Transfer Learning for Breast Tumor Segmentation in DCE MRI

Breast cancer ranks as the most prevalent form of cancer among women. Various screening methods, including mammography ultrasound and Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI), were commonly utilized for detection. DCE-MRI is recognized as a highly effective diagnostic tool for detecting and diagnosing Breast Cancer (BC). This paper aims to enhance breast tumor segmentation in DCE-MRI images using transfer learning and fine-tuning techniques on the BreastDM dataset [1]. It involves fine-tuning a pre-trained Hierarchical U-Net model. The results of this approach significantly improve the performance of DCE-MRI segmentation. Notably, the model achieves a Dice Similarity Coefficient (DSC) of 0.801 and an accuracy of 0.814. Additionally, it attains a Jaccard Index of 0.622 and a loss of 0.024. These metrics surpass those of the pre-trained model and all other segmentation methods utilized in the BreastDM dataset, highlighting the efficacy of the proposed approach in enhancing breast tumor segmentation accuracy in DCE-MRI images.