Development and validation of a deep learning-based automatic segmentation and classification of cerebral white...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Cerebral Small Vessel Disease

Background:

• White matter hyperintensities (WMH) are key neuroimaging markers of cerebral small vessel disease.
• Accurate quantification of WMH is crucial for diagnosing and managing cognitive impairment.

Purpose of the Study:

• To develop and validate a deep learning model for simultaneous, automatic WMH segmentation and classification.
• To assess the model's performance in patients with cognitive impairment.

Main Methods:

• A retrospective study utilizing a deep learning framework (UNet with Resnet-34 encoder) for segmentation.
• A multi-task multiple instance learning framework (MTMIL) for classification based on the Fazekas scale.
• Internal and external validation datasets comprising patients with cognitive impairment.

Main Results:

• The segmentation model achieved Dice scores between 0.72 and 0.88.
• Classification accuracy for periventricular and deep WMH ranged from 0.68 to 0.88.
• The model demonstrated robust performance on both internal and external datasets.

Conclusions:

• The developed deep learning model exhibits high performance in WMH segmentation and classification.
• The model shows potential as an accurate diagnostic support tool for quantified WMH evaluation.
• This technology can aid in the management of cognitive impairment associated with WMH.