cmTriage

cmTriage is a software device that utilizes an artificial intelligence algorithm to analyze 2D FFDM screening mammograms, aiming to improve radiologist workflow by passively notifying them of potentially suspicious cases for prioritized review via their PACS worklist. Its methodology involves secure cloud-based image processing and was validated through a retrospective study demonstrating high sensitivity and specificity in identifying suspicious findings.

CureMetrix's cmTriage is a radiological computer-assisted triage and notification software designed to help radiologists prioritize their worklist for 2D Full-Field Digital Mammography (FFDM) screening mammograms. The system operates by receiving DICOM images from a Picture Archiving and Communication System (PACS), de-identifying and encrypting them, and then transmitting them to a secure cloud environment. Within the cloud, an artificial intelligence algorithm analyzes the images to identify studies suggestive of suspicious findings. The result, a DICOM Structured Report (SR) file indicating either "Suspicious" or blank, is then sent back to the facility's PACS. This allows the cmTriage code to be displayed in the worklist, enabling radiologists to sort and potentially review suspicious cases sooner. cmTriage provides a passive notification, does not alter the original medical images, and is not intended for diagnostic use; it solely aids in workflow prioritization. The performance of cmTriage was evaluated in a retrospective, blinded, multi-center study using a dataset of 1255 mammographic studies, including 400 biopsy-proven cancers and 855 normal studies with two-year negative follow-up. This dataset was enriched with cancer cases and represented a typical screening population in terms of age, breast density, and lesion type. Key performance metrics included a mean sensitivity of 86.9% (95% CI: 83.6%-90.2%) and a mean specificity of 88.5% (95% CI: 86.4%-90.7%), exceeding pre-defined targets based on Breast Cancer Surveillance Consortium (BCSC) data. The overall area under the curve (AUC) was 0.951, demonstrating robust performance across different breast densities and lesion types (masses and calcifications). The system also demonstrated clinically acceptable processing times, averaging 3.35 minutes per study for analysis and notification.

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. How does the performance of cmTriage (sensitivity, specificity, AUC) compare when applied to digital breast tomosynthesis (DBT) images, given its current limitation to 2D FFDM?; What is the real-world impact on radiologists' reading times and diagnostic accuracy when cmTriage prioritization is integrated into their workflow, beyond the standalone performance metrics?; How does the AI algorithm differentiate between various types of suspicious findings (e.g., masses vs. microcalcifications), and are there specific lesion characteristics where it performs better or worse?; What are the computational and infrastructure requirements for a healthcare facility to implement the cmTriage system, particularly concerning data transfer to and from the cloud and PACS integration?; Could the "Suspicious" notification from cmTriage introduce an anchoring bias in radiologists, and what measures could be taken to mitigate this if observed?