Surface-based cortical thickness and gyrification mapping with data-driven prediction of cognitive impairment in prediabetes

AbstractBackground Prediabetes is a serious health condition characterized by blood glucose levels that are higher than normal but not high enough for a diagnosis of type 2 diabetes. It remains unclear whether alterations in cortical morphology occur during the prediabetic stage. This study aimed to investigate changes in cortical thickness and gyrification in individuals with prediabetes, and to explore whether these changes can predict cognitive performance using a machine learning approach. Methods T1-weighted MRI scans were acquired from 48 patients with prediabetes and 42 healthy controls. Surface-based morphometric analyses, including cortical thickness and the local gyrification index (LGI), were performed using FreeSurfer. Group comparisons were conducted. Neuropsychological assessments included the Mini-Mental State Examination (MMSE), the Montreal Cognitive Assessment (MoCA), and the Trail Making Test (TMT) Parts A and B. Pearson correlation analyses were conducted to examine associations between morphometric changes and cognitive performance. Furthermore, PyCaret, a machine learning framework, was applied to evaluate the predictive power of cortical features and clinical variables in predicting cognitive performance. Results Compared with controls, individuals with prediabetes exhibited significantly reduced cortical thickness in the left inferior temporal gyrus (ITG) and decreased LGI in the left precentral gyrus. TMT-A and TMT-B scores were significantly higher in the prediabetes group, indicating poorer cognitive performance. Cortical thickness in the left ITG was negatively correlated with TMT-B performance (r = −0.54, 95% CI: −0.71 to −0.31, p = 0.0001). Machine learning analysis identified the Gradient Boosting classifier as the best-performing model (AUC = 0.87, accuracy=0.80). Conclusion Our findings suggest that cortical alterations in the ITG and precentral gyrus are evident during the prediabetic stage and relate to early cognitive dysfunction. These results highlight the potential of combining neuroimaging biomarkers and AI models have the potential for early detection and intervention in prediabetes-associated cognitive decline.