Transpara 1.7.0

Transpara® 1.7.0 is a software device utilizing deep learning algorithms to assist physicians in interpreting digital mammography and tomosynthesis images by detecting and characterizing suspicious lesions. Its updated algorithms demonstrate improved, non-inferior performance in identifying calcifications and soft tissue abnormalities, including newly added support for Fujifilm DBT, aiming to enhance breast cancer screening accuracy and workflow.

Transpara® 1.7.0 is a software-only medical device designed as a concurrent reading aid for physicians interpreting screening full-field digital mammography (FFDM) and digital breast tomosynthesis (DBT) exams. Its purpose is to improve the detection and characterization of abnormalities suspicious for breast cancer and enhance workflow. The system employs 'deep learning' algorithms, trained on a large, biopsy-proven dataset of breast cancer, benign abnormalities, and normal tissue, to analyze FFDM images and DBT slices. It identifies suspicious calcifications and soft tissue lesions (including densities, masses, architectural distortions, and asymmetries), providing computer-aided detection (CAD) marks, region-specific suspicion scores (0-100), links between corresponding views, and an overall exam score (1-10) indicating cancer likelihood. Transpara® processes DICOM inputs and outputs results in a standardized mammography CAD DICOM format to clinical systems like PACS and RIS, operating as a virtual machine with user interface implementation handled by third parties. Non-clinical performance testing on an independent, multi-center dataset (7882 non-cancer, 1240 cancer exams) demonstrated Transpara® 1.7.0's effectiveness. For 2D FFDM, it achieved 94.7% sensitivity for calcifications at 0.11 false positives (FP)/image, and up to 92.6% sensitivity for soft tissue lesions at 0.17 FP/image. For DBT, overall sensitivity was 91.3% at 0.3 FP/volume. Compared to its predicate (Transpara® 1.6.0), version 1.7.0 showed non-inferior and improved Area Under the Curve (AUC) values on a common testset: 0.949 for 2D (vs. 0.929) and 0.931 for DBT (vs. 0.917). This updated version also introduces validated support for Fujifilm DBT systems, for which it demonstrated an AUC of 0.952. These enhancements are attributed to algorithmic changes aimed at improving detection accuracy across FFDM and DBT modalities.

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 was the specific composition and diversity of the "large database of biopsy-proven examples" used for training the deep learning algorithms, particularly regarding rare cancer subtypes or challenging benign mimics?; How does the "exam score" calibration, aiming for approximately 10 percent of non-cancer mammograms in each category, account for variations in baseline population cancer prevalence and image acquisition protocols across different clinical sites?; What are the specific algorithmic changes in Transpara® 1.7.0 compared to 1.6.0 that led to the improved detection accuracy for FFDM and DBT, and the addition of Fujifilm DBT support?; The document mentions false positive rates per image for FFDM and per volume for DBT; what is the variability of these false positive rates across different breast densities and lesion conspicuity levels?; Given that the user interface implementation is by third parties, what are the critical performance characteristics or integration guidelines ScreenPoint Medical provides to ensure optimal and consistent presentation of Transpara's outputs to the end-user radiologist?