Diagnosis of Suspected Pituitary Microadenomas: The Complementary Role of Habitat-based Non-enhanced MRI Radiomics Features.

To develop and validate a combined model for detecting pituitary microadenomas in suspected cases using non-contrast T1-weighted sagittal (T1SAG) magnetic resonance imaging (MRI), and to assess its performance as a non-contrast imaging-based tool for clinical decision support.This retrospective two-center study enrolled 519 patients with suspected pituitary microadenomas, divided into a training set (n = 321), an internal validation set (n = 80), and an external test set (n = 118). Radiomics features were extracted from both the entire pituitary gland and its habitat subregions, generated via unsupervised K-means clustering. Eight predictive models were constructed using logistic regression: a clinical model, an intra-pituitary radiomics model, three individual habitat models (Habitat 1, 2, and 3), an early-fusion model combining all habitats, an ensemble model integrating the probability outputs of the three habitat models, and a combined model that integrated clinical, intra-pituitary, and the optimal fusion model. Model performance was evaluated using receiver operating characteristic curves (ROC), calibration curves, and decision curve analysis.Among all models, the combined model demonstrated the best predictive performance across all three cohorts, achieving areas under the curve (AUCs) of 0.885 (95% CI: 0.850-0.920) in the training set, 0.788 (95% CI: 0.690-0.885) in the internal validation set, and 0.796 (95% CI: 0.712-0.880) in the external test set. The combined model also achieved the highest accuracy and sensitivity across all datasets.The combined prediction model showed robust performance in distinguishing pituitary microadenomas in suspected cases, offering enhanced clinical decision support while reducing dependence on contrast-enhanced MRI examinations.