Multiparametric MRI-Based Deep Learning Radiomics Model for Assessing 5-Year Recurrence Risk in Non-Muscle Invasive
Haolin Huang1,2, Yiping Huang3, Joshua D Kaggie4
1School of Biomedical Engineering, Fourth Military Medical University, Xi'an, Shaanxi, China.
Journal of Magnetic Resonance Imaging : JMRI
|August 21, 2024
Summary
Integrating multiparametric MRI radiomics and deep learning with clinical factors significantly improves 5-year recurrence risk assessment for non-muscle-invasive bladder carcinoma (NMIBC). This approach enhances prediction accuracy beyond traditional models for better NMIBC management.
Area of Science:
- Oncology
- Radiology
- Medical Imaging
Background:
- Accurate 5-year recurrence risk assessment is critical for managing non-muscle-invasive bladder carcinoma (NMIBC).
- Current models, such as the European Organization for Research and Treatment of Cancer (EORTC) model, demonstrate suboptimal performance in predicting NMIBC recurrence.
- There is a need for improved methods to enhance the accuracy of NMIBC recurrence risk stratification.
Purpose of the Study:
- To evaluate the efficacy of integrating multiparametric MRI (mp-MRI) with clinical factors for improving 5-year recurrence risk assessment in NMIBC.
- To compare the performance of a combined clinical-radiomics-deep learning model against traditional models.
Main Methods:
- A retrospective study involving 191 patients with NMIBC who underwent mp-MRI and had at least 5-year follow-up.
- Radiomics and deep learning (DL) features were extracted from the combined region of interest (cROI) on mp-MRI.
- Four models were developed: clinical, cROI-based radiomics, DL, and a combined clinical-radiomics-DL (CRDL) model. Statistical analyses included ROC curves, Cox regression, and SHAP values.
Main Results:
- The CRDL model demonstrated superior performance in the testing cohort, achieving an AUC of 0.909 for 5-year recurrence assessment in NMIBC.
- The CRDL model also achieved the highest Harrell's concordance index (0.804) for estimating recurrence-free survival.
- SHapley Additive Explanations (SHAP) analysis indicated that radiomics features played a substantial role (22%) in NMIBC recurrence prediction.
Conclusions:
- Integrating cROI-based radiomics and DL features from preoperative mp-MRI with clinical factors significantly enhances 5-year recurrence risk assessment for NMIBC.
- The developed CRDL model offers a more accurate and robust tool for predicting NMIBC recurrence compared to existing methods.
- This advanced approach holds promise for optimizing NMIBC patient management and treatment strategies.
