AI-Based Intelligent-Annotation Algorithm for Medical Segmentation from Ultrasound Data.

The management of prostate cancer, a prevalent source of mortality in men, calls for meticulous delineation of the prostate in transrectal ultrasound (TRUS) images for effective treatment planning. This paper introduces a hybrid artificial intelligence approach for prostate delineation, leveraging prior information from experts, a machine learning model, and a quantum-inspired evolutionary network to augment the accuracy of prostate segmentation. The approach incorporates three novel elements: 1) limited prior information from expert and adaptive polygon tracking (APT) module for initial segmentation; 2) a novel historical storage-based quantum-inspired evolutionary network (HQIE) mechanism to search for the optimal neural network and enhancing solution diversity and capacity to address unimodal and multimodal challenges, and 3) a unique mathematical formulation denoted by parameters of the neural network is used to achieve smooth prostate periphery. The method was evaluated across various noise conditions and against several state-of-the-art methods using a multi-center dataset. In addition, an ablation study was performed to evaluate the efficacy of each component. The results demonstrated the superior performance of the hybrid AI method (Dice index: 96.4 ± 2.4%) against state-of-the-art deep learning methods (e.g., UTNet, Dice index: 90.1 ± 5.7%). The hybrid method also showed higher robustness to image noise than traditional methods. This study suggests new insights and technical approaches in the field of prostate segmentation using hybrid artificial intelligence methods.