Mammography, one of the most widely performed imaging tests in women, may contain latent vascular and cardiometabolic information beyond oncological findings. Today, Doctor Viana Copeland (Chaim Sheba Medical Center, Tel Aviv University - Ramat Gan, Israel) presents research evaluating the performance of a deep learning-based algorithm for detecting hypertension, ischaemic heart disease (IHD) and cerebrovascular accident (CVA) from mammograms.
This retrospective cohort study included 29,921 women who underwent 97,364 mammography examinations and for whom clinical data on hypertension, IHD and CVA were extracted from electronic medical records, medication prescriptions, procedural and imaging findings, and documented in-hospital measurements.
A convolutional neural network was developed to predict the presence of hypertension, IHD and CVA directly from mammographic images. The model’s performance was evaluated using areas under the receiver operating characteristic curve (AUROC) for each disease.
Based on medical records, the prevalence was 16% for hypertension, 2.5% for IHD and 2.5% for CVA among this cohort who had a median age of 54 years. The deep learning model had significant discriminative performance, yielding AUROCs of 0.79 for hypertension, 0.78 for IHD and 0.86 for CVA. Prespecified sensitivity analyses based on cancer status, mammographic view and age showed consistent results.
The researchers are now working to improve the model’s accuracy and reduce both false positives and false negatives. They also plan to investigate whether mammograms could help identify other cardiovascular conditions. They conclude that mammography may serve as a dual-purpose screening platform – extending beyond cancer detection – and provide a scalable approach for opportunistic CVD identification to reduce persistent underdiagnosis in women.