DeepCatch

DeepCatch is medical imaging software that employs automated algorithms to segment anatomical structures from whole body CT scans, providing quantitative measurements of volume and proportion, along with 3D visualizations. Its scientific basis lies in processing DICOM CT data for accurate anatomical delineation, validated through comprehensive performance testing against ground truth and predicate devices across diverse patient datasets to support clinical assessment.

DeepCatch is medical image processing software designed to analyze whole body CT images for the automatic segmentation of various anatomical structures, including skin, bone, muscle, visceral fat, subcutaneous fat, internal organs, and the central nervous system. The software processes DICOM-compliant CT data using auto-segmentation algorithms to identify and delineate these tissues. Its primary outputs are quantitative measurements, such as volume and proportions of each structure, delivered as CSV data, and corresponding 3D models for visualization, intended to support clinical assessment and potentially serve as base data for diverse medical applications. The software's efficacy has been established through rigorous non-clinical testing across diverse international datasets (Korean, French, and US-based). Performance evaluations demonstrated high accuracy, with Dice Similarity Coefficients (DSC) for segmentation consistently exceeding 0.900 when compared to ground truth data. Key quantitative outputs, including volumetric and area measurements, showed mean differences of less than ±10% from ground truth, while ratio differences were less than ±1%, and body circumference differences remained below ±5%. Comparative studies against existing predicate devices confirmed that DeepCatch's performance is not inferior, and it exhibited enhanced results in specific areas such as muscle segmentation and certain fat area measurements. The system is designed to be used adjunctively, requiring physician oversight for final confirmation of results.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. What specific deep learning architecture or image processing algorithms does DeepCatch utilize for its auto-segmentation, and how were these algorithms trained and validated prior to the performance tests described?; How does the software handle variations in CT image quality, scanner protocols, and patient anatomical abnormalities (e.g., presence of implants or pathological changes) that might affect segmentation accuracy?; The performance tests mention ground truth created by a licensed physician; what was the inter-observer variability in creating these ground truths, and how was this variability accounted for when assessing DeepCatch's accuracy?; Beyond volume and proportion, does DeepCatch offer texture analysis features or other quantitative imaging biomarkers from the segmented regions that could be clinically relevant for conditions like sarcopenia or detailed adiposity assessment?; Given its performance on US, Korean, and French datasets, what is the anticipated generalizability and robustness of DeepCatch's segmentation accuracy across broader global populations and different CT scanner vendors or models not explicitly included in the validation studies?