Exploring Robustness of Cortical Morphometry in the presence of white matter lesions, using Diffusion Models for Lesion Filling

TL;DR

White matter lesions bias MRI-based cortical thickness, motivating lesion-filling approaches. The authors develop diffusion-model–based lesion filling using conditional and unconditional DDPMs with 2D and pseudo-3D U-Nets, and assess robustness across five cortical-thickness pipelines. They show that deep-learning-based pipelines (DL+DiReCT, ANTsPyNet, FastSurfer) exhibit substantially greater robustness to lesions than traditional methods, with conditional 3D diffusion models and circle-mask training yielding the best lesion-fill quality. The results suggest that DL-based morphometry may reduce or even obviate the need for lesion filling in many contexts, while providing guidance on when and how to apply lesion filling in lesion populations.

Abstract

Cortical thickness measurements from magnetic resonance imaging, an important biomarker in many neurodegenerative and neurological disorders, are derived by many tools from an initial voxel-wise tissue segmentation. White matter (WM) hypointensities in T1-weighted imaging, such as those arising from multiple sclerosis or small vessel disease, are known to affect the output of brain segmentation methods and therefore bias cortical thickness measurements. These effects are well-documented among traditional brain segmentation tools but have not been studied extensively in tools based on deep-learning segmentations, which promise to be more robust. In this paper, we explore the potential of deep learning to enhance the accuracy and efficiency of cortical thickness measurement in the presence of WM lesions, using a high-quality lesion filling algorithm leveraging denoising diffusion networks. A pseudo-3D U-Net architecture trained on the OASIS dataset to generate synthetic healthy tissue, conditioned on binary lesion masks derived from the MSSEG dataset, allows realistic removal of white matter lesions in multiple sclerosis patients. By applying morphometry methods to patient images before and after lesion filling, we analysed robustness of global and regional cortical thickness measurements in the presence of white matter lesions. Methods based on a deep learning-based segmentation of the brain (Fastsurfer, DL+DiReCT, ANTsPyNet) exhibited greater robustness than those using classical segmentation methods (Freesurfer, ANTs).

Figures (14)

• Figure 1: Diffusion process from Data to Noise and reverse process from Noise to Data.
• Figure 2: Creation of the training dataset
• Figure 3: Training step involving the conditional mixture model. An MR-image and its corresponding lesion mask are sampled from the dataset. Alternatively, with a 50% probability, a random circle mask is sampled instead. This mask is used to void the portion of the image, which requires inpainting. Additionally, a random timestep $t$ and random noise matching the image shape are sampled. These are used to generate a noisy image as described in Section \ref{['sec:DDPM']}. The mask, the noisy image, and the voided image are concatenated and fed into the diffusion model, which aims to predict the sampled noise. The predicted and sampled noise are used to calculate the MSE.
• Figure 4: T1w before lesion filling with conditional mixture model
• Figure 5: T1w after lesion filling with conditional mixture model