An interpretable machine learning model for identifying granulation patterns in somatotroph tumors: A multi-center...

Area of Science:

• Radiology
• Machine Learning
• Oncology

Background:

• Distinguishing sparsely granulated (SGST) and densely granulated (DGST) somatotroph tumors is crucial for acromegaly management.
• Current methods may lack non-invasive accuracy for preoperative classification.

Purpose of the Study:

• To develop and validate non-invasive, interpretable machine learning (ML) models for differentiating SGST and DGST somatotroph tumors.
• Leverage preoperative multiparametric MRI and clinical data for tumor subtype classification.

Main Methods:

• Retrospective analysis of 201 patients with somatotroph tumors across four institutions.
• Extraction of 3004 radiomic features from MRI, followed by feature selection (LASSO, Boruta).
• Development and evaluation of six ML algorithms, with a support vector machine (SVM) model selected and interpreted using SHAP.

Main Results:

• The SVM model, integrating eight radiomic and five clinical features, achieved AUCs of 0.828 (training) and 0.820 (validation).
• SHAP analysis identified key radiomic and clinical predictors, enhancing model interpretability and transparency.

Conclusions:

• A multi-center, radiomics-based SVM model demonstrates high accuracy and generalizability for preoperative somatotroph tumor classification.
• This interpretable, non-invasive tool can improve personalized treatment planning and acromegaly management.