Related Experiment Video
Updated: Aug 29, 2025
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
Prostate cancer classification using radiomics and machine learning on mp-MRI validated using co-registered
Ryan Alfano1, Glenn S Bauman2, Jose A Gomez3
1Baines Imaging Research Laboratory, 790 Commissioners Rd E, London, ON N6A 5W9, Canada; Lawson Health Research Institute, 750 Base Line Rd E, London, ON N6C 2R5, Canada; Western University, Department of Medical Biophysics, 1151 Richmond St., London, ON N6A 3K7, Canada.
This study developed a machine learning system for prostate cancer (PCa) detection using multi-parametric MRI (mp-MRI). The system achieved an AUC of 0.80, addressing biases in computer-aided diagnosis models.
Area of Science:
- Radiomics and Machine Learning in Medical Imaging
- Prostate Cancer Detection and Diagnosis
Background:
- Multi-parametric magnetic resonance imaging (mp-MRI) shows promise for prostate cancer (PCa) detection but has limitations.
- Computer-aided diagnosis (CAD) systems aim to improve PCa detection but often suffer from inherent biases in model development and validation.
Purpose of the Study:
- To develop and validate a radiomics-based machine learning system for classifying prostate cancer (PCa) versus non-PCa tissue on mp-MRI.
- To investigate and mitigate potential biases in CAD system development for PCa detection.
Main Methods:
- Clinically significant PCa regions from histology were mapped to mp-MRI using a validated registration algorithm.
- A novel sampling algorithm was employed to match non-PCa regions by shape and size, reducing bias.
- Inter-zonal variability biases were assessed, and a 5-feature Naïve-Bayes classifier was trained and validated using leave-one-patient-out cross-validation.
Main Results:
- The Naïve-Bayes classifier achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.80.
- Classifier performance was invariant to shape differences between PCa and non-PCa lesions (AUC: 0.82 vs 0.82).
- Models trained and tested in the peripheral zone showed lower performance (AUC: 0.75) compared to the central gland (AUC: 0.95).
Conclusions:
- A radiomics-based machine learning system was successfully developed to differentiate PCa from non-PCa tissue on mp-MRI.
- The system was validated using accurately co-registered histology with a measured target registration error.
- Considerations for mitigating biases in future CAD system development for PCa detection were highlighted.
