A Novel Segment White Matter Hyperintensities Approach for Detecting Alzheimer

Segmentation has been an effective step that needs to be done before the classification or detection of an anomaly like Alzheimer's on a brain scan. Segmentation helps detect pixels of the same intensity or volume and group them together as one class or region, where in that particular region of interest (ROI) can be concentrated on, rather than focusing on the entire image. In this paper White Matter Hyperintensities (WMH) is taken as a strong biomarker that supports and determines the presence of Alzheimer's. As the first step a proper segmentation of the lesions has to be carried out. As pointed out in various other research papers, when the WMH area is very small or in places like the Septum Pellucidum the detection of the lesion is hard to find. To overcome such problem areas a very optimized and accurate Threshold would be required to have a precise segmentation to detect the area of localization. This would help in proper detection and classification of the Anomaly. In this paper an elaborate comparison of various thresholding techniques has been done for segmentation. A novel idea for detection of Alzheimer's has been presented in this paper, which encompasses the effectiveness of an optimized and adaptive technique. The Unet architecture has been taken as the baseline model with an adaptive kernel model embedded within the architecture. Various state-of-the-art technologies have been used with the dataset and a comparative study has been presented with the current architecture used in the paper. The lesion segmentation in narrow areas has accurately been detected compared to the other models and the number of false positives has been reduced to a great extent.