Automated Aqueductal CSF Flow Analysis in Spontaneous Intracranial Hypotension: Hemodynamic Quantification and...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Spontaneous intracranial hypotension (SIH) diagnosis relies on clinical and MRI signs, lacking objective physiological monitoring tools.
• Cine phase-contrast MRI (PC-MRI) quantifies aqueductal CSF dynamics but faces challenges in SIH due to small aqueduct size and reduced pulsatility.

Purpose of the Study:

• To develop and evaluate an automated AI framework for robust CSF dynamics quantification and waveform morphology analysis in SIH using cine PC-MRI.
• To explore the potential of aqueductal CSF waveform morphology as an exploratory biomarker for SIH.

Main Methods:

• An end-to-end automated framework integrating cascade localization-segmentation (Tiny YOLOv4, MultiResUNet) and physiology-informed pulsatility-based segmentation (PUBS).
• 1D convolutional neural networks (1D-CNNs) were used to extract waveform morphology features from cardiac-cycle velocity data.
• The framework was tested on 59 cine PC-MRI scans from controls, pre-treatment SIH, and post-treatment recovery groups.

Main Results:

• The automated framework significantly improved segmentation robustness and reduced quantification errors compared to baseline methods.
• PUBS refinement enhanced diagnostic performance for hemodynamic parameters.
• Exploratory analysis showed potential for waveform morphology to distinguish SIH from controls, though out-of-sample classification performance was modest (AUC 0.646).

Conclusions:

• Aqueductal CSF waveform morphology may contain SIH-related physiological information beyond flow magnitude.
• Current AI-based morphology features are exploratory biomarker candidates, not standalone diagnostic tools.
• Larger independent cohorts are necessary to confirm reproducibility, physiological relevance, and clinical utility.