A novel CNN based Alzheimer’s disease classification using hybrid enhanced ICA segmented gray matter of MRI

• Published in: Computerized medical imaging and graphics Vol. 81; p. 101713
• Main Authors: Basheera, Shaik, Satya Sai Ram, M.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.04.2020
• Subjects: Alzheimer’s disease; CNN; Gaussian mixture model; Independent component analysis; Internal Medicine; Other; CNN; Alzheimer’s disease; Gaussian mixture model; Independent component analysis
• ISSN: 0895-6111; 1879-0771; 1879-0771
• DOI: 10.1016/j.compmedimag.2020.101713

[Display omitted] •We have proposed a model for early diagnosis of AD from MCI and CN.•Skull stripping is performed to remove the unwanted tissues like skull, dura, eyes, skin and fat from MRI Slices.•Hybrid enhanced Independent Component Analysis is used to collect Segmented Grey Matter MRI Image.•Proposed model is tested by subject wise spitting, 10 fold cross validation is used to analyse the classifier performance.•We obtained better classification accuracy of CN-MCI is 98.0 %. Our model gives better accuracy at early diagnosis of AD to assist physician. Predicting Alzheimer’s Disease (AD) from Mild Cognitive Impairment (MCI) and Cognitive Normal (CN) has become wide. Recent advancement in neuroimaging in adoption with machine learning techniques are especially useful for pattern recognition of medical imaging to assist the physician in early diagnosis of AD. It is observed that the early abnormal brain atrophy and healthy brain atrophy are same. In our endeavor, we proposed a model that differentiation MCI and CN more accurately to escalate early diagnosis of AD. In this paper, we applied both binary and multi class classification, 4463 Slide are divided in to two groups one for training and another for testing at subject level, achieves 100 % of accuracy, 100 % of sensitivity and 100 % of Specificity in the case of AD-CN. 96.2 % of accuracy, 93 % Sensitivity and 100 % Specificity in the case of AD-MCI. 98.0 % of accuracy, 96 % of sensitivity, 100 specificity in the case of CN-MCI. 86.7 % accuracy, 89.6 % of sensitivity, 86.61 % of specificity in the case of AD-MCI-CN. The model is further tested using 10 fold cross validation and obtained 98.0 % of accuracy, to differentiate CN and MCI. Our proposed framework generated results are significantly improving prediction of AD from MCI and CN than compare to the previous work flows and used to differentiate the AD at early stage.