Implementation of U-Net Architecture for Medical Image Segmentation in Breast Cancer Detection Using Ultrasound Dataset

Abstract

Medical image segmentation is one of the important methods in the field of image analysis to support automatic diagnosis of diseases. In this study, the application of U-Net architecture in segmenting breast cancer ultrasound images is proposed. The dataset used is a collection of ultrasound images with three categories: benign, malignant, and normal. Preprocessing is done by reducing the image size and removing classes without annotations. The U-Net architecture was built from scratch and trained using binary crossentropy loss function and accuracy metrics. Model evaluation is performed based on Mean Intersection over Union (IoU), Precision, Recall, and F1-Score metrics. The test results show that the model is able to perform tumor segmentation quite accurately on low-resolution medical images. This research shows that U-Net can be an efficient solution to assist ultrasound image-based breast cancer detection.