Organ boundary delineation for automated diagnosis from multi-center using ultrasound images

Abstract

Delineation of prostate boundary is a critical procedure in multi-modal prostate registration and patient-specific anatomical modeling for surgical planning and image-guided biopsy of prostate cancer. Accurate segmentation of the prostate from medical transrectal ultrasound (TRUS) images is challenging due to the presence of shadow artifacts that may cause missing or ambiguous boundaries of the prostate. In this work, we developed a novel hybrid model consisting of four primary merits to address this issue: (1) integrating the properties of the principal curve to fit the data center intelligently and of the artificial intelligence model to decrease model error; (2) developing a principal curve-based adaptive polygon tracking model to acquire the vertices sequence, which is designed via newly merging an automatic judgment of data radius algorithm; (3) introducing the quantum characteristics into the historical storage-based evolutionary model, which is proposed for the first attempt, whilst innovatively introduced new mutation technique and cuckoo search model; (4) presenting an interpretable mathematical model (explained via parameters of neural network) to express the smooth boundary. Promising experimental outcomes denote that our model has satisfactory segmentation performance. Future work will concentrate on improving the trustworthy performance of the model.