Study on the effect of extreme learning machine and its variants in differentiating Alzheimer conditions from selective regions of brain MR images

In this study, an attempt has been made to analyze the effect of the Extreme Learning Machine (ELM) classifier and its variants in the differentiation of Healthy Controls (HC) and Alzheimer's Disease (AD) using structural MR images. For this, sub-anatomic brain structures, namely Corpus Callosum (CC) and Lateral Ventricles (LV) are segmented and characterized using morphometric features. Significant features from these regions are subjected to ELM, online sequential ELM, and Self-adaptive Differential Evolution ELM (SaDE-ELM) classifiers for the differentiation of HC and AD. The activation functions and the number of hidden neurons in ELM classifiers are tuned by evaluating the performance using standard metrics. Results indicate that ELM and its variant classifiers are able to identify AD using morphometric features from CC and LV. ELM classifiers achieve an accuracy greater than 90% using the sigmoid activation function for both regions. Among the ELM variants, SaDE-ELM attains a maximum sensitivity of 97% and 94% using LV and CC features respectively with the lowest number of hidden neurons. The extracted features from LV demonstrate higher discriminative power than CC in AD classification. However, a specificity greater than 95% is observed in ELM for CC features. As the proposed approach is able to characterize and differentiate the morphometric alterations in CC and LV due to AD, the study seems to be clinically significant for the differentiation of HC and AD.

Automated summary

In this study, the authors analyzed the effectiveness of the Extreme Learning Machine (ELM) classifier and its variants in differentiating between Healthy Controls (HC) and Alzheimer's Disease (AD) using structural MR images. They found that ELM classifiers achieved an accuracy of over 90% in identifying AD, with SaDE-ELM classifier demonstrating the highest sensitivity.