Medical Image Segmentation with U-Net for Breast Cancer and Lump Type Prediction

Abstract

Today, it is known that, like other types of cancer, breast cancer cases are increasing every year. It is an indisputable fact that early diagnosis and correct diagnosis, which medical literature pays particular attention to, have positive effects on patient health in this field. Therefore, it is a great necessity to identify breast cancer cases correctly and not to confuse malignant masses with benign masses. At this stage, a study was carried out to support the physician's decision, and segmentation was carried out on images containing potential breast cancer with the U-Net model. In addition, a two-class classification model is proposed to determine whether ultrasound images containing masses are benign or malignant. While the detection rate of the current model for benign masses is 99.9%, it reaches 86.6% for malignant masses. In addition, various data augmentation techniques were used hybridly to increase malignant mass images containing fewer samples, and the overall prediction success was increased to 91.13%. © 2024 IEEE.