CT radiomics nomogram combining vertebral, muscle features shows AUC 0.956 for T2DM diagnosis

This retrospective study investigated the diagnostic value of CT-based radiomics features from vertebral bodies (VB) and paravertebral muscles (PVM) for type 2 diabetes mellitus (T2DM). The study included 160 cases: 80 patients with T2DM and 80 non-diabetic patients. Regions of interest for VB and PVM were delineated, and radiomics features were extracted. Patients were divided into a training group (n=112) and a validation group (n=48) in a 7:3 ratio. Key radiomics features were selected using independent samples t-test and LASSO algorithm. A k-nearest neighbor classifier was used to establish radiomics models, and radiomics scores were calculated. Clinical risk factors were identified via univariate and multivariate logistic regression to build a clinical model. A nomogram was developed by integrating the radiomics score with the clinical model using multivariate logistic regression. Diagnostic performance was evaluated using AUC, calibration curves, and clinical decision curves, with Delong's test for model comparison. In the training set, AUCs were 0.902 for the VB radiomics model, 0.948 for the PVM radiomics model, 0.952 for the VB-PVM combined radiomics model, 0.857 for the clinical model, and 0.956 for the radiomics-clinical combined model. In the validation set, corresponding AUCs were 0.873, 0.880, 0.894, 0.758, and 0.926. The radiomics-clinical combined model showed the best diagnostic performance. Calibration and decision curves indicated the nomogram had good consistency and clinical applicability. The study concluded the combined radiomics and clinical model based on CT images of VB and PVM has good diagnostic value for T2DM differential diagnosis.

Spotting type 2 diabetes early is crucial for getting ahead of it, but current methods aren't perfect. What if a scan you might already be getting for other reasons could offer a clue? This study looked at whether the subtle patterns in a CT scan of your spine and the muscles alongside it could help tell if someone has type 2 diabetes. The researchers analyzed scans from 80 people with the condition and 80 without. They used a computer to pick out specific texture features from the bone and muscle. They then combined these scan features with basic clinical information, like known risk factors, into a single diagnostic tool called a nomogram. The result? The combined model, using both the scan data and clinical info, performed the best at telling the two groups apart. It wasn't perfect, but it showed good accuracy and consistency in both the group used to build it and a separate group used to test it. This suggests that everyday CT images contain hidden signals about metabolic health that, when read the right way, could become a useful piece of the puzzle for doctors trying to diagnose type 2 diabetes.