Data-Driven Machine-Learning Quantifies Differences in the Voiding Initiation Network in Neurogenic Voiding...

Area of Science:

• Neuroscience
• Urology
• Medical Imaging

Background:

• Neurogenic lower urinary tract dysfunction (NLUTD) and voiding dysfunction (VD) are common in female multiple sclerosis (MS) patients.
• Understanding the neural underpinnings of voiding control in MS is crucial for effective treatment.

Purpose of the Study:

• To quantify the relative importance of brain regions in the Voiding Initiation Network (VIN) associated with reduced functional connectivity (FC) in female MS patients with NLUTD and VD.
• To utilize a data-driven machine-learning approach for quantifying these differences.

Main Methods:

• Functional magnetic resonance imaging (fMRI) and simultaneous urodynamic testing were performed on 27 ambulatory female MS patients with NLUTD (15 voiders, 12 VD).
• The VIN was identified from averaged fMRI activation maps.
• Four machine-learning algorithms were used to optimize the area under the curve (AUC) for identifying the relative importance of brain regions within the VIN.

Main Results:

• The VIN showed stronger FC in frontal regions for voiders and stronger disassociation in cerebellar regions for patients with VD.
• Machine learning algorithms, particularly 'random forests' (AUC 0.86) and 'partial least squares' (AUC 0.89), effectively differentiated the groups.
• Key brain regions contributing to these differences included frontal and cingulate areas, with a global effect observed across 186 of 227 VIN regions.

Conclusions:

• Distinct functional connectivity patterns in the VIN differentiate MS patients who can void from those with VD.
• Machine learning effectively identifies brain regions contributing to these differences.
• This knowledge may enable patient phenotyping for targeted therapies like cortical modulation.