Delineation of Prostate Cancer Via Enhanced AI-Based Algorithm In Ultrasound Images

Abstract

Delineation of prostate cancer (PCa) on ultrasound images has become an essential technique for early PCa treatment, which still faces several challenges, such as low image contrast and blurred organ boundaries. Facing the aforementioned issues, a novel coarse-fine segmentation framework method is adopted in this work. First, a deep learning algorithm is used for positioning the region of interest (ROI); then, the principal curve-related technique is adopted to fine-tune initial results. Thirdly, an enhanced quantum evolutionary algorithm is used to hunt for the optimal initialization of the backpropagation neural network. Finally, an interpretable mathematical model is introduced for representing the ROI contour by using the parameters of the neural network. A comprehensive evaluation using the clinical data proves the superiority of our method, with Dice similarity coefficient (DSC) of 83.48+3.44%, Jaccard similarity coefficient (Ω) of 82.25+4.12%, and accuracy (ACC) of 83.4+3.57%, respectively.