Comprehensive Summary
The use of transrectal ultrasound imaging for accurate prostate segmentation has the potential to be used in the development of AI algorithms for diagnosis and treatment of prostate diseases. Previous to this study, however, a large variability in segmentation methods, specifically manual vs. semi-automatic methods, made it difficult to determine the most reliable reference for future AI training. Results of this study suggest semi-automatic segmentation as a preferred standard for AI training in prostate segmentation as it demonstrated comparable accuracy to manual segmentation while having lower rates of inter- and intra-individual variability.
Outcomes and Implications
By highlighting the consistency of semi-automatic prostate segmentation methods, results from this study support the appeal of using semi-automatic methods in the clinical setting. With the potential to improve the accuracy of prostate cancer diagnoses and treatment planning, inter- and intra- individual variability may be reduced across hospital systems. Furthermore, with the potential for future AI algorithm training, other issues such as spatial variability and quality control can be addressed through processes such as optimization of TRUS image acquisition and the ability to alert operators of insufficient annotated points.