Development of a Data-driven Prediction Model for the Evolution of White Matter Hyperintensities using Deep Learning

This is ongoing work done in collaboration with the Centre for Clinical Brain Sciences, University of Edinburgh. Principal investigators:

• Muhammad Febrian Rachmadi (MFR) [BIA, RIKEN CBS]
• Maria del C. Valdés-Hernández (MCVH) [CCBS, UoE]

Previous studies have shown that the volume & shape of White Matter Hyperintensities (WMHs) on a patient may decrease, stay unchanged, or increase over time (i.e., evolution of WMH). WMHs are associated with dementia, Alzheimer’s Disease (AD), stroke, and multiple sclerosis (MS). In this study, we aim at predicting the evolution of WMHs. This is challenging because:

• the rate of WMH evolution varies across studies and patients,
• it involves a high degree of uncertainty, and
• influencing clinical factors are poorly understood.

Figure 1. Predicting Disease Evolution Map (DEM) by using neural networks..

• Clinicians can estimate the size and location of WMH in the future in relation to clinical health and disease indicators, for ultimately designing more effective therapeutic interventions for dementia/AD.
• Similar models can be applied for any disease progressions.

• Irregularity Map (IM) was proposed for richer representation of WMHs.

• Disease Evolution Map (DEM) for representing the evolution of WMHs across an entire brain MRI scan was proposed.

• Generative Adversarial Networks (GANs) were employed to make the predicted DEM is as close as to the real DEM.

• Probabilistic U-Net with Adversarial Training (i.e., GANs) was proposed to capture uncertainties in predicting shrinking and growing WMHs.

• Volume Interval Estimation (VIE) was proposed for better interpretation and higher confidence in the estimation of future volume of WMHs.

• Rachmadi, MF, et al. “Probabilistic Deep Learning with Adversarial Training and Volume Interval Estimation-Better Ways to Perform and Evaluate Predictive Models for White Matter Hyperintensities Evolution.” International Workshop on PRedictive Intelligence In MEdicine. Springer, Cham, 2021.
• Rachmadi, MF, et al. “Automatic spatial estimation of white matter hyperintensities evolution in brain MRI using disease evolution predictor deep neural networks.” Medical Image Analysis 63 (2020): 101712.
• Rachmadi, MF, et al. “Limited one-time sampling irregularity map (LOTS-IM) for automatic unsupervised assessment of white matter hyperintensities and multiple sclerosis lesions in structural brain magnetic resonance images.” Computerized Medical Imaging and Graphics 79 (2020): 101685.
• Rachmadi, MF, et al. “Predicting the evolution of white matter hyperintensities in brain MRI using generative adversarial networks and irregularity map.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2019.
• Rachmadi, MF, et al. “Automatic irregular texture detection in brain MRI without human supervision.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2018.

• JSPS Kakenhi Grant-in-Aid for Research Activity Start-up (Project No. 20K23356) (MFR)
• Row Fogo Charitable Trust (Grant No. BRO-D.FID3668413) (MCVH)
• Wellcome Trust (patient recruitment, scanning, primary study Ref No. WT088134/Z/09/A)
• Fondation Leducq (Perivascular Spaces Transatlantic Network of Excellence)
• EU Horizon 2020 (SVDs@Target)
• MRC UK Dementia Research Institute at the University of Edinburgh (Wardlaw programme)
• Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) from the Japan Agency for Medical Research and Development AMED (JP15dm0207001).