Comp2Comp: Open-Source Software with FDA-Cleared Artificial Intelligence Algorithms for Computed Tomography Image Analysis

TL;DR

This work tackles the transparency and validation gap in FDA-cleared AI for opportunistic CT by introducing Comp2Comp, an open-source platform housing two FDA 510(k)-cleared modules, AAQ for abdominal aorta diameter and BMD for vertebral bone density. The authors detail end-to-end pipelines—AAQ with a 3D nnU-Net that crops to L1–L5 and extracts the maximal minor-axis diameter, and BMD with L1–L4 vertebral HU-based calibration to flag low density—validated across four external institutions against radiologist measurements and DXA references, respectively, with strong performance metrics ($MAE=1.58$ mm, $ICC=0.985$, $S=81.0\%,\;SP=78.4\%$, $r=0.791$, AUROC continuous $=0.883$, binary $=0.797$). By making code and model weights openly accessible, the work emphasizes reproducibility and local validation to streamline adoption and support future FDA clearance of open AI tools. The findings suggest meaningful clinical impact for opportunistic AAA detection and osteoporosis risk screening, while acknowledging limitations such as post-endograft performance for AAQ and lack of contrast-enhanced CT validation for BMD. Overall, Comp2Comp advances transparent, open-source AI in radiology and provides a framework for broader, multi-institutional validation and registry-building in body composition analysis.

Abstract

Artificial intelligence allows automatic extraction of imaging biomarkers from already-acquired radiologic images. This paradigm of opportunistic imaging adds value to medical imaging without additional imaging costs or patient radiation exposure. However, many open-source image analysis solutions lack rigorous validation while commercial solutions lack transparency, leading to unexpected failures when deployed. Here, we report development and validation for two of the first fully open-sourced, FDA-510(k)-cleared deep learning pipelines to mitigate both challenges: Abdominal Aortic Quantification (AAQ) and Bone Mineral Density (BMD) estimation are both offered within the Comp2Comp package for opportunistic analysis of computed tomography scans. AAQ segments the abdominal aorta to assess aneurysm size; BMD segments vertebral bodies to estimate trabecular bone density and osteoporosis risk. AAQ-derived maximal aortic diameters were compared against radiologist ground-truth measurements on 258 patient scans enriched for abdominal aortic aneurysms from four external institutions. BMD binary classifications (low vs. normal bone density) were compared against concurrent DXA scan ground truths obtained on 371 patient scans from four external institutions. AAQ had an overall mean absolute error of 1.57 mm (95% CI 1.38-1.80 mm). BMD had a sensitivity of 81.0% (95% CI 74.0-86.8%) and specificity of 78.4% (95% CI 72.3-83.7%). Comp2Comp AAQ and BMD demonstrated sufficient accuracy for clinical use. Open-sourcing these algorithms improves transparency of typically opaque FDA clearance processes, allows hospitals to test the algorithms before cumbersome clinical pilots, and provides researchers with best-in-class methods.

Figures (7)

• Figure 1: (A) Abdominal aortic quantification (AAQ) pipeline architecture from input DICOM files of CT scan to output of aortic maximum diameter with associated statistics and visualizations. (B) Study layout, consisting of data acquisition for AAQ training, integration within the AAQ pipeline, and evaluation across multiple institutions. (C) Representative AAQ outputs from contrast enhanced and non-contrast enhanced CT scans. The red dot denotes the center of the segmentation slice, and the blue line denotes the reported diameter measurement. The dashed vertical line on the graph is at 3 cm, the threshold size for aortic aneurysm. (D) Example maximum intensity projection of an abdominal aortic aneurysm (not part of algorithm output). TS = TotalSegmentator30 DASA = Diagnósticos da América S.A. UAB = University of Alabama at Birmingham NC = North Carolina MedStar = MedStar Health
• Figure 2: Overview of the BMD Algorithm Development and Output. (A) Processing pipeline from DICOM series input to BMD classification (normal or low density). The pipeline segments the L1-L4 vertebrae and generates ROIs in the body of the vertebrae. The Hounsfield Unit (HU) values are then calibrated by segmented visceral adipose tissue and a ROI placed in air. The final output is a binary prediction of an estimated DXA T-score $\ge -1$ (normal) or $< -1$ (osteopenic). (B) Development and FDA approval cohorts, showing distributions by sex, imaging center, age, and CT scanner vendor. (C) Algorithm output displays coronal and sagittal views with overlaid spine segmentations and ROIs, including median HU values for both the complete vertebra and the ROI. VerSe = Vertebrae labeling and segmentation benchmark dataset31 UCSF = University of California San Francisco TS = TotalSegmentator30 DASA = Diagnósticos da América S.A. UAB = University of Alabama at Birmingham UH = University Hospitals Cleveland Medical Center MedStar = MedStar Health
• Figure 3: (A) Bland-Altman plot demonstrating relationship between algorithm accuracy and aortic diameter. (B) Scatter plot demonstrating close linear relationship between ground truth and AAQ measurement of aortic diameter. LOA = Limit of Agreement
• Figure 4: (A) Bland-Altman and (B) scatter plots of BMD continuous scores predicted by the model from CT scans against DXA T-score BMD ground truth. (C) AUROC of BMD model continuous score against DXA T-score, and (D) AUROC of BMD model binary score against DXA binary classification. AUROC = area under receiver-operator curve
• Figure 5: Selection of largest minor axis of cross-sectional aortic ellipse to accurately represent true aortic diameter.