An Explainable Radiomics-Based Classification Model for Sarcoma Diagnosis.

Area of Science:

• Radiomics and Medical Imaging
• Artificial Intelligence in Oncology
• Machine Learning for Disease Classification

Background:

• Sarcoma tumor classification relies heavily on subjective interpretation of MRI scans.
• Need for objective, automated methods to improve diagnostic accuracy and efficiency.
• Explainable AI (XAI) offers potential for transparent and trustworthy clinical decision support.

Purpose of the Study:

• To develop and validate an explainable, radiomics-based machine learning framework for automated sarcoma tumor classification using MRI.
• To reduce clinician dependence on subjective image interpretation.
• To enhance the interpretability of AI models in medical diagnosis.

Main Methods:

• Extraction of 851 radiomic features, including wavelet descriptors, from 186 MRI scans of sarcoma patients.
• Training a Random Forest classifier with hyperparameter tuning via nested cross-validation.
• Utilizing Feature Importance and Local Interpretable Model-agnostic Explanations (LIME) for model interpretability.

Main Results:

• The radiomics model achieved an F1-score of 0.742 and an accuracy of 0.724 on the test set.
• LIME analysis identified texture and wavelet-based radiomic features as key predictors.
• The framework demonstrated effective classification of sarcoma tumors from healthy tissue.

Conclusions:

• The proposed explainable AI framework enables accurate and interpretable classification of sarcomas from MRI.
• This non-invasive approach supports earlier, personalized, and precision-driven cancer diagnosis.
• The study underscores the value of explainable AI in enhancing clinical decision-making security.