Related Experiment Video
Updated: Dec 9, 2025
08:05
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
976
Radio-pathomic mapping model generated using annotations from five pathologists reliably distinguishes high-grade
Sean D McGarry1, John D Bukowy2, Kenneth A Iczkowski3
1Medical College of Wisconsin, Department of Biophysics, Milwaukee, Wisconsin, United States.
Journal of Medical Imaging (Bellingham, Wash.)
|September 14, 2020
Summary
This study mapped epithelium density using MRI, finding that pathologist variability impacts results. A consensus model improved accuracy, outperforming apparent diffusion coefficient (ADC) in classifying disease grades.
Area of Science:
- Radiology
- Pathology
- Medical Imaging Analysis
Background:
- Interobserver variability in pathology annotations can affect downstream analysis.
- Predictive mapping of epithelium density using magnetic resonance imaging (MRI) is an emerging area.
- Understanding variability is crucial for developing robust radiopathomic models.
Purpose of the Study:
- To predictively map epithelium density in MRI space.
- To quantify the impact of interobserver variability from multiple pathologists on these maps.
- To develop and evaluate a consensus model for improved accuracy.
Main Methods:
- Collected clinical MRI and postsurgical tissue data from 48 patients.
- Five pathologists annotated whole-mount stained tissue slides.
- Trained pathologist-specific radiopathomic models using partial least-squares regression.
- Developed and validated a consensus model, comparing it to apparent diffusion coefficient (ADC).
Main Results:
- Interobserver agreement among pathologists ranged from 0.31 to 0.69.
- Statistically significant differences in predicted epithelium density were observed between models.
- The consensus model achieved an area under the curve of 0.80, outperforming ADC (0.71).
Conclusions:
- Radiopathomic maps of epithelium density are sensitive to pathologist annotations.
- A consensus model demonstrated superior performance and robustness.
- The consensus model's performance matched the best individual model and exceeded ADC for disease classification.
