Deep learning-based super-resolution dynamic contrast-enhanced radiomics model for predicting NSMP endometrial cancer

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Endometrial cancer (EC) diagnosis relies on accurate molecular profiling.
• Nonspecific molecular profile (NSMP) EC requires effective noninvasive prediction methods.
• Dynamic contrast-enhanced (DCE) MRI coupled with radiomics shows potential for cancer characterization.

Purpose of the Study:

• To evaluate a deep learning-based super-resolution (SR) DCE radiomic model for predicting NSMP EC.
• To compare the performance of SR-DCE radiomics against original-resolution (OR) DCE radiomics.
• To assess the clinical utility of different machine learning algorithms in this context.

Main Methods:

• 140 EC patients were divided into training and testing cohorts.
• Deep learning SR techniques enhanced original-resolution DCE images.
• Radiomic features were extracted from both SR and OR images.
• Logistic regression, support vector machine, and multilayer perceptron models were developed and evaluated using AUC and DCA.

Main Results:

• The SR-DCE radiomic model significantly outperformed the OR-DCE model (P < 0.05).
• In the testing set, SR-DCE models achieved higher AUC values (e.g., 0.841 for LR) compared to OR-DCE models (e.g., 0.637 for LR).
• LR and SVM algorithms showed superior performance within the SR-DCE models compared to MLP.

Conclusions:

• Deep learning-based SR reconstruction enhances the diagnostic performance of DCE radiomics for NSMP EC.
• SR-DCE radiomics presents a potential noninvasive preoperative predictive tool for NSMP EC.
• This approach could aid in personalized treatment strategies for EC patients.