Automated MRI radiomics pipeline predicts acromegaly treatment response to somatostatin receptor ligands

Automated MRI radiomics pipeline predicted somatostatin receptor ligand response in acromegaly: best 3-feature logistic regression had out-of-fold AUC 0.798 (95% CI 0.671-0.918), correctly classifying 39 of 49 patients (79.6%).

• Automated tumor segmentation achieved a Dice coefficient of 0.795 on combined contrast-enhanced T1-weighted and T2-weighted MRI, decreasing to 0.765 in the acromegaly-specific cohort.
• Best model was a 3-feature logistic regression with out-of-fold AUC 0.798 (95% CI 0.671-0.918), balanced accuracy 0.793, and F1-score 0.815.
• The most consistent radiomic feature was log-sigma-3-0-mm-3D_glcm_ClusterShade, appearing in 98% of cross-validation folds and contributing most to discriminative signal.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.