A Machine Learning Framework for Assessment of Cognitive and Functional Impairments in Alzheimer’s Disease: Data Preprocessing and Analysis

Abstract

The neuropsychological scores and Functional Activities Questionnaire (FAQ) are significant to measure the cognitive and functional domain of the patients affected by the Alzheimer’s Disease. Further, there are standardized dataset available today that are curated from several centers across the globe that aid in development of Computer Aided Diagnosis tools. However, there are numerous clinical tests to measure these scores that lead to a challenging task for their assessment in diagnosis. Also, the datasets suffer from common missing and imbalanced data issues. In this paper, we propose a machine learning based framework to overcome these issues. Empirical results demonstrate that improved performance of Genetic Algorithm is obtained for the neuropsychological scores after Miss Forest Imputation and for FAQ scores is obtained after subjecting it to the Synthetic Minority Oversampling Technique.

Alzheimer’s Disease Assessments Optimised for Diagnostic Accuracy and Administration Time

“…Notably, our identified optimal set of assessment items included several items from the Functional Assessment Questionnaire (FAQ), which is often used in research study settings but less so in clinical practice. The effectiveness and high predictiveness of FAQ for AD diagnosis has also been identified in our previous work and the work of others [24], [23], [18].…”

“…In particular, we focus on the CFAs of the data as they form a key component of dementia clinical assessment [4]. Further, previous studies have shown that CFAs, when considered in machine learning approaches, can achieve relatively high accuracy in identifying AD severity [17], [23], [24], [19], [18], [25].…”

“…For example, [17] used false-discovery rate feature selection algorithm combined with expert input from neuropsychologists to select assessment items which are diagnostic of AD stage. [18] focused on data preprocessing issues such as missing data handling and class balancing and applied genetic algorithm to select features predictive of disease status at different stages. [19] trained a deep learning network and used Recursive Feature Elimination feature selection in building a classifer to diagnose AD stage, and found the time orientation features of the Mini Mental State Examination (MMSE) to be the most powerful predictors of the evaluated features.…”

Alzheimer’s Disease Assessments Optimized for Diagnostic Accuracy and Administration Time

“…In particular, we focus on the CFAs of the data as they form a key component of dementia clinical assessment [6] . Further, previous studies have shown that CFAs, when considered in machine learning approaches, can achieve relatively high accuracy in identifying AD severity [5] , [20] – [24] .…”

“…Another dementia progression study had performed meta-analysis across multiple studies that use CFAs to classify MCI and predict MCI-to-dementia progression, and found them to perform better for diagnostic than prognostic predictions [19] . In terms of data pre-processing, issues such as missing data handling and class balancing were addressed, and genetic algorithm was applied to select features predictive of disease status at different stages [20] . In terms of AD diagnosis, a study had trained a deep learning network and used Recursive Feature Elimination feature selection in building a classifier to diagnose AD stage, and found the time orientation features of the Mini Mental State Examination (MMSE) to be the most powerful predictors of the evaluated features [21] .…”