Multi-Features Fusion in Multi-plane MRI Images for Alzheimer’s Disease Classification
Authors: Cucun Very Angkoso, Hapsari Peni Agustin Tjahyaningtijas, Yudhi Adrianto, Anggraini Dwi Sensusiati, I Ketut Eddy Purnama, Mauridhi Hery Purnomo
Number of views: 86
Alzheimer's disease (AD) is characterized by severe memory loss, typical in dementia. This disease has serious public health consequences (high incidence, prevalence, and mortality rates), as well as significant health and social costs. Therefore, AD should be recognized as a disease rather than a natural occurrence that affects everyone, allowing for early detection and treatment to begin before the situation worsens. Previous studies have focused on shrinkage in certain brain locations (typically the hippocampus, cerebral cortex, and brain ventricles), ruling out the potential of atrophy in other areas. Therefore, our study proposed an algorithm for detecting and classifying AD which uses the whole brain area. While previous studies focused mostly on a single plane, we propose to exploit all planes (multi-plane) of the MRI, including the axial, coronal, and sagittal planes, to obtain more detailed whole-brain imaging characteristics. A multi-feature fusion of texture-based feature extraction, including First-order statistics (FOS), Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LPB), and Gray Level Run Length Matrix (GLRLM), were used as input of our classifier. We used T1 and T2-weighted structural MRI from the Alzheimer's Disease Neuroimaging Initiative (ADNI) MRI database and Brain-Atrophy (BA) MRI from the Airlangga University Hospital (AUH) to evaluate our proposed method. A significantly higher prediction accuracy confirms the effectiveness of the proposed method. As all of the extracted features were employed, the accuracy of the AD and CN classifiers with the multi-plane approach can be improved by up to 20.41% compared to the single-plane approach. In addition, our proposed method has the highest accuracy of 0.967 in binary classification tasks and 0.867 in multiclass classification tasks, outperforming previous works reported in the related references. Furthermore, the multi-plane analysis strategy has proven superior to the single-plane approach in all evaluations.