Towards Deep Learning-Driven Assessment of Lesion Biopsy in Breast MRI

Motivation: MR-guided biopsies are critical for breast cancer diagnosis, but accurate representation of lesions is challenging, often requiring follow-up imaging that increases patient burden and delays diagnosis. Goal(s): This study aims to develop a real-time tracking system by identifying lesions, position-marker, needle-tips during breast MR-guided biopsies. Approach: We employed a Template-Matching-Network, trained on 235 patients' breast-MRI scans to predict in real-time lesion location, position-marker, and needle-tip during MR-guided breast biopsy. Results: Our model demonstrated high localization accuracy within images, but requires further adjustment cross-images. Results indicate potential to streamline biopsies and minimize follow-up imaging. Future research will focus on assessing adequacy for breast-MRI biopsy. Impact: This study introduces a real-time MR-guided breast biopsies tracking system, enhancing localization accuracy of lesions, position-markers, and needle-tips. This approach aims to improve patient outcomes, optimize breast cancer diagnostic confidence, and reduce the need for follow-up examinations.