Transpara

Transpara™ 1.6.0 is a radiological computer-assisted detection and diagnosis software that utilizes deep learning algorithms, trained on diverse, biopsy-proven mammographic cases, to analyze full-field digital mammography and digital breast tomosynthesis exams for suspicious lesions indicative of breast cancer. It functions as a concurrent reading aid by providing physicians with CAD marks, lesion-specific suspicion scores, and an overall exam cancer likelihood score, with studies demonstrating its ability to enhance radiologist detection accuracy and improve reading efficiency in DBT interpretation.

Transpara™ 1.6.0 is a software application designed to assist physicians in the interpretation of screening full-field digital mammography (FFDM) and digital breast tomosynthesis (DBT) examinations. Its primary purpose is to act as a concurrent reading aid, helping to identify regions suspicious for breast cancer and assess their likelihood of malignancy. The system utilizes advanced 'deep learning' algorithms, which have been trained on a comprehensive database containing biopsy-proven examples of breast cancer, benign abnormalities, and normal breast tissue. These algorithms analyze FFDM images and DBT slices to recognize suspicious calcification groups and various soft tissue lesions, including densities, masses, architectural distortions, and asymmetries. The software processes mammograms or DBT exams provided in DICOM format and is compatible with imaging systems from multiple manufacturers, including Hologic, Siemens, General Electric, Philips, and Fujifilm for FFDM, and Hologic and Siemens for DBT. The output of Transpara™ 1.6.0 includes computer-aided detection (CAD) marks to highlight suspicious locations, region scores on a 0-100 scale indicating suspicion level, links between corresponding regions in different views, and an overall exam score (1-10) reflecting the likelihood of cancer in the exam. This information is sent to medical workstations, PACS, or RIS using the DICOM protocol. Non-clinical stand-alone performance testing confirmed the software's effectiveness in detecting lesions, showing non-inferior or improved performance over its predicate. A pivotal clinical reader study involving 18 radiologists and 240 DBT exams demonstrated that using Transpara™ significantly improved radiologists' cancer detection performance (AUC increased from 0.833 to 0.863), reduced reading time, and provided superior sensitivity without compromising specificity. The system is offered as a virtual machine, with the user interface implementation handled by third-party providers.

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 deep learning algorithm specifically differentiate between the morphological characteristics of various soft tissue lesions, such as masses versus architectural distortions, and how is this reflected in the region scores?; What were the specific performance metrics (e.g., sensitivity at different false positive rates, FROC curves) for calcification detection versus soft-tissue lesion detection in the standalone FFDM and DBT performance tests across the different validated imaging manufacturers?; The pivotal reader study demonstrated improved AUC and reduced reading time using Siemens DBT exams; are there plans for or existing data on similar reader studies using Hologic DBT systems, given the differences in their acquisition and reconstruction technologies?; The exam score (1-10) is calibrated so approximately 10% of mammograms without cancer fall into each category. How robust is this calibration across datasets with varying cancer prevalence and patient demographics, and what is the distribution of cancer cases across these score categories?; Considering that the user interface for displaying Transpara's outputs is provided by third parties, what guidelines or minimum requirements does ScreenPoint Medical provide to ensure consistent, effective, and safe presentation of CAD marks and decision support scores within diverse clinical environments?