Bypassing Pre-processing Method in Alzheimer’s Disease Diagnosing using Deep Learning Instance Segmentation

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that will cause the memory loss of patient and will progressively lead to loss of bodily function that will eventually lead to death. Therefore, diagnosing AD accurately is critical to provide the patients with suitable treatment to delay the progression of AD as well to facilitate the treatment interventions. Recent studies are more dependent on the Deep Learning Semantic Segmentation method to perform the Alzheimer's Disease diagnosis. However, semantic segmentation will segment every single pixel in the images which will affect the precision of the small targets like hippocampal region in MRI images, even though the overall loss is low enough. Therefore, a Deep Learning Instance Segmentation is introduced into the Alzheimer’s disease diagnosis field without using any pre-processing method. In this research, the Mask R-CNN will be used to localize the hippocampal region to do the segmentation, and then classified it as AD or NC. The dataset UTM_ADNI_RAW will be used in this study. The proposed method applied on UTM_ADNI_RAW shows the high accuracy of 92.67%. These results show that the proposed method to segment the hippocampal region without requiring pre-processing techniques has a good accuracy in classifying AD and NC subjects. In conclusion, the proposed Mask R-CNN generated a good result on segmenting the hippocampal region without requiring any pre-processing techniques.