FDA-Cleared CTA-Based Plaque Analysis

Supporting regulatory clearance through rigorous validation

This project is centered on validating the ITK-based image processing algorithm to characterize and quantify atherosclerotic plaque composition in the arterial vessel wall from routine contrast enhanced CT angiography (CTA) images. The work directly contributed to building the clinical evidence required for FDA 510(k) clearance, and led to the commercialization of the Elucid PlaqueIQ software.

The core challenge was establishing algorithm accuracy against a histology-based ground truth — the gold standard for tissue characterization — ensuring that automated plaque measurements met the performance thresholds demanded by clinical regulators.

Validation methodology and engineering

The validation begins with clinical analysts creating arterial vessel lumen and wall models from CTA image volumes and meticulously aligning the histopathology cross-sections to the locations in the corresponding CTA image volume. A subset of the data was used for tuning the parameters in the algorithm while another subset (blinded to engineering) was used for validation.

The image processing algorithm is containerized and can be deployed to run in any GPU-enabled environment. A Python-based tool executes parallel processing of the dataset to compare algorithm-derived area measurements of Calcified Plaque, Non-calcified Matrix, Lipid Rich Necrotic Core for each section against histology ground truth. Metrics including slope, bias, Pearson correlation coefficient are computed to evaluate performance for a given set of parameters.

• Methodology Run batch processing on the training set for parameter tuning to compute metrics for plaque area measurements against histological cross-sections; the optimal parameters are used to run against the validation set to create quantitative performance evaluation against pre-defined regulatory criteria
• Key Challenge Establishing ground truth correspondence between in-vivo imaging and ex-vivo histology while accounting for tissue deformation and processing artifacts
• Infrastructure Fully containerized analysis pipelines (Docker) deployed on AWS
• Regulatory Context FDA 510(k) submission pathway; work contributed to the algorithm performance section of the clinical evidence package