Intelligent Diagnosis of Prostate Tumors Based on Depth-wise Separable Convolutional Neural Network using Transrectal Ultrasound B-mode Images

Prostate cancer (PCa) is the most common cancer among men worldwide in terms of incidence. Transrectal ultrasound (TRUS) imaging is one of the most commonly used methods for diagnosing PCa, offering valuable features that aid in the diagnosis. This study uses convolutional neural network (CNN) to extract deep features from TRUS B- mode images, and establishes a novel neural network model for differentiating diagnosis of benign and malignant nature of prostate tumors. The proposed deep feature-based intelligent diagnostic method achieves an accuracy of 89.37%, a sensitivity of 86.23%, a specificity of 92.11%, an F1-score of 88.53%, and an area under the receiver operating characteristic curve (AUC) of 90.67%. The comparison with other commonly used CNN models and a non-deep learning methods based on manually extracted features reveals that our proposed method performs better across all the evaluation metrics on our dataset. In summary, the proposed method is a competitive tool for the intelligent diagnosis of benign and malignant prostate tumors.