A Hybrid Multi-Class Classification of Alzheimer Disease Based on OperativeDeep Learning Techniques: Xception-Fractalnet

Alzheimer's disease (AD) is a type of dementia that affects people as they get older and is one of the most frequent memory depletion diseases.  Early-stage  AD  identification is essential for  preventing  and intervening  the  disease  development.  But  it  is  a  challenging  task  due  to  the  complex  structure of  brain  and  its functions.  Hence,  the  research on  AD  has  increased  recently.  Therefore in  this  paper,  an  effective hybrid Xception  and  Fractalnet  based  deep learning  framework is  implemented  to  classify  the  stages  of  AD  into  five classes. To increase the performance of the classifier, effective pre-processing methods and Unet++ based segmentation technique are applied on Magnetic Resonance Imaging (MRI) images gathered from ADNI dataset. The  performance  of  the  proposed  approach is  analysed  based  on  Recall,  precision  and  accuracy  metrics.  The investigation  results  shows  that  the  proposed  technique  have  the  capacity  to  attain  98.30%  recall,  99.72% precision,  and  99.06%  accuracy  in  multiclass  classification.  The results  indicate  that  the  proposed  techniques combined with MRI images can be utilized to categorize forecast neurodegenerative brain illnesses like AD.

Alzheimer's disease (AD) is a type of dementia that affects people as they get older and is one of the most frequent memory depletion diseases.  Early-stage  AD  identification is essential for  preventing  and intervening  the  disease  development.  But  it  is  a  challenging  task  due  to  the  complex  structure of  brain  and  its functions.  Hence,  the  research on  AD  has  increased  recently.  Therefore in  this  paper,  an  effective hybrid Xception  and  Fractalnet  based  deep learning  framework is  implemented  to  classify  the  stages  of  AD  into  five classes. To increase the performance of the classifier, effective pre-processing methods and Unet++ based segmentation technique are applied on Magnetic Resonance Imaging (MRI) images gathered from ADNI dataset. The  performance  of  the  proposed  approach is  analysed  based  on  Recall,  precision  and  accuracy  metrics.  The investigation  results  shows  that  the  proposed  technique  have  the  capacity  to  attain  98.30%  recall,  99.72% precision,  and  99.06%  accuracy  in  multiclass  classification.  The results  indicate  that  the  proposed  techniques combined with MRI images can be utilized to categorize forecast neurodegenerative brain illnesses like AD.