# Cortical thickness analysis combined with CSF dynamics improves diagnostic stratification in idiopathic normal pressure hydrocephalus

## TL;DR

Combining brain imaging and spinal fluid analysis improves diagnosis and treatment decisions for a brain condition called idiopathic normal pressure hydrocephalus.

## Contribution

A new diagnostic approach combining cortical thickness and CSF dynamics achieves high predictive accuracy for iNPH treatment outcomes.

## Key findings

- Cortical thickness variations in specific brain regions correlate with poor treatment outcomes in iNPH patients.
- A machine learning model integrating cortical thickness and CSF data achieved 90% positive predictive value for treatment response.
- Combining neuroimaging and CSF analysis improves diagnostic precision over traditional methods.

## Abstract

The diagnostic landscape for idiopathic normal-pressure hydrocephalus is intricate, and there is a pressing need for accurate and cost-effective methods. Because of the lack of accurate diagnostic and prognostic quantitative biomarkers, the frequent presence of comorbidities, and the limited understanding of the pathophysiology of the disorder, only a minority of patients receive disease-specific treatment. While traditional neuroimaging offers insights, its isolated diagnostic precision can be enhanced. Emerging quantitative methods analyzing cortical thickness based on standard T1-weighted brain MR images offer new diagnostic possibilities. We analyzed 294 patients referred to our clinic from January 2015 until December 2022. After the exclusion criteria, the final sample consisted of 100 possible iNPH patients. Of these, 71 underwent ventriculoperitoneal shunt surgery, while 29 did not qualify post-evaluation. Cortical thickness was assessed using an advanced deep-learning neuroimaging pipeline. For predictive modeling, we employed a comprehensive set of Machine Learning algorithms, including Distributed Random Forests, Extremely Randomized Trees, Generalized Linear Model with Regularization, Gradient Boosting Machines, Extreme Gradient Boosting machines, and a fully connected multi-layer Artificial Neural Network. These algorithms were strategically combined into a Super Learner ensemble approach to harness their collective predictive power. Among patients with negative CSFTT outcomes or subpar VPS surgery responses, distinct cortical variations emerged, particularly in the caudal middle frontal, rostral middle frontal, superior frontal, and superior parietal regions. Our Super Learner model, integrating CSF dynamics and cortical thickness data, achieved a 90% positive predictive value, signifying a tangible advancement over traditional measures. Analyzing preoperative cortical thickness emerges as a viable strategy for streamlining therapeutic decisions for potential iNPH patients. Future endeavors should focus on large-scale multicentric studies to further delineate specific cortical thickness patterns, potentially enhancing the prediction accuracy for VPS surgery outcomes.