320 Deep-Learning Cortical Thickness Analysis as Predictor for Shunt Surgery Effectiveness in Possible iNPH Patients

Abstract

INTRODUCTION: While optimal treatment for iNPH is quite known, diagnosis remains a topic of debate, with different procedures suggested as potential predictors for shunt surgery effectiveness. Various tests, biomarkers, and neuroimaging techniques can be used; however, the diagnostic accuracy is below optimal when performed with neuroimaging techniques alone and not always cost-effective when achieved with CSF biomarkers. Quantitative computational methods for cortical thickness analysis have become available in the last decade, and they only require regularly available T1-weighted brain MR images. METHODS: We analyzed 294 total patients referred to our clinic from January 2015 until December 2021. After exclusion criteria, the final sample consisted of 98 possible iNPH patients: 64 were diagnosed as probable iNPH and underwent VPS surgery, while 34 did not receive a surgical indication after diagnostic procedures. We investigated differences in cortical thickness for all possible iNPH patients using a public deep-learning based neuroimaging pipeline composed by an advanced semantic segmentation neural network architecture and automated surface-based cortical thickness analysis. RESULTS: In 38 total patients with a negative CSFTT or not responsive to shunt surgery, a significant localized cortical thinning was seen in the lateral surface of the frontal and parietal lobes (bilateral superior and middle frontal gyrus, bilateral pars opercularis, bilateral superior parietal gyrus, bilateral supramarginal gyrus, left inferior parietal gyrus, bilateral precuneus). CONCLUSIONS: Preoperative cortical thickness is a feasible analysis that might be useful in defining the optimal therapeutic plan for possible iNPH patients. A large multicentric study is encouraged to elucidate specific patterns and ratios of cortical thickness that might be used as potential predictors for shunt surgery effectiveness.