MRI-Based Computer-Aided Diagnostic Model to Predict Tumor Grading and Clinical Outcomes in Patients With Soft Tissue...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Magnetic Resonance Imaging (MRI) is crucial for tumor characterization using advanced computer-aided diagnostic (CAD) methods.
• Soft tissue sarcomas (STSs) require accurate grading and prognostic assessment for effective treatment planning.

Purpose of the Study:

• To evaluate and validate MRI-based CAD models for differentiating low-grade and high-grade STSs.
• To investigate the prognostic capability of these models for patient survival.

Main Methods:

• A retrospective study involving 540 patients with STSs utilized 5-T MRI with T1 WI and fat-suppressed T2-weighted sequences.
• Radiomic features were extracted from manual regions of interest, alongside deep learning analysis using convolutional neural networks (CNNs).
• Support vector machines classified tumors, and a clinical-MRI nomogram integrated radiomics and deep learning signatures for grading and survival prediction.

Main Results:

• The clinical-MRI differentiation nomogram achieved an AUC of 0.870 in training and 0.855 in validation cohorts, with high accuracy, sensitivity, and specificity.
• The prognostic model demonstrated good performance for overall survival, with 3- and 5-year C-indices ranging from 0.642 to 0.722 across cohorts.

Conclusions:

• MRI-based CAD nomograms effectively classify low-grade and high-grade STSs.
• The developed MRI-based prognostic model shows favorable preoperative capacity for predicting long-term survival in STS patients.