Interpretable radiomics model based on dual-layer spectral CT iodine maps for predicting microsatellite instability...

To develop and validate an interpretable radiomics model using dual-layer detector spectral CT (DLSCT)-derived iodine maps to preoperatively predict Microsatellite instability (MSI) status in colorectal cancer (CRC).

A total of 255 CRC patients who underwent DLSCT were retrospectively included from two independent centers. Tumor iodine concentrations (IC) and normalized iodine concentrations (NIC) were measured and calculated. Iodine maps were analyzed to extract radiomics features. Clinical, radiomics, and combined clinical- radiomics models were constructed respectively. Model performance was evaluated using receiver operating characteristic curve and decision curve analysis. SHapley Additive exPlanation (SHAP) was employed to enhance interpretability.

Overall, 255 patients (mean 60.5 ± 12.9 years; 122 males) were divided into training (n = 149), internal testing (n = 60), and independent external testing (n = 46) sets. High-level MSI (MSI-H) status was present in 38 patients (14.9 %). CRC with microsatellite instability-low or microsatellite stability exhibited higher IC values in the arterial phase and NIC in the venous phase than those in MSI-H patients (p < 0.05). For MSI prediction, the clinical- radiomics models outperformed the clinical model (AUC = 0.84 vs 0.67, p＜0.001). Decision curve analysis confirmed the higher net benefit of the clinical- radiomics model compared to the other models across probability thresholds ranging from 0 to 0.40. SHAP analysis highlighted a wavelet-based feature and a shape feature as key contributors.

The interpretable clinical-radiomics model based on iodine maps effectively predicts MSI status in CRC preoperatively, facilitating personalized decision-making in clinical practice.