White Matter Lesions Segmentation of FLAIR MRI Based on U-Net Using Cascade Filtering

White matter lesions (WMLs) are critical neurological markers associated with a range of conditions, including dementia, reduced mobility, and mental health issues. Their prevalence increases with age, making early and accurate detection vital for effective medical intervention. Magnetic Resonance Imaging (MRI) serves as a cornerstone in medical imaging for this purpose, particularly using Fluid-Attenuated Inversion Recovery (FLAIR) sequences. However, the quality of MRI scans can vary due to machine brand, magnetic field strength, and image noise, affecting the accuracy of WMLs segmentation. While existing automated segmentation methods have made strides, they still lag behind expert manual segmentation in terms of accuracy. This paper introduces an innovative approach to WMLs segmentation in FLAIR images using a U-Net architecture enhanced with cascade filtering. Our proposed filtering model employs two filters-Gaussian and Wiener-arranged in serial to improve image quality while preserving essential features. We conducted an ablation study for hyperparameter tuning and compared the performance of our model against three filtering methods and baseline models with no-filtering. Our proposed method not only elevates image quality but also minimizes feature loss, thereby achieving a better Hausdorff distance segmentation score than other models. This research offers a robust and reliable tool for WMLs segmentation, with significant implications for both clinical practice and ongoing research in MRI medical image analysis.