Related Experiment Video
Updated: May 22, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
116
Interpretable multiparametric MRI radiomics-based machine learning model for preoperative differentiation between
Wenjun Zhou1,2, Zhangcheng Liu1,3, Jindong Zhang1
1Department of Urology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Frontiers in Oncology
|May 20, 2025
Summary
This study developed an interpretable machine learning model using multiparametric MRI radiomics to accurately differentiate benign from malignant prostate masses before surgery. The random forest model showed high predictive ability, aiding clinical decision-making.
Area of Science:
- Radiology and Medical Imaging
- Machine Learning in Medicine
- Oncology and Urology
Background:
- Accurate preoperative differentiation of prostate masses is crucial for appropriate treatment planning.
- Multiparametric MRI (mpMRI) provides detailed imaging information for prostate cancer assessment.
- Radiomics and machine learning (ML) offer potential for enhanced diagnostic accuracy.
Purpose of the Study:
- To develop and validate an interpretable ML model based on mpMRI radiomics for preoperative differentiation of prostate masses.
- To assess the model's performance in distinguishing between benign and malignant prostate lesions.
Main Methods:
- Retrospective analysis of mpMRI data from 567 patients across two hospitals.
- Extraction of radiomic features from T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) using PyRadiomics.
- Feature selection via LASSO, followed by building and validating five ML models, including Random Forest (RF), with interpretability analysis using SHAP.
Main Results:
- A total of 2,632 radiomic features were reduced to 18, forming the basis for five ML models.
- The RF model achieved high predictive performance with an Area Under the Curve (AUC) of 0.929 (internal) and 0.852 (external validation).
- Calibration and decision curve analyses confirmed the clinical utility of the RF model, with SHAP explaining feature contributions.
Conclusions:
- mpMRI-derived radiomic features combined with ML enable accurate preoperative evaluation of prostate mass malignancy.
- The interpretable ML model, particularly the RF model, demonstrates significant potential for clinical application in prostate cancer diagnostics.
- SHAP analysis enhances model transparency, potentially facilitating broader clinical adoption.
