Independent Validation of the volBrain CognitiveTree Automated MRI Pipeline in Alzheimer's Disease Diagnosis.
Authors
Affiliations (4)
Affiliations (4)
- Faculty of Medicine, Department of Anatomy, Karabük University, Karabük, Türkiye.
- Faculty of Medicine, Department of Anatomy, Istanbul Arel University, Istanbul, Türkiye.
- Faculty of Health Sciences, Department of Physiotherapy and Rehabilitation, Istanbul Bilgi University, Istanbul, Türkiye.
- Faculty of Medicine, Department of Anatomy, Sanko University, Gaziantep, Türkiye.
Abstract
In Alzheimer's disease, significant structural changes are observed in various brain regions, particularly in the cerebral cortex. The aim of this study is to contribute to the diagnosis of the disease by conducting comparative analyses between Alzheimer's patients and a healthy control group. Magnetic resonance images from Alzheimer's disease and healthy control groups, comprising 100 individuals in each group, were retrospectively obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Image analysis was performed using the CognitiveTree module within the volBrain software. In this context, percentage values for the right, left, and total volumes of 231 distinct anatomical regions of the brain were analyzed using various statistical methods. In the classification analysis of Alzheimer's disease and healthy control groups, the highest accuracy rate of 90% was achieved using the Random Forest and Bagging Classifier models. The parameter with the greatest impact on the classification analysis was the total volume of the inferior lateral ventricle. In the machine learning analysis conducted using a 20% test set, the highest AUC value of 0.95 was obtained with the Random Forest algorithm. A higher AUC value was observed compared to the Bagging model; however, the DeLong test determined that this difference was not statistically significant (p > 0.05). Findings from the classification analysis of Alzheimer's disease and healthy control groups indicate that machine learning-based models offer an effective approach for the rapid and practical prediction of the disease. It is anticipated that the integration of these methods into clinical practice, along with an evaluation of their applicability across different populations, could contribute to diagnostic processes.