WATUNet: Towards Enhancing Breast Cancer Segmentation in Rural Settings
Monday, April 8, 2024
2:56pm – 3:03pm
Location: 412
Authors: Donya Khaledyan, University of Rochester Thomas Marini, University of Rochester Kevin Parker, University of Rochester
Overcoming UNet's limitations, WATUNet integrates wavelet and attention gate modules, significantly improving lesion segmentation in breast ultrasound imagery. Its versatile architecture effectively captures multi-scale spatial context, accurately localizing lesions in various benign and malignant cases. Evaluated on diverse datasets of ultrasound, WATUNet demonstrates robust segmentation performance, achieving a Dice score of 0.94 and an F1 score of 0.94. By enabling precise breast segmentation, this method could impact breast cancer diagnosis, particularly in underserved communities, through the fusion of VSI technology and deep learning algorithms.