Evaluation of a Clinical Acceptability of Deep Learning-Based Autocontouring: An Example of The Use of Artificial Intelligence in Prostate Radiotherapy

Authors:
Open Access
Publisher website
Google Scholar
Abstract
This study aimed to evaluate the usability and benefit of a new generation of auto segmentation, that automatically identifies organs and auto-contours them directly at CT simulator before creating prostate radiotherapy plans. The prostates of 10 patients were automatically contoured using the DirectORGANS auto-segmentation algorithm at the CT simulator. The CT scans were imported into the Eclipse treatment planning system for contouring. On the same CT image sets, the prostate was manually contoured by a group of five experienced physicians. MR-guided prostate contours were delineated using MRI images and used as a reference structure. The volumes of the prostate were measured, and the Overlap index (OI), Dice similarity index (DSC), and Volume difference (Dv) were calculated based on contours. The Kruskal-Wallis H test was performed with SPSS (P<0.05). MR-based contouring was used as a reference, and the OI, DSC, Dv, and contouring time results of users and artificial intelligence were analyzed accordingly. There was a significant difference in OI, DSC, and Dv between the results of users and artificial intelligence. The most significant difference between users, artificial intelligence, and MR-based contouring was contouring time (p <0.001). MR- based contouring was time-consuming. Artificial Intelligence’s automatic contouring of the prostate required minimal modification.

This publication has 30 references indexed in Scilit: