Transpara

Transpara™ is a software device utilizing machine learning algorithms to assist radiologists in the detection and diagnosis of breast cancer from screening mammograms by providing CAD marks, region-specific scores, and an overall exam score. Clinical and non-clinical studies demonstrate its ability to significantly improve radiologists' detection performance, particularly by increasing AUC, with high sensitivity for suspicious lesions and calcifications, without substantially increasing reading time.

Transpara™ is a software-only medical device designed as a concurrent reading aid for radiologists interpreting screening mammograms to identify and assess regions suspicious for breast cancer. The system utilizes machine learning algorithms, trained on extensive databases of biopsy-proven cancer, benign lesions, and normal tissue, to detect suspicious calcifications and soft tissue lesions. It processes DICOM mammography images, offering features like computer-aided detection (CAD) marks for highly suspicious findings, interactive querying of regions for a suspiciousness score, and an overall exam-based "Transpara™ Score" (1-10) indicating cancer likelihood. The software includes image pre-processing to ensure compatibility across images from different mammography vendors. Non-clinical standalone performance testing on a multi-vendor dataset demonstrated high sensitivity for calcifications (94.1%) and soft tissue lesions (72.0%) with low false positive rates. The Transpara™ Score effectively stratified exams, with 78% of cancer cases scoring in the highest likelihood category. A pivotal clinical reader study involving 14 radiologists and 240 cases showed that using Transpara™ significantly improved their cancer detection performance, increasing the average area under the ROC curve (AUC) from 0.866 to 0.886 (P=0.0019) without significantly impacting reading time. This improvement was observed for both soft tissue lesions and calcifications, and Transpara's standalone detection performance (AUC=0.887) was comparable to the average performance of unaided radiologists.

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 "Transpara™ Score" calibration, aiming for approximately equal numbers of screening mammograms in each category, affect its discriminatory power for very rare or very common presentations of cancer in specific subpopulations?; What are the specific characteristics of the "large databases of biopsy proven examples" used for training the algorithms, particularly regarding diversity in patient demographics, breast density, and cancer subtypes?; The study showed an average AUC increase of 0.020; for the two radiologists where performance did not improve or slightly decreased, what were the potential contributing factors or case characteristics?; How does the software's pre-processing component standardize images from different vendors, and what is the performance variability, if any, when dealing with older versus newer imaging equipment or varying image acquisition protocols?; Given the standalone performance of Transpara™ approached that of radiologists, what are the ethical and practical considerations for its potential future role, beyond a concurrent reading aid, in the screening workflow?