Hybrid Deep Learning Models with Lesion Detection for Breast Cancer Diagnosis from Ultrasound Images

Breast cancer is a lethal type of cancer that can be treated with early diagnosis. In this study, a deep learning-based system capable of detecting lesions from ultrasound images for the diagnosis of breast cancer is proposed. The proposed system extracts features from the processed images using the ResNet- 101 architecture, through a lesion detection module and various preprocessing steps. The extracted features are evaluated using an artificial neural network and specific machine learning algorithms, classifying the images into three categories: benign, malignant, and normal. The experiments have shown that the system achieves an accuracy rate of 98-100%. The dataset used, data limitations, and the model's training process have been detailed, and the results have been compared with other studies. The results demonstrate that the proposed system is a reliable tool for diagnosing breast cancer.