An End to End Deep Learning model for Alzheimer s Disease Classification through Biomarker Segmentation

Abstract

Alzheimer s disease is an irreversible neuro-degenerative sickness causing cogni- newlinetive decline in older people. Early diagnosis of Alzheimer s may help doctors delay newlinedisease progression in a patient. Computer vision treats this as a classification problem newlinethat classifies a brain scan into various stages of the disease, such as cognitively normal newline, mild cognitive impairment, and alzheimer s disease. However, classical image pro- newlinecessing and machine learning techniques must undergo a formidable feature extraction newlinephase that would sometimes require domain expertise, which is subtle in the case of newlinemedical imaging. This limitation has drawn the neuroimaging researcher s attention newlinetoward deep learning where feature extraction is an automated process. Most of the newlineclassification models based on deep learning use transfer learning. They exploit the newlinepre-trained weights of the state-of-the-art classification models, which are trained upon newlinenatural images such as the ImageNet dataset. Despite accomplishing state-of-the-art newlineperformance, these models accuracy is suspected to be over-fitting. Hence an end-to- newlineend model trained upon the brain MR Images from scratch has been awaited. newlineOur research addresses above mentioned gap by proposing a deep-learning approach newlinethat performs Alzheimer s classification through biomarker segmentation. Though newlinemany researchers believe that the anatomical changes occur only at later stages of the newlinedisease, little evidence exists to vocalize that they institute at the early stage and persist newlinetill the final stage spreading across various brain regions. Therefore identifying those newlinebrain regions that get affected due to Alzheimer s, plays a crucial role in diagnosing newlineAlzheimer s. Our first contribution investigates the three significant brain regions called newlinethe gray matter , white matter, and cerebro spinal fluid as the volumetric biomarkers newlineof Alzheimer s and their contribution in classifying various stages of the disease. The multiclass classification of 4 stages of the disease, known as cognitively