Related Experiment Video
Updated: Oct 8, 2025
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
3.0K
Computer-aided segmentation on MRI for prostate radiotherapy, part II: Comparing human and computer observer
Jeremiah W Sanders1, Henry Mok2, Alexander N Hanania3
1Department of Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, United States.
Summary
Deep learning algorithms demonstrate superior performance in segmenting prostate radiotherapy structures compared to human observers. This finding is crucial for ensuring the quality and clinical application of AI in radiation oncology.
Area of Science:
- Medical Imaging
- Radiotherapy
- Artificial Intelligence
Background:
- Human interobserver variability is a significant source of noise in manual annotations for prostate radiotherapy.
- Evaluating deep learning (DL) algorithms against human performance is essential for clinical integration and quality assurance.
Purpose of the Study:
- To compare the segmentation accuracy of 114 DL algorithms against 5 human observers for prostate radiotherapy.
- To assess the clinical applicability and quality assurance of DL in prostate cancer treatment.
Main Methods:
- Developed 114 DL algorithms on 295 prostate MRIs for segmenting prostate, EUS, SV, rectum, and bladder.
- Used 50 MRIs from 25 patients as an independent test set.
- Computed spatial entropy (SE) and similarity metrics, comparing DL predictions to human contours.
Main Results:
- DL algorithms achieved statistically significantly higher similarity metrics for prostate segmentation than human observers.
- DL outperformed humans in segmenting the four organs at risk.
- Both DL and human annotators showed variability in similar anatomical regions, particularly at organ junctions.
Conclusions:
- Annotation quality is a critical factor in the development and clinical use of DL algorithms.
- DL algorithms show promise for improving segmentation accuracy in prostate radiotherapy.
- Further research is needed to optimize DL performance and ensure reliable clinical application.
