Improve the diagnosis of idiopathic normal pressure hydrocephalus by combining abnormal cortical thickness and ventricular morphometry.

Background: The primary imaging markers for idiopathic Normal Pressure Hydrocephalus (iNPH) emphasize morphological measurements within the ventricular system, with no attention given to alterations in brain parenchyma. This study aimed to investigate the potential effectiveness of combining ventricular morphometry and cortical structural measurements as diagnostic biomarkers for iNPH. Methods: A total of 57 iNPH patients and 55 age-matched healthy controls (HC) were recruited in this study. Firstly, manual measurements of ventricular morphology, including Evans Index (EI), z-Evans Index (z-EI), Cella Media Width (CMW), Callosal Angle (CA), and Callosal Height (CH), were conducted based on MRI scans. Cortical thickness measurements were obtained, and statistical analyses were performed using surface-based morphometric analysis. Secondly, three distinct models were developed using machine learning algorithms, each based on a different input feature: a ventricular morphology model (LVM), a cortical thickness model (CT), and a fusion model (All) incorporating both features. Model performances were assessed using 10-fold cross validation and tested on an independent dataset. Model interpretation utilized Shapley Additive Interpretation (SHAP), providing a visualization of the contribution of each variable in the predictive model. Finally, Spearman correlation coefficients were calculated to evaluate the relationship between imaging biomarkers and clinical symptoms. Results: iNPH patients exhibited notable differences in cortical thickness compared to HC. This included reduced thickness in the frontal, temporal, and cingulate cortices, along with increased thickness in the supracentral gyrus. The diagnostic performance of the fusion model (All) for iNPH surpassed that of the single-feature models, achieving an average accuracy of 90.43%, sensitivity of 90.00%, specificity of 90.91%, and Matthews correlation coefficient (MCC) of 81.03%. This improvement in accuracy (6.09%), sensitivity (11.67%), and MCC (11.25%) compared to the LVM strategy was significant. Shap analysis revealed the crucial role of cortical thickness in the right isthmus cingulate cortex, emerging as the most influential factor in distinguishing iNPH from HC. Additionally, significant correlations were observed between the typical triad symptoms of iNPH patients and cortical structural alterations. Conclusion: This study emphasizes the significant role of cortical structure changes in the diagnosis of iNPH, providing a novel insights for assisting clinicians in improving the identification and detection of iNPH.