16 Nov 20. Featured preprint

Automatic segmentation of white matter hyperintensities from brain magnetic resonance images in the era of deep learning & big data – a systematic review.

Link to paper on Preprints

Authors

Ramya Balakrishnan, Maria Valdes Hernandez, Andrew Farrall

Abstract

Background:

White matter hyperintensities (WMH), of presumed vascular origin, are visible & quantifiable neuroradiological markers of brain parenchymal change.

These changes may range from damage secondary to inflammation & other neurological conditions, through to healthy ageing.

Fully automatic WMH quantification methods are promising, but still, traditional semi-automatic methods seem to be preferred in clinical research.

We systematically reviewed the literature for fully automatic methods developed in the last five years, to assess what are considered state-of-the-art techniques, as well as trends in the analysis of WMH of presumed vascular origin.

Method:

We registered the systematic review protocol with the International Prospective Register of Systematic Reviews (PROSPERO), registration number - CRD42019132200.

We conducted the search for fully automatic methods developed from 2015 to July 2020 on Medline, Science direct, IEE Explore, & Web of Science.

We assessed risk of bias & applicability of the studies using QUADAS 2.

Results:

The search yielded 2327 papers after removing 104 duplicates.

After screening titles, abstracts & full text, 37 were selected for detailed analysis.

Of these, 16 proposed a supervised segmentation method, 10 proposed an unsupervised segmentation method, & 11 proposed a deep learning segmentation method.

Average DSC values ranged from 0.538 to 0.93, being the highest value obtained from a deep learning segmentation method.

Only four studies validated their method in longitudinal samples, & eight performed an additional validation using clinical parameters.

Only 8/37 studies made available their method in public repositories.

Conclusions:

Although deep learning methods reported highly accurate results, we found no evidence that favours them over the more established k-NN, linear regression & unsupervised methods in this task.

Data & code availability, bias in study design & ground truth generation influence the wider validation & applicability of these methods in clinical research.

Keywords

• Deep learning
• FLAIR hyperintensities
• Supervised segmentation
• Unsupervised segmentation
• White matter lesions
• White matter hyperintensities