AlzCoGame: A Serious Game using Machine Learning for the Early Diagnosis of Alzheimer's Disease (Preprint)
UNSTRUCTURED Older people can develop cognitive problems with a range of symptoms including memory, perception, and difficulty in solving problems known as Alzheimer's disease (AD). Early diagnosis of Mild Cognitive Impairment (MCI), which can lead to AD, plays an important role in the management of patients to slow the decline in cognitive function, as treatments are effective early in the course of the disease. For this, advanced computer technologies can provide a tool for early detection of AD and for predicting the progression of the disease. This article presents a serious game formed of 16 mini-games that aimed to detect AD or MCI at the mild stage, based on gamification techniques and machine learning (ML) by overcoming the traditional testing limitations. The gamified cognitive tool, named AlzCoGame, assesses the main cognitive domains considered to be the most relevant indicators for the cognitive impairment diagnosis: working memory, episodic memory, executive functions, visuospatial orientation, concentration, and attention. Six predictive ML models such as Random Forest (RF), Naïve Bayes (NB), Decision Tree (DT), Logistic Regression (LR), AdaBoost (AB), and Extra Tree (ET) have been implemented using the AlzCoGame dataset. To validate the model's performance, we used the K-fold cross-validation and classification metrics (accuracy, precision, specificity, sensitivity, F1-Score, and ROC curve (receiver operating characteristic)). The results obtained indicate that Random Forest classifier gave better performances with average Sensitivity = 0.89, Specificity = 0.93, Accuracy = 0.92, 1-Score = 0.91 and ROC = 0.91. We can conclude that the inclusion of machine learning techniques and serious games can be used to improve some aspects of the clinical diagnosis of cognitive impairment. Moreover, clinical trials are necessary to prove the effects of this gamified program on cognitive functions and to assess usability measures.