Validate Pipeline

Perform a static analysis of the full pipeline from DICOM loading through export. This does NOT run the app — it reads the code and checks for integration issues.

Validation Checks

1. Function Signature Matching

For every function call from main.py into core modules, verify:

• • All arguments passed exist in the function signature
• • All required parameters (no default) are provided at the call site
• • Argument types are consistent (not passing a string where int expected)
• • No deprecated parameter names from old versions

Check these call chains:

• •main.py -> dicom_utils.py (loading)
• •main.py -> core/metal_detection.py (detection)
• •main.py -> core/discrimination.py (discrimination)
• •main.py -> contour_operations.py (segmentation)
• •main.py -> dicom_export.py (export)
• •main.py -> visualization.py (display)

2. UI Parameter Coverage

Scan all Streamlit widgets in main.py (sliders, checkboxes, selectboxes, number_inputs). For each:

• • Is the value actually used downstream? (not just displayed)
• • Does changing the value affect the output? (trace to computation)
• • Is it gated behind the correct conditional (e.g., only shown when star profiles enabled)?

Flag any widget whose value is never passed to a core function.

3. Mask Shape Consistency

Trace mask creation and usage:

• • Metal mask shape matches volume shape (num_slices, rows, cols)
• • ROI masks are per-component, not global
• • Boolean operations (AND, OR, NOT) are between same-shaped arrays
• • No 2D mask applied to 3D volume without proper broadcasting
• • Russian Doll exclusion: dark AND metal = empty, bright AND metal = empty, bone AND artifact = empty

4. Pipeline Stage Dependencies

Verify execution order makes sense:

• • Metal detection runs before discrimination (discrimination needs metal mask)
• • ROI creation runs before discrimination (discrimination is per-ROI)
• • Body mask available before discrimination (used for bounds)
• • All required session_state keys are set before they're read

5. Import Consistency

• • All imports resolve (no importing removed functions)
• • No circular imports between modules
• • Optional imports (cupy) have proper try/except fallbacks

6. Known Issue Check

Cross-reference against known issues in CLAUDE.md:

• • bone_low/bone_high slider connectivity status
• • Star profile discrimination HU integration status
• • Any workarounds documented still match current code

Output Format

PIPELINE VALIDATION REPORT
==========================

1. FUNCTION SIGNATURES
   [PASS] main.py:423 -> metal_detection.detect_metal_3d() -- all args match
   [ISSUE] main.py:910 -> discrimination.discriminate() -- bone_low not passed

2. UI PARAMETERS
   [PASS] metal_threshold (slider) -> metal_detection.py:45
   [ORPHAN] bone_low (slider) -- defined at main.py:355, never reaches computation

3. MASK CONSISTENCY
   [PASS] All masks are (num_slices, H, W)
   [WARN] Line 892: 2D operation on potentially 3D mask

4. PIPELINE ORDER
   [PASS] All stage dependencies satisfied

5. IMPORTS
   [PASS] All imports resolve

6. KNOWN ISSUES
   [CONFIRMED] bone_low/bone_high still disconnected (CLAUDE.md matches code)

SUMMARY: X passes, Y issues, Z warnings

Important Notes

• •This is a READ-ONLY analysis. Do not modify any files.
• •Focus on actionable issues -- don't flag style preferences or minor warnings.
• •If an issue was already documented in CLAUDE.md, mark it as [CONFIRMED] rather than [ISSUE].
• •Prioritize issues that would cause silent incorrect results (wrong segmentation) over issues that would cause visible errors (crashes).