Deep Learning Fusion Model Enhances Breast Lesion Diagnosis

In a recent study, researchers aimed to enhance breast lesion diagnosis and decision-making by developing a deep learning fusion model that integrates ultrasound (US) and mammography (MG) images. The study focused on cases with discordant Breast Imaging Reporting and Data System (BI-RADS) classifications between US and MG.

Breast cancer screening with mammography is limited in women with dense breasts, while ultrasound is less effective in detecting microcalcifications. Combining US and MG could potentially improve cancer detection, especially in women with dense breasts. However, discordant BI-RADS classifications between US and MG can lead to unnecessary biopsies and anxiety.

To address these challenges, the study developed a DL-UM network that integrates US and MG images. The network showed superior performance in sensitivity and specificity compared to DL-U and DL-M networks. In cases with discordant BI-RADS classifications, DL-UM achieved high diagnostic accuracy.

Collaborating with the DL-UM network improved the diagnostic performance of radiologists with varying levels of experience. It led to increased AUC values, specificity, and reduced unnecessary biopsies, especially for junior radiologists. The DL-UM outputs enhanced radiologists’ trust and improved interobserver agreement between US and MG.

The study highlights the potential of DL-UM in optimizing breast lesion diagnosis and management, potentially reducing unnecessary biopsies. The integration of DL models with radiologists can enhance diagnostic accuracy and decision-making in cases with discordant BI-RADS classifications. The findings underscore the value of AI in supporting radiologists and improving patient care in breast cancer diagnosis.