Related Experiment Video
Updated: Jun 11, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.7K
Navigating the gray zone: Machine learning can differentiate malignancy in PI-RADS 3 lesions.
Emre Altıntaş1, Ali Şahin2, Seyit Erol3
1Department of Urology, Selcuk University School of Medicine, Konya, Turkey.
Urologic Oncology
|September 29, 2024
Summary
Machine learning, specifically the random forest model, accurately predicts prostate cancer in PI-RADS 3 lesions. This approach can help prevent unnecessary biopsies by identifying malignancy using MRI and clinical data.
Area of Science:
- Urology
- Radiology
- Machine Learning
- Oncology
Background:
- Prostate cancer diagnosis often involves assessing PI-RADS 3 lesions on multiparametric MRI (mpMRI).
- Distinguishing between benign and malignant PI-RADS 3 lesions is challenging, leading to potential overtreatment or delayed diagnosis.
- Machine learning (ML) offers a promising avenue for improving diagnostic accuracy in these ambiguous cases.
Purpose of the Study:
- To develop and evaluate machine learning models for predicting the probability of prostate cancer in PI-RADS 3 lesions.
- To identify key clinical and radiological parameters that contribute to accurate malignancy prediction.
- To assess the performance of different ML algorithms, including the random forest model, for this diagnostic task.
Main Methods:
- A cohort of patients with PI-RADS 3 lesions on mpMRI who underwent fusion biopsy was retrospectively analyzed.
- Radiological parameters (e.g., Apparent Diffusion Coefficient (ADC), Ktrans, lesion size) and clinical data (e.g., age, PSA density, neutrophil-lymphocyte ratio (NLR)) were collected.
- Six distinct machine learning models were trained and validated using these parameters to predict cancer probability.
Main Results:
- The random forest model demonstrated superior performance, achieving the highest accuracy (0.86), F1 score (0.91), and AUC (0.92).
- SHAP analysis identified tumor ADC, tumor ADC/contralateral ADC ratio, and PSA density as the most influential predictors of malignancy.
- Inflammatory markers like systemic inflammatory index and NLR showed higher predictive value than traditional parameters such as total PSA and lesion size.
Conclusions:
- The developed random forest model effectively predicts malignancy in PI-RADS 3 lesions, potentially reducing the need for unnecessary biopsies.
- This ML-based approach shows clinical utility and warrants further validation in multicenter studies with larger patient cohorts.
- Integration of advanced imaging and clinical data through ML can enhance diagnostic precision in prostate cancer assessment.
