A Predictive MRI Radiomics Model for Histologic Differentiation in Soft Tissue Sarcomas

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Soft tissue sarcomas (STS) are a heterogeneous group of tumors requiring accurate subtyping for effective treatment.
• Non-invasive methods for differentiating STS subtypes and benign mimics are crucial for improving diagnostic accuracy and patient management.
• Current diagnostic approaches may involve invasive procedures, highlighting the need for advanced imaging techniques.

Purpose of the Study:

• To develop and validate a radiomics-based classification model using pre-treatment MRI.
• To non-invasively differentiate between major soft tissue sarcoma (STS) subtypes and a benign intramuscular myxoma.
• To assess the model's performance using combined T1 and T2 MRI sequences.

Main Methods:

• Retrospective analysis of 332 patients with biopsy-proven STS subtypes and intramuscular myxoma.
• Extraction of 1240 radiomic features from pre-treatment T1-weighted and T2-weighted MRI sequences.
• Training and validation of an XGBoost classifier using bootstrap framework and nested cross-validation.

Main Results:

• The combined T1+T2 radiomics model achieved an accuracy of 0.68 ± 0.04 and an AUC of 0.92 ± 0.02.
• Highest balanced accuracy was observed for intramuscular myxoma (0.91), dedifferentiated liposarcoma (0.84), and leiomyosarcoma (0.83).
• SHAP analysis identified key radiomic features correlating with specific tumor subtypes, aligning with pathological characteristics.

Conclusions:

• Radiomics analysis of pre-treatment MRI provides a robust method for non-invasive classification of STS subtypes.
• The developed model demonstrates significant clinical potential for enhancing diagnostic confidence.
• This approach can aid in optimizing triage strategies for patients with soft tissue tumors.