading, chamber volumes, valve geometry, repair constraints, calculation keys or results. Newly fitted image-informed cases show both shells by default. Synthetic cases can turn each one on from the procedural view.

Image-derived surface and editing

The image model extracts the 0.5 boundary of a binary tissue mask with six tetrahedra per grid cell and shared edge vertices. Coordinates use the declared DICOM or ultrasound-local transform: image position plus column index × column spacing × row direction, plus row index × row spacing × column direction. Actual slice positions supply depth. Lengths remain in mm and coordinates retain their declared LPS or ultrasound-local frame; the source is not replaced by a synthetic valve template.

The reconstructed boundary is a closed shell around selected tissue, not automatically a leaflet midsurface or a mechanical volume mesh. Exterior zeros close the mask at the crop border. If selected tissue reaches that border, a warning identifies the artificial closure: expand the crop or correct the mask. The midpoint surface and apparent thickness depend on voxel resolution and segmentation; they are not measured tissue boundaries. A maximum of 80,000 triangles bounds the prototype.

Rotate, wheel/pinch zoom, Drag tissue and Trim surface edit a separate geometry copy. A visible footprint shows the local brush; surface radius uses mm and Shape strength adjusts displacement. Trim follows face-connected tissue rather than removing a nearby disconnected sheet. Drag tissue spreads a smooth displacement over connected mesh edges, with a bounded step and local triangle-orientation checks. These are editing safeguards, not physical elasticity, self-intersection proof or a biomechanical repair calculation. Trimming can still open the shell, and edited geometry can be implausible.

Export surface saves local JSON containing the mask, physical grid, modality/phase and original/edited surfaces. Export is blocked during reconstruction and whenever the mask differs from the built surface; rebuild first. Undoing back to the exact built mask restores consistency. STL review also excludes unfinished geometry strokes and stale surface sources. It does not export source pixels or patient identifiers, and it is separate from the synthetic case format. Source-derived geometry can still be sensitive. No valve labels, chords, unloaded shape, material properties, motion, repair success, regurgitation or hemodynamic response are inferred from this surface. The synthetic pressure/hinge solver does not run directly on it. Fitting a procedural estimate uses its geometric bounds and stated priors to build a separate editable parametric valve.

Image-editor controls and limits

Original outline shows the unchanged reconstructed surface. Surface Undo/Redo keeps up to 20 completed edits; Reset surface returns to the reconstructed original and can itself be undone. These temporary surface and mask histories are separate and are not exported. Escape or a cancelled pointer stroke restores its starting geometry. Rebuilding from the mask creates a new original and replaces surface edits. Source pixels and the segmentation are not changed by surface editing.

With the surface canvas focused, V rotates, G selects Drag tissue, T selects Trim surface, and [ / ] changes radius. Ctrl/Command Z undoes; Ctrl/Command Shift Z or Ctrl/Command Y redoes. With the source-slice canvas focused, B paints, E erases and [ / ] changes the pixel brush radius; the same undo, redo and cancellation shortcuts apply. Shortcuts ignore text fields and editable text.

The shared editing.ts helper uses frontmost triangle picking, bounded stroke sampling, connected face/edge neighborhoods and finite geometry checks. Shape displacement has a smooth radial falloff and is limited to the smaller of 5 mm or half the brush radius before strength is applied. A local triangle-orientation check repeatedly halves a rejected displacement. The step limit and preservation checks are original numerical safeguards; they do not preserve tissue volume, prove mesh quality or establish clinical anatomy.

Export image geometry to CAD or a slicer

After reviewing the tissue mask and editing its reconstructed surface, open 3D Print / CAD. Download STL (mm) exports the current edited triangles; Download print manifest records mm, the declared LPS or ultrasound-local frame, 1:1 scale, dimensions and mesh checks. The original reconstruction remains unchanged. STL has no standardized units field, so choose millimetres at import. This is a triangle surface, not a parametric CAD or STEP model.

Export is disabled while reconstructing or editing, after unreconstructed mask changes, or when selected tissue reaches the crop boundary. Expand or correct the mask and rebuild instead of treating its artificial crop closure as anatomy. Empty, invalid, open, nonmanifold, degenerate, duplicate or inconsistently wound geometry also blocks STL export. Checks include coordinate precision and zero enclosed volume.

Separate closed parts and inward-wound parts are reported for review; the exporter does not silently weld or repair them. These checks do not detect all self-intersections, inadequate wall thickness, nesting errors or manufacturing problems. Inspect segmentation against source images and the final mesh in CAD or a slicer. Export carries geometry only, not patient mechanics, clinical suitability or a validated implant design.

Reduced mechanics and contact

The regional-hinge-v1 backend minimizes a common quasi-static energy with one rotation per leaflet region. Components are hinge bending ½k(θ−θ₀)², symmetric neighboring-angle coupling, approximate pressure work ΔP·A·L·0.42·cos(θ) and tension-only chord energy ½kc·max(0,l−l₀)². Pressure work samples continuous regional angles. 1 mmHg = 0.000133322 N/mm². Dimensions, hinge coefficients, coupling and the 0.42 lever factor are original illustrative choices, not fitted clinical measurements.

Same-leaflet boundary interpolation couples adjacent regions, and all affected chord energies enter each coordinate minimization. Trimming and resection scale pressure area by the removed mesh-cell fraction and hinge stiffness by 1−0.25×that fraction. These are reduced-order assumptions rather than shell constitutive laws.

Contact is a unilateral central exclusion envelope: leaflet radial reach is limited to 0.15 mm from the valve axis. It prevents crossing the central axis but does not provide general leaflet-to-leaflet collision, contact stress or an excessive-penetration guarantee. General penetration is marked unavailable, never reported as zero. Free-edge proximity is measured separately in 3D.

Deterministic bounded coordinate minimization uses 28 golden-section steps per coordinate and repeats until the greatest angular update is below tolerance (default 0.0005 rad, maximum 80 sweeps). Convergence is a numerical stopping condition, not evidence of anatomical accuracy or a global minimum. Failed calculations provide no current frames. Previous results remain stored as stale; the viewport uses edited reference geometry until recalculated.

Prescribed pressure and coupling boundaries

LA and RA means are prescribed constant loads. LV and RV pressures use a squared-sine systolic pulse lasting 42% of the cycle, between the entered diastolic and systolic values. Echo EF, FAC and wall-motion inputs do not rescale these pressures. Arterial pulses lag by 4.5% of a cycle, allowing the early systolic opening gradient. The default solve uses 24 phases. Heart rate sets displayed cycle duration; this quasi-static solver has no inertial rate dependence.

Each comparison and alternative uses the same loading inputs. Baseline, higher afterload, higher filling pressures, low pressure and pulmonary-pressure challenges are available; custom values allow repeated load sweeps. Inputs are prescribed, not the result of solved circulation. A separate educational assessment uses solved geometry and an explicit ejection budget to estimate relative flow burden. It does not replace the prescribed pressures or infer effective orifice area, tissue stress or strain.

The aortic and mitral structures share coordinates. Aortic orientation and root dimensions affect the synthetic LVOT clearance measurement; there is no myocardial deformation coupling or complete aorto-mitral continuity tissue model. Annular and papillary positions are fixed within each pressure experiment; anatomy edits and the documented echo-derived offsets change them between experiments. Time-varying boundary motion and a dynamically coupled circulation are future extensions. The shared-flow assessment below conserves an assumed serial flow budget without solving chamber pressure or volume feedback. No wedge-to-LA adapter is currently used.

Flow and acceleration: what is estimated

The colour-flow overlay is an original moving illustration attached to the displayed valve geometry. Forward flow follows filling for mitral/tricuspid valves and ejection for aortic/pulmonic valves; residual reverse flow uses the opposite phase. Read its colour convention and limits below. Numerical flow estimates come from the separate geometry/budget calculation, not the displayed jet pixels.

The assessment uses the complete solved cycle, with 24 phases by default. A pristine synthetic reference is solved with the same entered loading and solver settings but without echo-derived tethering; it is distinct from the untouched pathological baseline used to compare repairs. At each valve, positive prescribed upstream-minus-downstream pressure defines the opening phase. An ideal-orifice capacity sample is 6 × projected opening in cm² × √(positive pressure difference/4), in L/min. This is a dimensional Bernoulli/continuity illustration, averaged across the complete cycle; no vena-contracta contraction coefficient or pressure recovery is solved.

Closing-phase leakage uses a geometric excess relative to the pristine same-phase reference without echo-derived tethering. Subtract reference uncovered fraction × current annular area and a 2% current annular-area allowance from the projected opening. Compare this with excess free-edge gap above the larger of reference gap or contact tolerance, multiplied by the representative closure span and divided by 100 to give cm². The larger nonnegative excess is capped at annular area. The span is the mitral diameter, or three mean annular radii for a three-region valve. These are original regularizers for coarse geometry; the result is not EROA, vena contracta or a clinical regurgitant orifice. Closed-phase summary values are weighted by √(closing pressure difference/4).

Each valve receives a reverse-to-forward capacity ratio k. The forward denominator has a 0.01 L/min floor; k is bounded from 0 to 9 and saturation is flagged. These numerical bounds prevent unstable estimates; they are not physiologic limits. The model uses the ratio to apportion the assumed flow budget, rather than treating its raw geometric capacity as a measured output ceiling.

The displayed velocity ratio compares Q/A with the pristine synthetic reference: (current forward flow / gross budget) × (reference open fraction × reference mean open area) / (current open fraction × current mean open area). The relative gradient index is its square, following the proportional relationship in simplified Bernoulli. These dimensionless ratios do not report Doppler velocity, a mean gradient in mmHg or a catheter pressure difference. Geometric and effective opening differ; mean clinical gradient averages instantaneous squared velocities. A zero budget makes both ratios unavailable, and an unresolved opening makes the assessment unavailable.

Read the colour-flow illustration

The camera acts as a virtual probe: the dominant red hue indicates motion toward it and blue away. Rotating the view can therefore change the colour of the same anatomical flow. This adopts the common Doppler colour convention, not an oxygenation, disease or severity code. Real ultrasound colour maps can be configured differently.

Real Doppler jet appearance depends on gain, velocity scale, beam alignment, pressure/loading and receiving-chamber constraints. Wall-hugging or eccentric jets can appear deceptively small. Clinical regurgitation grading integrates multiple measurements and anatomy; a jet image alone is insufficient. Stenosis assessment also considers flow, valve morphology and quantitative measurements, because velocity and gradient change with flow.

Here, the colour field, moving texture and sparse contrasting mosaic accents are generated artwork. A bounded tube follows the geometric path, brightens around a schematic narrowing and fades when viewed across its axis. The texture is deterministic and repeats with cardiac phase. It is clipped by the selected view and opaque tissue. These choices do not result from an ultrasound beam, sampled blood velocities, measured Nyquist limit, actual aliasing, turbulence calculation or CFD/FSI. The drawing cannot supply a vena contracta, PISA, EROA, regurgitant volume or clinical grade. The separate suspected-grade categories use explicit synthetic rules, not pixel colour, brightness or jet area.

How the four valves interact

The assumed ejection budget B defaults to 5 L/min and may be zero. When supplied from LV end-diastolic and end-systolic volumes, B = (LVEDV − LVESV) × heart rate / 1000. This is a gross LV ejection rate, not forward cardiac output in the presence of regurgitation. Exact paired LV volumes are not reduced again by EF or RV multipliers. Otherwise EF uses entered/contextual EDV × EF/100 × heart rate/1000; regional motion without EF uses a 5 L/min reference multiplied by the LV contextual function factor; absent function data use the editable 5 L/min default. An available RV contextual budget of 5 × RV contextual function factor may cap only these assumed fallback budgets. This RV cap never overwrites the paired-LV-volume total and is not measured RV output. The final common budget bounds both ventricular turnovers in the flow illustration.

Using the four reverse-to-forward ratios, common net flow F = B / [1 + max(kmitral + kaortic, ktricuspid + kpulmonic)]. For each valve, gross forward flow is F × (1+k) and reverse flow is F × k, leaving the same net F at all four valves. Left ventricular turnover is F plus mitral and aortic recirculation; right ventricular turnover is F plus tricuspid and pulmonic recirculation. The opposite side is assumed to adjust to the serial-flow limit. No shunt, transient chamber storage or unequal sustained pulmonary/systemic net flow is introduced.

This bookkeeping illustrates why a regurgitant lesion can reduce forward perfusion, while mixed stenosis/regurgitation changes the gross flow passing through each valve. Repair of one valve may change the estimated velocity and gradient at another without changing its geometry. It is not direct tissue coupling, remodeling or a clinical prediction of multivalve repair. Prescribed pressures do not automatically normalize after an edit.

Read the repair comparison

The summary evaluates mitral, aortic, tricuspid and pulmonic valves against the untouched baseline under the same inputs. Read opening/narrowing, residual leak proxy, edge proximity and flow burden together. The overall statement describes a comparative geometric tradeoff, not surgical success, durability, a treatment recommendation or a probability of benefit.

A smaller residual gap can coexist with reduced forward opening. Reduced leakage can lower recirculating flow and thus lower a gradient even if an opening has narrowed; conversely a small opening can raise the velocity index. Clinical post-repair evaluation integrates regurgitation, valve gradients and multiple imaging/hemodynamic findings. The prototype table cannot substitute for that assessment.

The reduced-opening flag means mean projected area below 80% of the pristine reference. A residual-gap flag means the weighted leak proxy exceeds 0.5% of annular area. Free-edge separation is flagged above the selected contact tolerance. These are original geometric conventions, not clinical cutoffs. Changes above 5% of baseline opening, 0.5% of annular area for leakage or 0.1 mm of gap inform the comparative statement. Favorable geometry, Trade-off and Needs revision describe only these rules.

Coaptation here means sampled free-edge proximity, not measured leaflet contact area or sealing force. Prolapse alone is not assigned a regurgitation grade. Stale or failed mechanical results must be recalculated before this assessment is current; patient-derived image surfaces remain geometry-only and never enter these flow or repair estimates.

Suspected categories: original teaching rules

Each valve has separate suspected stenosis and regurgitation categories: None detected, Mild, Moderate, Severe or Indeterminate. These are unvalidated model categories shared across all four valves, not ASE severity thresholds, a clinical diagnosis or a recommendation. None detected does not establish a normal valve or exclude disease. Jet colour, area and brightness do not enter the rules.

Suspected stenosis first compares mean pressure-open projected area with the pristine synthetic reference. A ratio of at least 0.80 gives None detected. Below that, an explicit restriction/fusion input, mitral type IIIa or aortic type III mechanism permits the ordinal opening bins below. Without those features, a ring or nonzero-stiffness clip/suture requires a valid untouched baseline and a current/baseline opening ratio below 0.80; the category then uses that baseline ratio, not the pristine ratio. Otherwise the result is Indeterminate. Redundant prolapsing tissue can overlap in projection without true obstruction, so a low projected opening alone cannot establish stenosis or prove that a repair caused it.

Suspected regurgitation first requires valid geometry, a positive ejection budget, forward geometric capacity above the 0.01 L/min numerical floor and an uncapped reverse/forward capacity ratio. Invalid evidence, zero budget or a raw capacity ratio above 9 makes the result Indeterminate. With resolved evidence, None detected requires both a closed-phase excess leak proxy at or below 0.5% of annular area and modeled reverse/gross forward flow k/(1+k) at or below the original 1% teaching tolerance. If either floor is exceeded, the category uses the modeled reverse/gross fraction. A gap within its sampling floor cannot conceal substantial modeled return. This capacity-derived index depends on prescribed loading and geometry; it is not a measured regurgitant fraction, EROA or lesion-specific clinical cutoff.

The untreated Barlow example therefore has an indeterminate stenosis category when only redundant projected coverage suggests narrowing. Its suspected regurgitation category comes from resolved modeled return burden, not from prolapse alone. Stenosis categories remain geometry-based at zero ejection budget; regurgitation categories become indeterminate. Changing a positive budget alone scales flows without changing these categories. The rules do not alter prescribed pressure, solved tissue mechanics or the shared circulation budget.

Category thresholds

Apply these bins only after the eligibility checks above. They are original ordinal conventions for this prototype; clinical grading requires integrated valve-specific assessment.

Read the measurements

Hover a result for its method and units. Evaluate the cardiac phase and numerical status before comparing.

Case format and safeguards

Version 1 procedural-case JSON stores provenance, units, coordinate system, baseline anatomy, alternative plans, complete operation histories and cursors, material and pressure inputs, optional echo measurements and wall-motion scores, mechanism metadata, solver settings, phase, camera, chamber envelopes, markings and results. Image-informed cases additionally retain their fitted dimensions, source-frame association, landmarks and assumptions. The file excludes DICOM pixels, segmentation and registered image-derived surfaces. Import accepts synthetic and validated image-informed parameter cases, validates active and redo history, limits file size and rejects invalid units, nonfinite values, unsupported operations or attachment references. Image-surface export is a separate geometry-only format, not a mechanical-case import. Named G4 device identities persist with their repair operation; a legacy clip with no device identity retains its generic single-pair constraint. Invalid device identities or a valve/device mismatch are rejected.

Editing changes the calculation key. Results with a different key are not current. A browser worker keeps calculations off the rendering thread; request IDs reject superseded replies. The untouched baseline never inherits repair operations. Numerical inputs are available alongside viewport picking, and the viewer supports keyboard rotation and zoom. The active editor keeps up to 30 plan snapshots for Undo/Redo of revisions and removals; these temporary stacks are not part of saved JSON. Saved operation sequences and their cursor remain replayable after import. New edits clear the corresponding redo path. Switching plans or starting a new case resets temporary editor history. editor-history.ts validates every replay boundary, including undone operations, before restoring a plan or removing an edit; dependency failures leave the accepted plan unchanged.

Verification is not clinical validation

Synthetic-model checks cover finite indexed geometry, shared scallop boundaries, retained attachments, atomic rejected edits, fixed native rest lengths, tension-only behavior, load response, deterministic recalculation, case round trips, stale keys and forced nonconvergence. Image tests use synthetic pixels to check spatial calibration, stack rejection, source preservation, seeded connectivity, mask edits and surface topology. These checks verify implementation properties, not whether a valve was correctly segmented. Benchmarks and release verification are retained with the project.

Future validation needs independently segmented imaging, bench closure/contact measurements under recorded loads, mesh and solver convergence studies, material sensitivity, uncertainty propagation and blinded comparison with observed repairs. Clinical planning requires a substantially higher-fidelity validated model and governance. Published valve-modeling work informs the architecture; it does not validate this implementation.

Next milestones and imaging

Next: robust cut remeshing and stitched boundary reconstruction, general leaflet contact, moving annular/papillary boundaries and measured material models. Then extend tricuspid repair and add semilunar plication/commissural procedures and replacement with documented device data.

A future coupled-circulation adapter must distinguish prescribed and resulting pressure/flow, document any PAWP-as-LA approximation, and conserve flow. A nonlinear FE/FSI backend can implement the existing solver boundary without replacing the editor.

The current local image workflow provides source measurements and user-reviewed intensity masks with editable surfaces for compatible calibrated stacks. The experimental Philips reader supports the recognized private format described above. The local imaging project preserves immutable source blobs or explicit relinking records with hashes, anatomical labels, phase associations, repair variants and review status. It does not identify valve anatomy automatically or register the segmented surface to the reduced mechanical model. Procedural fitting creates a separate parameterized approximation. Patient-specific mechanics and validated registration remain future work.

Proposed pipeline: validate formats → segmentation and clinician correction → landmarks and registration → surface reconstruction → simulation mesh/attachment quality checks → phase correspondence → independent validation. SlicerHeart offers valve segmentation/analysis and FEM export, including synthetic inferred chords and VTK output. Prefer VTK plus explicit coordinate/units/provenance metadata; unstructured STL alone cannot preserve these semantics. Vendor-specific echo formats need separate compatibility checks.