Dilated Saliency U-Net for White Matter Hyperintensities Segmentation using Irregularity Age Map

Abstract

White matter hyperintensities(WMH) appear as regions of abnormally high signal intensity on T2-weighted magnetic resonance image(MRI) sequences. In particular, WMH have been noteworthy in age-related neuroscience for being a crucial biomarker for Alzheimer s disease and brain aging processes. However, the automatic WMH segmentation is challenging because of the variable intensity range, size and shape. U-Net tackled this problem through the dense prediction and showed competitive performances on not only WMH segmentation/detection but also on varied image segmentation tasks, but it still accompanies a high complexity of the network architecture. In this study, we propose to use Saliency U-Net architecture and irregularity age map(IAM) to decrease the U-Net complexity without a performance loss. We trained Saliency U-Net using both T2-FLAIR MRI sequence and IAM. Since IAM guides where irregularities, in which WMH is possibly included, exist on the MRI slice, Saliency U-Net performs better than the original U-Net trained only using T2-FLAIR. The better performance was achieved with fewer parameters and shorter training time. Moreover, the application of dilated convolution enhanced Saliency U-Net to recognise the shape of large WMH more accurately by learning multi-context on MRI slices. This network named Dilated Saliency U-Net improved Dice coefficient score to 0.5588 which is the best score among our experimental models, and recorded a relatively good sensitivity of 0.4747 with the shortest train time and the least number of parameters. In conclusion, based on the experimental results, incorporating IAM through Dilated Saliency U-Net resulted an appropriate approach for WMH segmentation.