Unsupervised machine learning identifies clinically relevant patterns of CSF dynamic dysfunction in normal pressure...

Area of Science:

• Neurology
• Radiology
• Medical Imaging Analysis

Background:

• Idiopathic normal pressure hydrocephalus (iNPH) presents with non-specific symptoms, complicating diagnosis.
• Accurate diagnosis of iNPH is crucial for effective treatment and patient management.

Purpose of the Study:

• To investigate if data-driven patterns of cerebrospinal fluid (CSF) distribution can predict iNPH diagnosis.
• To assess the utility of these patterns in predicting treatment response in iNPH patients.

Main Methods:

• Non-negative Matrix Factorization (NMF) applied to CSF probability maps from MRI scans.
• Analysis of CSF distribution patterns and component loads to predict diagnosis and clinical outcomes.
• Comparison with manual measurements like Evan's index and callosal angle.

Main Results:

• NMF identified 7 CSF distribution patterns in 98 iNPH patients and 98 controls.
• The automated NMF model achieved high accuracy (96%/97%) in predicting iNPH diagnosis.
• Patterns associated with disproportionately enlarged subarachnoid space hydrocephalus (DESH) showed the highest diagnostic potential.

Conclusions:

• NMF-based CSF distribution patterns accurately differentiate iNPH patients from controls.
• The NMF algorithm, particularly when focusing on DESH features, shows promise as a diagnostic aid.
• The proposed tool complements, rather than replaces, clinical expertise for iNPH diagnosis.