A Novel Approach for the Prediction of Conversion from Mild Cognitive Impairment to Alzheimer's disease using MRI Images

• Published in: Advances in Electrical and Computer Engineering Vol. 17; no. 2; pp. 113 - 122
• Main Authors: AYUB, A., FARHAN, S., FAHIEM, M. A., TAUSEEF, H.
• Format: Journal Article
• Language: English
• Published: Suceava Stefan cel Mare University of Suceava 01.05.2017
• Subjects: Accuracy; Algorithms; Alzheimer's disease; Automation; Biomarkers; Brain research; Cerebrospinal fluid; Classification; Cognitive ability; computer aided diagnosis; Conversion; Diagnosis; feature extraction; image analysis; image classification; Image processing; Impairment; Literature reviews; Longitudinal studies; Machine learning; Magnetic resonance imaging; Medical imaging; Methods; Pattern recognition; Search algorithms
• ISSN: 1582-7445; 1844-7600; 1844-7600
• DOI: 10.4316/AECE.2017.02015

The main objective of our research is to introduce an approach that uses noninvasive MRI images to predict the conversion from mild cognitive impairment to Alzheimer's disease at an early stage. It detects normal controls that are likely to develop Alzheimer's disease and mild cognitive impairment patients that are likely to establish Alzheimer's disease within two years or, contrarily, their stage remains same. The proposed approach uses two types of features i.e. volumetric features and textural features. Volumetric features consist of volume of grey matter, volume of white matter and volume of cerebrospinal fluid. A total of 364 textural features have been calculated. To avoid the curse of dimensionality, textural features are reduced to 15 features using gain ratio, a ranking based search algorithm. All features are tested against four classifiers i.e. AODEsr, VFI, RBF and LBR. Leave-OneOut cross validation strategy is used for the evaluation of proposed approach. Results show accuracy of 98.33% with volumetric features and 100% with textural features using VFI and LBR. Our approach is innovative because of its higher accuracy results as compared to existing approaches yet with a smaller feature set.