AI-enhanced Mammography With Digital Breast Tomosynthesis for Breast Cancer Detection: Clinical Value and Comparison
Daphne Resch1, Roberto Lo Gullo1, Jonas Teuwen1
1From the Department of Biomedical Imaging and Image-guided Therapy, Division of Molecular and Gender Imaging, Medical University of Vienna, Austria (D.R.); Department of Radiology, Breast Imaging Service, Memorial Sloan-Kettering Cancer Center, New York, NY (R.L.G., J.T.); Center for Medical Physics and Biomedical Engineering, Medical University Vienna, Vienna, Austria (F.S., J.H.); St Francis Hospital Vienna, Vienna, Austria (A.R.); Sigmund Freud University Medical School, Vienna, Austria (A.R.); and Department of Radiology, Division of Breast Imaging, Columbia University Irving Medical Center, 161 Fort Washington Ave, New York, NY 10032 (K.P.).
Two artificial intelligence (AI) systems for mammography showed high accuracy in detecting breast cancer using digital breast tomosynthesis (DBT). However, their performance was lower than that of experienced radiologists performing double-reading.
Area of Science:
- Radiology and Medical Imaging
- Oncology
- Artificial Intelligence in Healthcare
Background:
- Mammography with digital breast tomosynthesis (DBT) is a key tool for breast cancer screening.
- Artificial intelligence (AI) systems are increasingly being developed to aid in mammographic interpretation.
- Evaluating the performance of commercially available AI systems against human readers is crucial for clinical adoption.
Purpose of the Study:
- To compare the diagnostic performance of two AI systems (Transpara and ProFound AI) for breast cancer detection using DBT.
- To benchmark the AI systems' performance against radiologists performing standard-of-care double-reading.
- To assess AI performance across different breast densities.
Main Methods:
- Retrospective analysis of mammography with DBT examinations from 2019-2020.
- Evaluation of 419 patient examinations using Transpara 1.7.0 and ProFound AI 3.0.
- Comparison of AI performance with human double-reading using receiver operating characteristic (ROC) analysis and the DeLong test.
Main Results:
- Both AI systems demonstrated high areas under the ROC curve (AUC) for malignancy detection: Transpara (0.86) and ProFound AI (0.93).
- Human double-reading achieved a higher AUC (0.98).
- Specific sensitivity and specificity thresholds were identified for both AI systems, with ProFound AI showing higher specificity in rule-out criteria.
Conclusions:
- Commercially available AI systems demonstrate significant capabilities in detecting breast cancer with DBT.
- While effective, the performance of these AI systems currently lags behind that of expert human double-reading.
- Further research and development may enhance AI performance for breast cancer screening.
