Canada

Anxiety disorder (AD) is a condition associated with feelings of dread and an incapability to perform daily tasks. Symptoms can be physical, such as a rapid pulse and sweating, and emotional, including constant stress and difficulty concentrating. Anxiety disorders are subjective, variable and often overlap with underlying conditions, complicating detection and treatment. This study focuses on three factors, implicit memory, reaction time, and reaction to a perceived threat—all measured in previous studies and correlated with anxiety levels. Our goal is to determine if anxiety presence can be identified, using three games and one simulation. The first game centers around implicit memory, enabling us to perform tasks subconsciously. Anxious individuals exhibit an implicit memory bias for threat-related information. The second game will associate a level of anxiety to reaction times. The third game first makes the user complete a regular task and progressively introduces a threat to the successful completion of the task. Attentional bias shows that anxious people are more likely to respond to threat-related stimuli rather than stick to their task. Finally, the user will take part in an artificial anxious situation. In the first stage, the user plays a regular game. In the second, distracting stimuli appear. In the third, the user’s available resources decrease. Throughout the stages, both the user’s effectiveness and efficiency will be evaluated. If the user experiences anxiety, efficiency should decrease in stage 2 and effectiveness should decrease in stage 3. Based on their results, they will receive an assessment of their optimal stress level for highest performance. The results from all of three games will be compiled to provide an anxiety assessment. The player will then be prompted to complete the Hamilton Anxiety Rating Scale (HAM-A) which will be used to verify whether the game results match their true anxiety levels.


Leveraging Open Data Analytics and Machine Learning to Improve Mental Health Research and Innovation

Using Interactive Games as Tools for Gauging Anxiety

technical paper

The University Big Data Challenge (IUBDC) successfully concluded its latest edition, bringing together national health industry leaders, policymakers, and dignitaries for a conference centered on mental health research and innovation. Participants had the opportunity to showcase their projects in a poster session, engaging with representatives from industry and academia. The event provided a rich network of industry and academic professionals, offering pathways for scholarly publication, access to curated datasets, learning materials materials, and workshops.

Held at the Microsoft Headquarters in Toronto on July 26, 2024, the challenge featured hands-on data science and AI research projects focused on mental health. For over seven years, IUBDC has fostered collaboration between data science and health professionals and motivated university students. This year’s projects, which spanned topics from dementia to addictions, aimed to enhance our understanding of mental health, improve patient lifestyles, and reduce disease risk. Abstracts from teams were published in the NRC Research Press STEM Fellowship Journal, with all team's works featured by Underline publications.

2024 Inter-University Big Data And AI Challenge -

Thee challenge featured hands-on data science and AI research projects focused on mental health. 

Autism Spectrum Disorder (ASD) is a neurological development condition that affects approximately 1 in every 54 children and youth in Canada. Individuals diagnosed with ASD may experience challenges in communication and behaviour, which may vary depending on the individual. Often, such challenges are coupled with comorbidities like mental health conditions, which are often more prominent in individuals on the spectrum. As such, the detection of early signs of mental health issues and the proper diagnosis of ASD is crucial. This project explores the overlap between ASD and mental health conditions by utilizing a mixed-methods analysis. By analyzing survey-based and omics data on ASD, the authors hope to conduct a comprehensive examination of Autism Spectrum Disorder from both a psychological and biological perspective. For the first part of this project, questionnaire-based datasets from Thabtah F. et al. on individuals sharing ASD traits were retrieved for analysis. The datasets consist of 10 questions related to ASD traits and demographic information across developmental stages. The concurrence of ASD and mental health conditions will be analyzed through machine learning, data analysis, and visualization. Subsequent stages will involve omics data to deepen our understanding of ASD.

Exploring the Intersection of Autism Spectrum Disorder and Mental Health Through a Mixed-Method Analysis

Substance treatment programs are essential for addressing substance addiction issues. Several considerations may explain why completion rates are lower than predicted. Previous studies have emphasized the impact of demographic and socioeconomic factors on treatment outcomes. However, there is insufficient research on utilizing advanced machine learning algorithms to predict adherence and discover trends in large datasets, particularly considering mental health concerns. The dataset was obtained from SAMHSA which contains extensive information on demographics, geography, education, employment, drug history, mental health status, and treatment outcomes. The study uses sophisticated machine learning algorithms to predict adherence to drug therapies, employing Random Forest and XGBoost models to identify significant characteristics influencing patient treatment completion. These models are quite accurate and could improve treatment outcomes through personalized methods. We investigated Random Forest and XGBoost models using cross-validation and several metrics such as accuracy, precision, recall, F1-score, and AUC-ROC. The Random Forest model achieved an accuracy of 85%, with a precision of 80%, recall of 82%, an F1-score of 81%, and an AUC-ROC score of 0.88. The XGBoost model performed even better, with an accuracy of 87%, precision of 82%, recall of 84%, an F1-score of 83%, and an AUC-ROC score of 0.90. Key predictors of treatment adherence included age, marital status, employment status, mental health status, and the number of previous treatment episodes. We also observed significant disparities in adherence rates among ethnic minorities and individuals with lower educational levels. The research shows that machine learning algorithms may accurately predict adherence to drug treatment programs. The identified predictors provide useful information for creating focused, individualized treatment methods. Future studies should use longitudinal data to evaluate post-treatment results and investigate other socioeconomic and psychological aspects. Healthcare practitioners may employ machine learning to enhance intervention efficacy, improve patient outcomes, and reduce the social impact of substance misuse.


Improving Prediction Adherence to Mental Health Treatment Programs Using Machine Learning Algorithms

The Internet provides a platform for individuals to share their thoughts and interact with others, enabling a dynamic exchange of information. In fact, one's online activities can reflect one's mental state, moreover, it may even influence other's mental state by their digital interactions. In this paper, we aim to leverage machine learning to monitor societal mental health by tracking users’ online activities. Using social media as a case study, we employ sentiment analysis to evaluate comments and assess users' online behaviors. Our analysis focuses on the frequent words used, the tone of the comments, and their impact on the audience. The NLP model is designed to monitor societal mental health trend by examining the correlation between online speech and mental health. This enables us to intervene in the content that negatively impacts social mental health, fostering constructive discussions and solutions. Our primary objective is to see a strong connection between hate speech and mental health, demonstrating its influence on the targeted individuals and the surrounding audience. However, our model has limitations. Social media users, particularly those on platforms like YouTube, represent only a small fraction of the overall internet population. Additionally, our model is designed to analyze English-language content, which excludes non-English-speaking internet users.

Social Media: The Influencer of Mental Health

We utilized the Behavioral Risk Factor Surveillance System (BRFSS) dataset for the following observational machine learning (ML) study. In contrast to recent literature using ML to predict specific outcomes, we used causal machine learning (CML) algorithms (LRSRegressors, DRLearner, Rlearner, Xregressor, and Tregressor) to quantify the cause-and-effect relationships between adverse childhood experiences (ACEs) and two different mental health targets (number of bad mental health days and whether the respondent had a depressive disorder). Specifically, we report each ACE’s average treatment effect (ATE) on each target. We also used Uplift Random Forest trees to calculate uplift scores. We found that growing up in a household where one parent had a depressive disorder and being forced into unwanted sex had the most significant effects on our chosen mental health targets. Growing up in a household where one parent had a depressive disorder increased the likelihood of being diagnosed with a depressive disorder by 30% and added 5.07 days of bad mental health in the last month. Repeatedly being forced to have sex added 5.77 days of poor mental health and increased the likelihood of being diagnosed with a depressive disorder by 33%. Uplift modeling indicated that individuals over 35 years of age earning under $100,000 USD annually were most susceptible to the effects of a parent diagnosed with depression by a factor of 1.1 compared to the average population. Male college graduates over 44 years of age earning under $100,000 USD were found to be most susceptible to the effects of childhood sexual abuse by a factor of 1.38. Future research should refine these models and employ more complicated algorithms to gain accurate, interpretable measurements and to understand the relationship between ACEs and mental health.

Quantifying Childhood Trauma: Causal Machine Learning Approaches to Mental Health Outcomes

Impostor syndrome is defined as a combination of self-perceived fraudulence, achievement pressure, and negative emotions [3], which creates doubts in one's capacity, leading to unfavorable comparisons with others. Our study represents an approach to evaluate individuals based on a set of questions from the Young Impostor Syndrome (YIS) scale [4] built within a web application. These questions assist users in identifying impostor syndrome symptoms, such as self-doubt, and use Likert-scale responses ranging from ‘agree’ to ‘disagree’ to calculate a more accurate impostor syndrome metric. Individuals are encouraged to provide personal information such as age and years of experience in the field, enabling users to attribute their accomplishments to their own ability. Furthermore, our initial data, from a dataset predicting impostor syndrome in medical students [2], is processed as the basis for our analysis. We then expand this data with online user responses, using a cloud solution to store their inputs. The data is then presented as clusters to the user so they can observe how many other individuals are dealing with similar kinds of issues. This enables the investigation of relationships between impostor syndrome traits and demographics. We use clustering algorithms [1] to place users into groups according to their profiles. Users can see where those clusters fit in relation to others by viewing them in 2D and 3D scatter plots. Visualizations help to better understand the frequency and the variability of this mental health phenomenon among demographic groups, in addition to quantifying it. This study adds to our understanding of the intricacy and effects of impostor syndrome. Users’ experience can be contextualized by the visual representation, which promotes awareness and debate of the psychological phenomenon. Our method emphasizes how sophisticated data analytics and psychological testing may be used to create a comprehensive tool that helps research and individuals. Keywords: impostor syndrome, data analysis, clustering algorithms, visualization, web application, mental health, Likert scale

Imposter Syndrome Comparison Map

With the advent of machine learning becoming more and more integrated into healthcare, this study explores the efficacy of machine learning models in classifying and predicting various brain disorders and diseases such as schizophrenia, Alzheimer’s disease and Parkinson’s disease. Our review highlights the potential that ML has in identifying effective biomarkers, with cortical thickness emerging as a more accurate indicator for schizophrenia. Additionally, the connection between Alzheimer's disease and metabolic syndrome was made much more clear through ML algorithms, identifying eight diagnostic genes with high AUC values. The versatility of ML is further demonstrated through the application of natural language processing to analyze social media data, uncovering mental health issues in patients with diabetes. The challenges associated with ML use in a healthcare setting were also analyzed; the performance of ML models can drop substantially when encountering new data significantly different from the training set. While ML has great promise, it is recommended to use as a supplementary tool in collaboration with clinicians for the greatest efficacy. Continuous improvement in algorithms, data quality, and model validation will be essential to fully realize the potential of ML in improving patient care and outcomes.

Machine Learning in Neurological Disorders: A Comprehensive Review of Diagnostic and Predictive Applications

Stress affects over 79% of individuals globally, presenting a significant challenge for timely and accurate identification within mental health care. Traditional methods, which rely on self-reported data and sporadic clinical assessments, often result in delayed diagnosis and inadequate stress management. To address these challenges, this project employs artificial intelligence (AI) to develop a real-time stress level detection model using unsupervised learning algorithms. By integrating data from wearable devices, such as Apple Watches, with self-reported mood and activity logs, our model provides continuous and comprehensive stress analysis. Achieving an accuracy of 60% upon validation, the model demonstrates a significant step toward more effective stress management. This AI-driven approach shows promise in enabling early stress detection and personalized interventions. The findings underscore the potential for AI to enhance stress monitoring accuracy, thereby contributing to improved individual well-being and broader accessibility to quality mental health care. Future research will focus on refining the model to improve its accuracy and exploring additional data sources to further enhance its predictive capabilities.

Enhancing Wellbeing: An AI-Driven Stress Level Identifier

Music therapy is a recognized field of therapeutic intervention used in a diverse variety of contexts, including hospital settings, retirement homes, palliative care, and community programs. It leverages intrinsic qualities of music to address mental health issues and enhance overall mental health. A challenge for music therapists working in hospice is lack of suitable musical repertoires, called “music unpreparedness”. This study aims to use a machine learning model to evaluate repertoires and predict their effects on an individual’s mental health given their age and mental health condition, if any. Streaming services such as Spotify can also implement this model to improve the overall effects of services on listeners’ mental health–for instance, promoting playlists that will likely improve, rather than worsen, the user’s mental health on their home page. To train the model, an open-access dataset was used containing features such as favorite genre, age, and mental health condition. Categorical variables including ‘Fav genre’ and ‘Frequency [genre]’ were prepared using one-hot encoding and the dataset was split into training and testing sets, using a stratified approach to maintain the distribution of mental health conditions and music preferences. A decision tree model was chosen due to its interpretability and ability to handle categorical data. The model was trained with the entropy criterion and a maximum depth of 6, achieving a training accuracy of 82% and a validation accuracy of 75%. Key features influencing the model included listening hours/frequency, favorite genres, and mental health condition levels. The model provides insights into how these factors contribute to the perceived effect of music on mental health, offering valuable predictions that can enhance therapeutic interventions and user experiences on streaming platforms. Future improvements include incorporating more diverse datasets to improve the model's generalizability. Additionally, other machine learning algorithms can be explored to enhance prediction accuracy.

Predictive Models on the Therapeutic Effects of Diverse Music Repertoires on Mental Health

Mental Health is a subjective phenomenon for which there is only one way to diagnose, and that is knowing the root cause of this health problem. In many cases, people hesitate to consult for treatment due to discrimination, stigma, or fear of the unknown. This study shows how a Natural Language Processing (NLP) based Chatbot can be an interactive medium for patients to seek guidance for their mental health-related issues without hesitation and in a healthy and safe environment that will protect their privacy. NLP-based chatbots are governed by an artificial neural network (ANN) that is deployed as an app. The model is trained on a dataset with various questions, patterns, and responses. Based on the trained data, it provides 86.04% accurate responses after the training and could increase if its knowledge base is increased. This Chatbot can address issues like anxiety, depression, phobia, post-traumatic stress disorder (PTSD), and Schizophrenia. However, there is a need to expand the volume and variety of data and include a learning curve in the model to improve the responses. Long Short-Term Memory (LSTM) or GPT-3 can make the NLP-based Chatbot play a crucial role in elevating efficiency as it can foster personalized support with tailored needs of individuals.

AI For Emotional Wellbeing: Designing an NLP-based Mental Health Chatbot

Stress is a common experience reported to impact 30% of the global population with significant negative mental health effects and has sparked an increase in research aimed at stress detection. This study explores the effectiveness of using electrocardiogram (ECG) and electroencephalography (EEG) signals as predictors for stress detection using a multi-perceptron neural network. The model is trained and validated using a dataset from a study with 40 participants, 21 female and 19 male, at the Prince of Songkla University in Thailand. The participants were between the ages of 18 and 25 and performed mental arithmetic tasks (MAT) while their EEG and ECG signals were measured. The data is divided into 3 classes: EO, A1, A2 representing normal stress, high stress, and low stress levels in participants respectively. In addition to the participant’s gender included in both the EEG and ECG data, the EEG data contains the Fp1, Fp2, F3, F4, T3, T4, P3, P4 while the ECG data contains mean heart rate, AVNN, SDNN, NN50, pNN50, RMSSD, LF, LF Norm, HF, HF Norm, and L/F ratio. The model’s architecture including its learning rate, activation function, dropout rate, epoch, and layers were adjusted and evaluated to currently achieve 62% accuracy. In comparison to existing studies using the same dataset with various machine learning algorithms such as Adaptive Boosting, RF, SVM, k-Nearest Neighbor, LR, and NB, this model achieves similar accuracy, indicating competitive performance of the multi-perceptron with existing stress detection models. Future iterations of the model aim to improve the test accuracy and decrease the loss of the model by creating a training set using the nearest centroid neighbor (NCN) approach. The scarcity of data on EEG, ECG, and stress while training the model suggest a need for further data collection to further improve the model’s training set and accuracy.

Using Interactive Games as Tools for Gauging Anxiety

Downloads

Next from Leveraging Open Data Analytics and Machine Learning to Improve Mental Health Research and Innovation

Exploring the Intersection of Autism Spectrum Disorder and Mental Health Through a Mixed-Method Analysis

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES