Brain ventricle morphology markers in predicting shunt surgery outcome in idiopathic normal-pressure hydrocephalus.
Authors
Affiliations (7)
Affiliations (7)
- School of Computing, University of Eastern Finland, Kuopio, 70211, Finland. [email protected].
- Faculty of Health Sciences, University of Eastern Finland, Kuopio, 70211, Finland.
- Institute of Biomedicine, University of Eastern Finland, Kuopio, 70211, Finland.
- Institute of Clinical Medicine-Radiology, University of Eastern Finland, Kuopio, 70211, Finland.
- Combinostics Ltd., Tampere, 33100, Finland.
- A.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, 70211, Finland.
- School of Computing, University of Eastern Finland, Kuopio, 70211, Finland.
Abstract
Idiopathic normal pressure hydrocephalus (iNPH) is characterized by a clinical triad of symptoms: abnormal gait, memory problems, and urinary incontinence. Neuroimaging plays a crucial role in diagnosing iNPH. However, current radiological markers, though indicative, are not definitive, suggesting the limited capacity of these indices to capture mechanisms associated with iNPH and the reversibility of the symptoms. This study aims to (1) determine the geometric features of the lateral ventricles, (2) develop a quantitative method for three-dimensional analysis, and (3) test the ability to predict response to shunt surgery. By examining these features as potential diagnostic markers, this research seeks to enhance the understanding of morphometric characteristics in iNPH, thereby paving the way for improved patient selection for surgical intervention. Our study contained 170 patients (95 shunt responders and 75 non-responders) from the Kuopio NPH registry. Our inclusion criteria required pre-surgery and one-year post-surgery symptom assessments alongside preoperative anatomical magnetic resonance imaging (MRI). Volumetric brain segmentations were performed using cNeuro software on T1-MRI images, followed by the generation of 3D lateral ventricle meshes for geometric feature extraction. The classification task employed the LogitNet machine learning model to analyze 27 geometric features. Model performance evaluation utilized repeated nested cross-validation (10 rounds) with five inner folds for parameter tuning and five outer folds for model evaluation. Additionally, we generated a ranking of feature importance based on the LogitNet L1 regularization coefficients. Our analysis revealed that LogitNet achieved an AUC of 0.661 (SD = 0.066) performance across 10 rounds of cross-validation in predicting the shunt surgery response. The most prominent feature contributing to the model’s prediction was asphericity. Our analysis suggests that the proposed set of features, especially asphericity, effectively captures valuable information linked to the reversibility of iNPH.