Premium
Deep learning based Alzheimer's disease early diagnosis using T2w segmented gray matter MRI
Author(s) -
Basheera Shaik,
Ram M Satya Sai
Publication year - 2021
Publication title -
international journal of imaging systems and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.359
H-Index - 47
eISSN - 1098-1098
pISSN - 0899-9457
DOI - 10.1002/ima.22553
Subject(s) - computer science , artificial intelligence , cognitive impairment , mri scan , alzheimer's disease neuroimaging initiative , pattern recognition (psychology) , data set , imaging biomarker , alzheimer's disease , neuroimaging , nuclear medicine , magnetic resonance imaging , cognition , medicine , neuroscience , pathology , radiology , disease , psychology
Diagnosing Alzheimer's disease at early stage required an effective classification mechanism to differentiate mild cognitive impairment from cognitive normal and AD. In this paper, we used data set collected from ADNI and OASIS. Instead of using the whole volume of the MRI, high informative slices are selected using entropy. The selected slices are pre‐processed by removing unwanted tissues using skull stripping algorithm and extracted gray matter using EICA. In this work, we used CNN model with inception blocks to extract deep features from the GM slices used to predict AD at early stage. The model avoids data leakage by considering all the slices of an MRI as a unit. The model trained with 80% of ADNI subject MRI volumes and tested with the remaining 20% subject MRI volumes, to provide great variance in training and testing data, the model further tested with OASIS data sets. 10‐fold cross‐validation is used to test the model without biasing. The model performance is evaluated using accuracy. The model achieves 98.73%, 100%, 93.72%, and 95.6% of accuracy for differentiating CN‐MCI, CN‐AD, AD‐ MCI and CN‐MCI and AD. At 10‐fold cross‐validation it gives 92.92 ± 3%, 98 ± 2%, 90 ± 4% and 94.9 ± 2% accuracy to differentiate CN‐MCI, CN‐AD, AD‐MCI, and CN‐MCI‐AD using ADNI. We further tested the model with 135 MRI volumes selected from OASIS data set, we achieved 92%, 91.76%, 88.23%, 81.48% of accuracy with CN‐AD, MCI‐AD, CN‐MCI, and three‐way classification. The model gives good accuracy and sensitivity of early AD Diagnosis.