Mild Cognitive Impairment Detection Using Machine Learning Models Trained on Data Collected from Serious Games

A unified Foot and Mouth Disease dataset for Uganda: evaluating machine learning predictive performance degradation under varying distributions

In Uganda, the absence of a unified dataset for constructing machine learning models to predict Foot and Mouth Disease outbreaks hinders preparedness. Although machine learning models exhibit excellent predictive performance for Foot and Mouth Disease outbreaks under stationary conditions, they are susceptible to performance degradation in non-stationary environments. Rainfall and temperature are key factors influencing these outbreaks, and their variability due to climate change can significantly impact predictive performance. This study created a unified Foot and Mouth Disease dataset by integrating disparate sources and pre-processing data using mean imputation, duplicate removal, visualization, and merging techniques. To evaluate performance degradation, seven machine learning models were trained and assessed using metrics including accuracy, area under the receiver operating characteristic curve, recall, precision and F1-score. The dataset showed a significant class imbalance with more non-outbreaks than outbreaks, requiring data augmentation methods. Variability in rainfall and temperature impacted predictive performance, causing notable degradation. Random Forest with borderline SMOTE was the top-performing model in a stationary environment, achieving 92% accuracy, 0.97 area under the receiver operating characteristic curve, 0.94 recall, 0.90 precision, and 0.92 F1-score. However, under varying distributions, all models exhibited significant performance degradation, with random forest accuracy dropping to 46%, area under the receiver operating characteristic curve to 0.58, recall to 0.03, precision to 0.24, and F1-score to 0.06. This study underscores the creation of a unified Foot and Mouth Disease dataset for Uganda and reveals significant performance degradation in seven machine learning models under varying distributions. These findings highlight the need for new methods to address the impact of distribution variability on predictive performance.

Unveiling peripheral neuropathy and cognitive dysfunction in diabetes: an observational and proof-of-concept study with video games and sensor-equipped insoles

Background

Proactive screening for cognitive dysfunction (CD) and peripheral neuropathy (PNP) in elderly patients with diabetes mellitus is essential for early intervention, yet clinical examination is time-consuming and prone to bias.

Objective

We aimed to investigate PNP and CD in a diabetes cohort and explore the possibility of identifying key features linked with the respective conditions by machine learning algorithms applied to data sets obtained in playful games controlled by sensor-equipped insoles.

Methods

In a cohort of patients diagnosed with diabetes (n=261) aged over 50 years PNP and CD were diagnosed based on complete physical examination (neuropathy symptom and disability scores, and Montreal Cognitive Assessment). In an observational and proof-of-concept study patients performed a 15 min lasting gaming session encompassing tutorials and four video games with 5,244 predefined features. The steering of video games was solely achieved by modulating plantar pressure values, which were measured by sensor-equipped insoles in real-time. Data sets were used to identify key features indicating game performance with correlation regarding CD and PNP findings. Thereby, machine learning models (e.g. gradient boosting and lasso and elastic-net regularized generalized linear models) were set up to distinguish patients in the different groups.

Results

PNP was diagnosed in 59% (n=153), CD in 34% (n=89) of participants, and 23% (n=61) suffered from both conditions. Multivariable regression analyses suggested that PNP was positively associated with CD in patients with diabetes (adjusted odds ratio = 1.95; 95% confidence interval: 1.03-3.76; P=0.04). Predictive game features were identified that significantly correlated with CD (n=59), PNP (n=40), or both (n=59). These features allowed to set up classification models that were enriched by individual risk profiles (i.e. gender, age, weight, BMI, diabetes type, and diabetes duration). The obtained models yielded good predictive performance with the area under the receiver-operating-characteristic curves reaching 0.95 for CD without PNP, 0.83 for PNP without CD, and 0.84 for CD and PNP combined.

Conclusions

The video game-based assessment was able to categorize patients with CD and/or PNP with high accuracy. Future studies with larger cohorts are needed to validate these results and potentially enhance the discriminative power of video games.

Use of artificial intelligence techniques for detection of mild cognitive impairment: A systematic scoping review

AIMS AND OBJECTIVES

The objective of this scoping review is to explore the types and mechanisms of Artificial intelligence (AI) techniques for detecting mild cognitive impairment (MCI).

BACKGROUND

Early detection of MCI is crucial because it may progress to Alzheimer's disease.

DESIGN

A systematic scoping review.

METHODS

Five-step framework of Arksey and O'Malley was used following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for scoping reviews checklist. A total of 11 databases (PubMed, EMBASE, CINAHL, Cochrane Library, Scopus, Web of Science, IEEE Explore, Science.gov, ACM digital library, arXIV and ProQuest) was used to search from inception till 17th December 2021. Grey literature and reference list were searched. Articles screening and data charting were conducted by two independent reviewers.

RESULTS

There were a total of 70 articles included from 2011 to 2022 across 16 countries. Four types of AI techniques were found, namely machine learning (ML), deep learning (DL), fuzzy logic (FL) and technique combinations. Herein, ML detects similar pattern within preselected data to classify subjects into non-MCI or MCI groups. Meanwhile, DL performs classification based on data patterns and data analyses are performed by themselves. Furthermore, FL utilises human-defined rules to decide the degree to which a person has MCI. A combination of AI techniques enhances the feature preparation phase for ML or DL to perform accurate classification.

CONCLUSION

Although AI-based MCI detection tool is critical for healthcare decision-making, clinical utility and risks remain underexplored. Hopefully, this review equips clinicians with background AI knowledge to address these clinical concerns. Hence, future research should explore more techniques and representative datasets to improve AI development.

RELEVANCE TO CLINICAL PRACTICE

Results of this review can increase the knowledge of AI-based MCI detection tools.

REVIEW REGISTRATION

This study protocol was registered in the Open Science Framework Registries (https://osf.io/45rdt).

The limitations and challenges in the assessment of executive dysfunction associated with real-world functioning: The opportunity of serious games

Nowadays, there is a broad range of methods for detecting and evaluating executive dysfunction ranging from clinical interview to neuropsychological evaluation. Nevertheless, a critical issue of these assessments is the lack of correspondence of the neuropsychological test's results with real-world functioning. This paper proposes serious games as a new framework to improve the neuropsychological assessment of real-world functioning. We briefly discuss the contribution and limitations of current methods of evaluation of executive dysfunction (paper-and-pencil tests, naturalistic observation methods, and Information and Communications Technologies) to inform on daily life functioning. Then, we analyze what are the limitations of these methods to predict real-world performance: (1) A lack of appropriate instruments to investigate the complexity of real-world functioning, (2) the vast majority of neuropsychological tests assess well-structured tasks, and (3) measurement of behaviors are based on simplistic data collection and statistical analysis. This work shows how serious games offer an opportunity to develop more efficient tools to detect executive dysfunction in everyday life contexts. Serious games provide meaningful narrative stories and virtual or real environments that immerse the user in natural and social environments with social interactions. In those highly interactive game environments, the player needs to adapt his/her behavioral performance to novel and ill-structured tasks which are suited for collecting user interaction evidence. Serious games offer a novel opportunity to develop better tools to improve diagnosis of the executive dysfunction in everyday life contexts. However, more research is still needed to implement serious games in everyday clinical practice.

Serious Games and Mixed Reality Applications for Healthcare

Serious games are games in which the main goal is not entertainment, but a serious purpose ranging from the acquisition of knowledge to interactive training, to name just a few [...]