RECOMIA

RECOMIA provides automated CT organ segmentation and quantitative extraction for PET/CT imaging to support research-grade medical image analysis.

Key Features:

• Local De-Identification: Local de-identification of medical images prior to transfer to cloud infrastructure is implemented to remove patient identifiers.
• Secure Image Transfer: Secure transfer of image data to the platform's cloud-based infrastructure is supported for collaborative analysis workflows.
• Manual Annotation Tools: Manual annotation of organs and pathologies within images is supported for customized labeling and review.
• Deep Learning-Based Organ Segmentation: AI-driven segmentation of 100 organs (77 bones and 23 soft tissue organs) in CT scans using two convolutional neural networks: one for organs with multiple similar instances (e.g., vertebrae and ribs) and one for other organs.
• Quantification of Segmented Volumes: Tools quantify segmented volumes to enable extraction of measurements and imaging biomarkers.
• Export Functionality: Quantitative results from segmentation and measurements can be exported for downstream analysis and reporting.

Scientific Applications:

• PET/CT Quantification: Extraction of standardized uptake values (SUVs) from PET images using CT-derived organ segmentations enables quantitative PET/CT analyses.
• Imaging Biomarker Research: Volume quantification and segmentation outputs support research into imaging biomarkers in oncology, cardiology, and other clinical research areas.

Methodology:

Segmentation is performed by two convolutional neural networks trained on CT studies from 339 patients with annotations by experienced radiologists; performance on a manually annotated test set reported a mean Dice index of 0.93 for all foreground voxels, with mean Dice indices of 0.86 overall, 0.82 for soft tissue organs, and 0.90 for bones across ten representative organs.