Class Imbalance Applied to Medical Neuroimaging for Classification of Alzheimer’s Disease

Abstract

Class imbalance is an issue that naturally occurs when a database is sparse or incomplete. This can occur

in medical diagnostics when a large percentage of tests ran to return negative results rather than a positive.

Classifification models are sensitive to an imbalanced training set, and training on one can cause undesirable

biases. This work presents an overview of the effffects of class imbalance on classifification models in

Alzheimer’s detection utilizing voxel based-morphometry (VBM). MRI scans are processed by FreeSurfer

where cerebral volumetric and thickness are taken as feature vectors. The effcts of class imbalances on

multiple machine learning models were compared to one another. Furthermore, different biomarkers were

studied for their effect on different metrics of trained models. The classifification models were trained to detect

the following categories: Alzheimer’s disease (AD), mild cognitive impairment (MCI), and normal controls

(NC). SVM, KNN, MLP, Random Forest, etc. algorithms were evaluated for the prediction analysis. It was

observed that class imbalance did not produce any significant effects on the disease classification process.