CogNet QmTRIAGE

CogNet QmTRIAGE is a software as a medical device that utilizes a machine learning algorithm to analyze 2D FFDM screening mammograms for suspicious findings, aiming to assist radiologists by prioritizing their worklist. Its methodology involves passive notification of flagged exams to a PACS/workstation, enhancing triage efficiency without altering the standard diagnostic pathway or providing diagnostic information.

CogNet QmTRIAGE is a software tool designed to assist MQSA qualified interpreting physicians in managing their workflow for 2D Full-Field Digital Mammography (FFDM) screening exams. It employs an artificial intelligence algorithm to analyze these mammograms at the exam level, identifying and flagging studies that exhibit findings suggestive of potential cancer. This process operates in a parallel workflow, meaning it doesn't interfere with the standard review process or remove any exams from the physician's worklist; instead, it provides a passive notification to a PACS or workstation, allowing for the prioritization of exams deemed suspicious. The system is specifically intended for triage and prioritization purposes only and does not offer diagnostic information, nor should it replace a full patient evaluation or be used to confirm a diagnosis. The AI algorithm was trained on diverse global datasets, independent of the testing data. The performance of CogNet QmTRIAGE was validated using a retrospective cohort of 800 anonymized 2D FFDM mammograms, sourced from the USA and Germany, which included 399 biopsy-confirmed cancer cases and 401 negative cases with two-year follow-up. The validation study demonstrated strong performance metrics, achieving an overall Area Under the Receiver Operating Characteristic curve (AUROC) of 0.9569. Furthermore, the software achieved an overall sensitivity of 87% and a specificity of 89% across the entire test dataset, values that reportedly exceed the standard of care benchmarks from the Breast Cancer Surveillance Consortium study. The study considered various confounding factors such as lesion type, breast density, age, and race, and utilized images from multiple FFDM system vendors, with Hologic Selenia Dimension specified for the initial release.

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 algorithm's performance (sensitivity/specificity) vary across different breast density categories and specific lesion types (e.g., microcalcifications vs. soft tissue masses)?; What were the specific characteristics of the training dataset (e.g., size, geographic distribution beyond "worldwide", cancer prevalence, vendor distribution) and how might these influence generalizability to diverse patient populations and imaging equipment not heavily represented?; The document mentions the exclusion of BI-RADS 0 cases that were determined to be benign after diagnostic workup from the test set; how would the inclusion of such cases, which often require further investigation, impact the software's real-world triage effectiveness and false positive notification rate?; Beyond the Hologic Selenia Dimension, what is the anticipated timeline or validation process for extending compatibility and ensuring consistent performance with other FFDM systems from vendors like GE, Siemens, and Fujifilm mentioned in the performance study?; Could the prioritization feature inadvertently lead to increased radiologist fatigue or bias if a high volume of cases are flagged, and are there any studies planned to assess its long-term impact on radiologist reading times and diagnostic accuracy for non-prioritized cases?