Machine learning models using multiparametric MRI for preoperative risk stratification in endometrial cancer

Area of Science:

• Oncology
• Radiology
• Medical Imaging
• Machine Learning
• Data Science

Background:

• Accurate preoperative risk stratification is crucial for optimizing endometrial cancer management.
• Traditional methods for assessing histopathologic features and FIGO stage can be invasive and may not capture the full tumor complexity.
• Multiparameter MRI offers a non-invasive window into tumor characteristics, but extracting predictive information requires advanced analytical techniques.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) and radiomics applied to preoperative multiparameter MRIs for predicting endometrial cancer risk.
• To assess the ability of these models to differentiate between low- vs. high-risk histopathologic features and early vs. advanced FIGO stages (IA vs. IB or higher).
• To explore the potential of MRI-based radiomics for personalized preoperative risk stratification in endometrial cancer.

Main Methods:

• Retrospective analysis of 110 radiomic features extracted from preoperative multiparameter MRIs (T2WI, CE-T1WI, DWI) in 110 endometrial cancer patients.
• Development of initial models using features from individual imaging sequences and a combined model incorporating features from all three sequences.
• Performance evaluation using the area under the receiver operating characteristic curve (AUC) to assess predictive accuracy for histopathologic features and FIGO stage.

Main Results:

• The combined radiomics model, integrating 38 features (12 from T2WI, 17 from CE-T1WI, 9 from DWI), demonstrated high predictive performance.
• Achieved AUCs for predicting 5 specific histopathologic features ranged from 0.87 to 0.90, indicating strong discriminatory power.
• The model showed significant potential in differentiating between early and advanced FIGO stages and low- and high-risk histologic markers.

Conclusions:

• An MRI radiomics-based model effectively predicts high-risk histopathologic features and advanced FIGO stage in endometrial cancer.
• This non-invasive approach holds promise for preoperative risk stratification, potentially guiding personalized clinical decision-making.
• The findings support the integration of ML and radiomics into routine preoperative assessments for endometrial cancer patients.