Radiomics-based machine-learning method for prediction of distant metastasis from soft-tissue sarcomas.

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Distant metastasis (DM) significantly impacts prognosis in soft-tissue sarcoma (STS).
• Accurate preoperative prediction of DM is crucial for effective treatment planning.

Purpose of the Study:

• To develop and validate a radiomics-based machine learning (ML) model for predicting DM in STS.
• To evaluate the performance of different ML algorithms and feature selection methods.

Main Methods:

• Seventy-seven STS cases were analyzed, split into training (n=54) and validation (n=23) sets.
• Feature selection methods including ReliefF, LASSO, and UDFS were compared with classifiers like RF, LOG, KNN, and SVMs.
• The synthetic minority oversampling technique (SMOTE) was employed to address data imbalance.

Main Results:

• The combination of LASSO feature selection and SVM classification, utilizing SMOTE, demonstrated superior performance.
• This optimized model achieved an AUC of 0.9020 and an accuracy of 91.30% in the validation dataset.

Conclusions:

• Radiomics-based ML models show promise for predicting DM in STS.
• This predictive capability can inform and personalize treatment strategies for STS patients.