WMH-DualTasker: A Weakly Supervised Deep Learning Model for Automated White Matter Hyperintensities Segmentation and...

Area of Science:

• Neuroimaging and computational neuroscience
• Artificial intelligence in medicine
• Vascular cognitive impairment research

Background:

• White matter hyperintensities (WMH) are key neuroimaging indicators of cognitive decline risk.
• Current WMH assessment methods (visual scales, volumetric segmentation) have limitations in descriptive power and scalability.
• Supervised segmentation requires extensive manual annotation, hindering large-scale studies.

Purpose of the Study:

• To develop an automated, minimally supervised deep learning model for accurate and holistic WMH severity quantification.
• To assess the clinical utility of the model in identifying mild cognitive impairment (MCI) and predicting dementia conversion.

Main Methods:

• Developed WMH-DualTasker, a deep learning model performing simultaneous voxel-wise segmentation and visual rating score prediction.
• Employed self-supervised learning with transformation-invariant consistency, using clinical WMH visual ratings as the sole supervisory signal.
• Evaluated model performance on MICCAI-WMH and SINGER datasets for volumetric quantification and on an external dataset for clinical rating agreement.

Main Results:

• WMH-DualTasker achieved volumetric quantification performance comparable or superior to existing supervised methods (MICCAI-WMH: Dice=0.602, SINGER: Dice=0.608).
• The model demonstrated strong agreement with clinical visual rating scales (SINGER: MAE=1.880, K=0.77).
• WMH severity metrics improved prediction for MCI classification (AUC=0.718) and MCI conversion (AUC=0.652) on the ADNI dataset.

Conclusions:

• WMH-DualTasker significantly reduces reliance on manual annotations, enabling efficient and scalable WMH severity quantification.
• The model enhances the assessment and management of vascular cognitive impairment associated with WMH.
• This approach holds potential for advancing preventive and precision medicine in cognitive health.