MRI radiomics-based machine-learning classification of bone chondrosarcoma.

Area of Science:

• Orthopedic Oncology
• Radiology
• Machine Learning in Medicine

Background:

• Distinguishing low-grade from high-grade cartilaginous bone tumors is crucial for treatment planning.
• Magnetic Resonance Imaging (MRI) is a key modality for evaluating bone tumors.
• Radiomic analysis of MRI can potentially enhance diagnostic accuracy.

Purpose of the Study:

• To assess the diagnostic performance of machine learning (ML) in classifying low-grade versus high-grade cartilaginous bone tumors.
• To evaluate radiomic features extracted from unenhanced MRI for tumor discrimination.
• To compare ML performance against expert musculoskeletal radiologist interpretation.

Main Methods:

• Retrospective analysis of 58 patients with histologically confirmed cartilaginous tumors.
• Radiomic features were extracted from T1- and T2-weighted MRI images.
• A Random Forest wrapper and AdaboostM1 classifier were used for feature selection and model training, with performance validated via cross-validation and an independent test set.

Main Results:

• The ML model, using 4 selected radiomic features from T1-weighted images, achieved 85.7% accuracy on the training set and 75% on the test set.
• Area under the receiver operating characteristic curve (AUC) values were 0.85 for training and 0.78 for testing.
• The ML classifier's performance was not significantly different from that of an experienced radiologist (81.3% accuracy).

Conclusions:

• Machine learning utilizing radiomic parameters from unenhanced MRI demonstrates robust diagnostic performance for classifying cartilaginous bone tumors.
• This ML approach can serve as a valuable tool for preoperative tumor characterization.
• Further validation may integrate ML into routine clinical practice for improved diagnostic decision-making.