QA • Medical Imaging

Accuracy Validation of Manufacturer-Specific Flow Calculation Algorithms

Duration: 6 weeks • Role: Lead QA

This project involved analyzing cardiac blood flow calculation dynamics across multiple Regions of Interest (ROIs) over different phases of the cardiac cycle. The dataset was generated from phase-resolved flow measurements and used to evaluate forward flow, backward flow, regurgitation fraction, velocity, acceleration, and pressure gradients.

Context

The analysis was critical because inaccuracies in flow calculation or phase alignment just because of difference in manufacturer could lead to misinterpretation of regurgitation severity and cardiac efficiency, directly affecting clinical conclusions.

Blood Flow Calculation Algorithm Validation Medical Image Data Analysis Quantitative Review Mathematical Models Clinical Phantom 4D MRI Metric Correlation

Primary Risks Identified

  • Phase misalignment risk affecting forward/backward volume accuracy
  • ROI consistency risk across all cardiac phases
  • Timing approximation risk due to non-exact temporal resolution
  • High regurgitation fraction misclassification risk
  • Heavy reliance on legacy formulas without version control or audit trail

Testing Approach

Testing focused on comparing the calculated Excel results against "Phantom" (simulated) datasets with known flow rates to validate the formulas, with inter-Phase Consistency Check for verifying that the integral of flow over a full cycle (Stroke Volume) remained consistent across different ROIs in the same vessel and also sensitivity Analysis for adjusting ROI boundaries by $\pm 1$ pixel to determine the impact of segmentation variance on the final flow volume.

  • User Acceptance Testing (UAT): Conducted sessions with 5 target users to gather feedback on the UI flow for manufacturer-specific outcomes.
  • Compared global volumes vs ruler-constrained volumes
  • Validated inside vs outside ruler volumes for boundary leakage
  • Reviewed physiological plausibility of: Velocity extrema, Acceleration symmetry, and Pressure gradient ranges

AI/ML Validation Excel Power BI DICOM Cardiac Flow Physiology Models Python

Outcome

Key Calculations Implemented and Validated:

  • Regurgitation Fraction: =ABS(Total Volume - Forward Volume)/Total Volume × 100
  • Pressure Gradient: =4 × (Max Velocity)² / 10000
  • Flow Acceleration: =(Mean Velocityₜ₂ - Mean Velocityₜ₁)/(Time₂ - Time₁)

Key Lessons

Manually processing ROIs across 20+ phases is prone to human error; scripting the calculation in Excel/Python reduced processing time by 70%. If a flow curve doesn't return to zero or show expected retrograde flow, the data must be re-validated against the originalimaging parameters.

  • Successfully automated the calculation validation of Stroke Volume (SV) and Cardiac Output (CO) for 4 distinct ROIs.
  • Identified a 13% variance in peak systolic velocity that was previously obscured by manual averaging.
  • Created a "Master Dashboard" that visualizes flow curves, allowing clinicians to see "Phase-Shift" anomalies at a glance.
  • (Documentation) Comprehensive formula documentation was essential for clinical audit purposes

Have a system like this?

If you have no idea of testing your current system, let's talk through what's actually going on.

Contact me