Khám phá ứng dụng kỹ thuật AI trong phân tích dữ liệu để chẩn đoán bệnh tâm thần phân tích hệ thống Exploring the application of ai techniques in data analysis to diagnose mental illnesses a systematic analysis
LITERATURE REVIEW
AI technology and applications of AI in medicine
In 1956, the Dartmouth Society officially introduced AI, a crucial area of computer science, now regarded as one of the top three technologies in the world [1]
AI is often described as having the ability to replicate human brain functions, enabling machines to reason, solve problems, recognize speech and objects, infer environmental conditions, and make decisions These capabilities encompass listening, seeing, speaking, thinking, and decision-making, mirroring essential human abilities.
AI is categorized into two types based on intelligence level: weak AI and strong AI Weak AI focuses on specific tasks, such as autonomous vehicles, Apple's Siri, and Amazon's Alexa, utilizing language and image recognition as well as intelligent translation It is designed to assist humans rather than replicate human intelligence, relying on medical and logical algorithms In contrast, strong AI denotes machines with human-like intelligence and capabilities, while Artificial Super Intelligence refers to systems that exceed human abilities Currently, strong AI remains a theoretical concept.
AI has several subdomains, including ML, DL, and Neural Networks (NN)
Machine Learning (ML), a subset of weak AI, encompasses various methods that enable algorithms to learn and adapt It drives numerous contemporary applications, such as email spam filters, search engine suggestions, online shopping recommendations, and speech recognition technology in smartphones.
Deep Learning (DL), a branch of Machine Learning (ML), simulates human brain functionality using Artificial Neural Networks (ANN) with multiple layers to automate predictions from training datasets An ANN is considered "deep" when it contains several hidden layers.
AI is revolutionizing early disease detection and enhancing our understanding of disease progression Despite limitations in capacity and access to knowledge, AI-powered systems can rapidly analyze and synthesize vast amounts of medical information from numerous sources.
Machine learning (ML) is a branch of artificial intelligence that develops statistical models and algorithms, enabling computers to perform tasks independently without human intervention These systems analyze data to identify patterns, which they then use to make informed decisions Arthur Samuel, a pioneer in the field, described ML as a discipline that empowers computers to learn autonomously without explicit programming.
Machine Learning (ML) has diverse applications across various industries In finance, ML algorithms enhance predictive analytics for stock market forecasting and credit scoring, aiding in risk management and investment decisions In healthcare, ML analyzes extensive medical datasets to predict disease outbreaks and assist in patient diagnoses Additionally, Natural Language Processing (NLP), a subset of ML, drives innovations such as chatbots, language translation services, and sentiment analysis tools, enabling machines to effectively understand and respond to human language For example, chatbots enhance customer service by delivering instant responses, while sentiment analysis tools track social media for customer feedback and emerging market trends.
Recommendation systems are a vital application of machine learning, utilized by major companies such as Netflix, Amazon, and Spotify to provide tailored suggestions for movies, products, and music according to individual user preferences By analyzing user behavior and preferences, these systems deliver personalized recommendations that significantly enhance user experience and engagement.
Deep Learning (DL) is a specialized branch of Machine Learning (ML) that employs deep neural networks (NNs) with multiple layers, designed to replicate the structure and function of the human brain This approach enables DL models to learn and make decisions from vast amounts of data, making them particularly effective for complex tasks like image and speech recognition.
Deep learning (DL) has transformed image and speech recognition, significantly enhancing technologies such as facial recognition systems used in security and personal device authentication, which depend on DL algorithms to accurately identify individuals by their facial features Additionally, voice assistants like Siri and Alexa utilize DL for natural language understanding and voice command processing, allowing for seamless user interaction and efficient task execution.
Deep learning (DL) plays a crucial role in the development of autonomous vehicles, allowing self-driving cars to analyze large volumes of data from sensors and cameras for navigation and real-time decision-making Leading companies such as Tesla and Waymo utilize DL algorithms to interpret visual data, identify objects, and anticipate their movements, ensuring safe and efficient driving.
Deep learning (DL) has significantly advanced medical diagnostics by enabling models to analyze medical images, including X-rays, MRIs, and CT scans, with high precision Research indicates that DL algorithms can equal or even outperform human specialists in diagnosing diseases such as cancer, facilitating more timely and accurate treatment options.
1.1.2 Overview of AI technology in healthcare
AI is revolutionizing healthcare by enhancing diagnostics, personalizing treatments, and improving operational efficiency Utilizing machine learning (ML) and natural language processing (NLP), AI analyzes vast amounts of medical data to enable early disease detection and accurate diagnoses, particularly in radiology through the identification of anomalies in X-rays and MRIs Furthermore, AI customizes treatments by examining genetic information, health records, and lifestyle factors, which not only boosts treatment efficacy but also minimizes adverse effects This innovative approach allows for the prediction of individual responses to medications, paving the way for personalized medicine.
AI enhances clinical practice by automating routine tasks, optimizing workflows, and aiding in decision-making It facilitates scheduling, billing, and patient triage based on urgency, which improves patient flow and minimizes waiting times in emergency departments Additionally, AI is vital in public health, as it analyzes data from various sources to predict and manage disease outbreaks, enabling timely interventions and effective responses.
The adoption of AI in healthcare can lead to substantial cost reductions by streamlining billing processes, improving resource allocation, and boosting operational efficiency This optimization not only ensures that healthcare facilities are well-staffed and properly equipped but also enhances patient care, ultimately resulting in significant cost savings.
[19] Additionally, AI's ability to handle large datasets and provide real-time insights supports the ongoing transformation and efficiency of healthcare systems [17].
Mental illness and mental illness data analysis
1.2.1 Overview of Mental Illness and Mental Illness data analysis
Mental illnesses encompass a diverse range of conditions that significantly impact mood, thought processes, and behavior, often hindering daily functioning and personal relationships Common types include depression, anxiety disorders, bipolar disorder, schizophrenia, and eating disorders These illnesses typically stem from a complex mix of genetic, biological, environmental, and psychological influences The severity of mental health conditions can vary widely, with some individuals experiencing mild symptoms, while others endure severe, chronic issues that necessitate ongoing treatment and management.
Data analysis is essential in mental health research for understanding the prevalence, risk factors, and outcomes of mental illnesses Systematic reviews and meta-analyses play a key role in synthesizing findings from various studies, such as demonstrating the effectiveness of physical activity interventions in alleviating symptoms of depression, anxiety, and psychological distress Furthermore, interventions that target stigma reduction among healthcare professionals have proven successful in decreasing self-stigma and promoting help-seeking behaviors Community mental healthcare for individuals with severe mental illness focuses on integrating medical and social services to enhance overall well-being and recovery Additionally, combining social justice initiatives with mental health services can significantly improve outcomes, making data-driven insights crucial for shaping public health strategies and informing effective policy-making to tackle mental health challenges.
1.2.2 Technologies and tools used for analysis
Big data analysis in healthcare leverages advanced tools to manage and interpret extensive datasets Apache Hadoop, utilizing HDFS and MapReduce, facilitates efficient parallel processing, while Apache Spark enhances big data processing speed through in-memory computing.
MongoDB handles large volumes of unstructured data flexibly and scalably
[27] Python and R are popular for data analysis and visualization, with libraries like Pandas, NumPy, ggplot2, and caret [28]
Tableau creates understandable visual reports from raw data [29] TensorFlow and PyTorch are essential for building and training NNs [15] SAS supports advanced analytics and data-driven decision-making in healthcare [30]
These tools transform raw data into actionable insights, enhancing patient outcomes and public health strategies
1.2.3 Existing limitations and difficulties in Mental illnesses data analysis
Analyzing data on mental illnesses poses significant challenges due to the heterogeneity of mental health conditions, complicating diagnosis and classification The wide range of symptoms and severity levels hampers the establishment of standardized diagnostic criteria and treatment protocols Furthermore, mental health data frequently depend on self-reported measures, which can introduce bias and inaccuracies because of the subjective nature of the information.
Significant data gaps in mental health research arise from underreporting and stigma, which hinder comprehensive analysis Inconsistent data collection methods across studies complicate comparisons of findings Additionally, privacy concerns regarding sensitive mental health information necessitate strict confidentiality measures, limiting detailed data availability for research Furthermore, the absence of longitudinal studies restricts our understanding of the progression and outcomes of mental health conditions over time.
AI technology in the study of mental illnesses
AI technology is revolutionizing the study and diagnosis of mental illnesses, providing innovative solutions for understanding and treating these conditions Machine learning (ML) and deep learning (DL) techniques are significantly improving diagnostic accuracy and efficiency by analyzing large, complex datasets that traditional methods often find challenging to handle.
Machine learning algorithms are increasingly being used to forecast the onset and progression of mental disorders by examining various data sources, including electronic health records and social media activity These advanced models can detect patterns and risk factors linked to conditions such as depression and anxiety, enabling timely intervention strategies.
Deep learning models have shown exceptional proficiency in analyzing neuroimaging data for diagnosing mental health disorders These models have been effectively utilized in MRI and fMRI scans to accurately identify conditions such as Alzheimer's disease and schizophrenia Such technological advancements enable the detection of subtle alterations in brain structure and function that signify various mental health issues.
Natural Language Processing (NLP) is a vital AI technique that enhances mental health research by analyzing speech and text to uncover linguistic patterns associated with mental health issues AI-driven chatbots and virtual therapists leverage NLP to engage with patients, offering support and recognizing indicators of mental distress.
Research methodology systematically overview, significance in medicine and
1.4.1 Systematic review analysis in medicine
A systematic review in medicine is a study designed to address a specific research question by synthesizing all relevant empirical data that meets established criteria The process involves a thorough and organized literature search, followed by a critical evaluation and integration of data from selected studies Systematic reviews are valued for their ability to minimize bias through structured methodologies, making them high-quality sources of evidence, especially when used to guide clinical guidelines and healthcare decisions.
1.4.2 Systematic review analysis in the field of mental health
Systematic reviews in mental health synthesize extensive research to provide a comprehensive understanding of various issues and identify research gaps Utilizing rigorous methodologies, these reviews gather and summarize findings from multiple studies, which are essential for identifying trends and evaluating the effectiveness of interventions They often reveal significant associations between mental health outcomes and various factors; however, inconsistencies in study designs and measurement tools highlight the need for standardized methodologies to achieve more reliable conclusions.
1.4.3 Reasons for conducting a systematic review study
The advancement of science and technology has led to numerous studies on similar topics by various research groups globally, enhancing the accuracy and reproducibility of research results However, this abundance of studies can complicate the synthesis of information and hinder the identification of unexplored research directions Therefore, conducting systematic reviews is crucial, especially before initiating new research projects Systematic reviews offer a comprehensive overview of current research trends, helping scientists pinpoint gaps and make informed decisions about future research directions, ultimately saving time and resources while improving research quality In our systematic review, we aimed to consolidate published findings on the application of AI techniques in analyzing data for diagnosing mental illnesses, providing recommendations for suitable AI tools and methods to guide clinical trials and facilitate practical applications Additionally, we identified gaps in data usage and AI application results, suggesting potential avenues for future research in this area.
1.4.4 Steps by steps to perform system overview
A systematic review can be carried out in seven phases, according to Petticrew and Olgilvie (2005) [38]:
1 Clearly state the hypothesis or research question
2 Determine the kinds of research required to carry out the investigation;
3 Carry out an exhaustive literature search required to find relevant studies
4 Screen studies to see whether they fit the selection criteria; if not, do additional analysis
5 Reexamine papers critically before incorporating them into the systematic review
6 Compile research and evaluate its consistency
RESEARCH METHOD
Research content
This study was conducted with an aim to explore and analyze the application of AI in data analysis to diagnose mental illnesses
A systematic review was conducted to assess the evidence and research gaps regarding the application of AI in diagnosing mental health issues Data was collected from various databases, filtered, and analyzed to identify the AI models and methods used, as well as their effectiveness in biomedicine and mental health The review also explored AI's role in enhancing treatment and prognosis for mental health conditions This analysis highlighted significant research gaps in the use of AI for data analysis in mental health diagnosis.
Research Method
The research focuses on reputable articles and research papers that explore the use of AI approaches in mental health diagnosis, accessible through various databases A systematic process of searching, collecting, and selecting these articles is integral to the research methodology.
The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were adhered to in this systematic review, making them the leading framework for conducting such studies These principles enable researchers to maintain a high standard of objectivity and accuracy throughout the systematic review process, encompassing research question selection, data collection, evaluation, and the synthesis and analysis of results.
To decide which studies to include in the analysis, the following criteria were applied (PICOTS) [39]:
This article focuses on the population of scientific research concerning the application of artificial intelligence in mental health and illness data analysis It specifically examines English-language articles available as free full texts from the PubMed and ScienceDirect databases.
- I - Intervention: With or without intervention
- C - Comparison: Whether or not there was a comparison group in the study
- O - Outcome: Application of AI techniques in data analysis to diagnose mental illness
- T – Time: All articles published to date
- S - Study Design: There is no limit to research design, except for Review, Systematic review, Protocol, and research using the same data
- Analytical studies do not refer to the application of AI techniques in data analysis to diagnose mental illnesses
- Analytical studies do not refer to diagnose methods but to other methods
- Review studies, systematic reviews, meta-analyses, etc
2.2.4 Data sources and techniques for searching
- The data source is taken from two research databases: PubMed and ScienceDirect
- Searching based on keywords “AI” AND (“Mental Health” OR “Mental Illness”) AND “Data Analysis” AND (“Diagnosis” OR “Diagnose”) contained in tilte and abstract of articles related to the topic
- Using Zotero software Version 6.0.37 of the Corporation for Digital Scholarship
Organization to store downloaded articles and research newspapers related to the topic
- Using Microsoft Excel to manage and analyze articles and research newspapers
2.2.5 Research screening and selection process
The procedure for filtering and selecting studies for review is outlined below:
Step 1: Combine search data, eliminate duplicate articles, and exclude those without abstracts
Step 2: Screen based on titles and abstracts to remove reports unrelated to the application of AI techniques in diagnosing mental health conditions
Step 3: Retrieve full texts of the remaining articles
Step 4: Screen the full texts Evaluate them according to the inclusion and exclusion criteria
Step 5: Contact authors when necessary to clarify research objectives or request additional information
Step 6: Make the final decision on which studies to include and begin data collection
All selected texts will be meticulously evaluated by two independent reviewers, Hoang Thai Van Bui and Thanh Nhat Hoang In case of any disagreements during the selection process, discussions will be held, or a third reviewer, Ph.D Dinh Toi Chu, will be consulted The final report will comprehensively detail the search results and include a PRISMA diagram for clarity.
RESULTS AND DISCUSSION
Search results and Characteristics of studies
The systematic review process starts with a thorough search across various databases, resulting in 2,546 records, including 1,546 from PubMed and 1,000 from Science Direct After eliminating 80 duplicate records, 2,466 unique articles remain These records are then screened by title and abstract, leading to the exclusion of 2,424 articles that either lack relevance to the study topic or are not available in free full text Subsequently, a detailed eligibility assessment is performed on the full-text articles.
Out of 42 full-text articles evaluated, only those that met the PICOTS criteria were included in the review Ultimately, 9 articles were selected for their direct relevance and high-quality evidence regarding the role of AI in diagnosing mental health issues, forming the core findings of the review In total, 33 articles were incorporated into the systematic analysis.
Table 3.1 provides a comprehensive overview of the features of 33 selected research papers, detailing each article's citation ID, publication year, country of origin, type of illness investigated, data utilized, main research objective, primary experimental design, and sample size.
Table 3.1 Characteristics of articles selected for the study
ID Year Disease Name Objective Country Sample
[40] 2023 Mental Disorder develop DL models to predict mental disorder diagnosis and severity spanning multiple diagnoses Denmark 63535 Genetic Data
[41] 2019 Bipolar Disorder develop a new Bipolar Diagnosis Checklist in Chinese (BDCC) by using ML to shorten the Affective Disorder Evaluation scale (ADE)
Disorder reducing the number of questions while preserving the classification given by the full ADI-R USA 1976 Genetic Data
In 2022, a study focused on children with Attention-Deficit Hyperactivity Disorder (ADHD) aimed to summarize their brain electrical activity and clinical characteristics Utilizing long-range video graphics data, the research explored the clinical significance of long-range EEG in diagnosing ADHD in children, highlighting its potential to enhance diagnostic accuracy and understanding of the condition.
EEG Data, demographic data, clinical data
[44] 2021 Anxiety Disorders find the variable importance hierarchy of biomarkers for anxiety disorder and a particular type of anxiety disorder, not the prediction of one particular anxiety disorder
The Netherlan ds 11081 Biomarker data
Disorder evaluating how the mentioned hybrid methodology proposal can make the protocol derived from ICD-10 more efficient, providing agility to diagnosing Autism Spectrum Disorder by observing a minor symptom
To develop ML models to explore factors contributing to misdiagnosis and seeking help behaviors in individuals experiencing low mood
Schizophrenia to develop an integrated multimodal data ML model to predict the diagnosis of bipolar disorder and schizophrenia France 416 Biomarker data
Attention deficit/hyperactivity disorder to develop interpretable ML models for aiding in the diagnosis of
ID Year Disease Name Objective Country Sample
Depression to evaluate the effectiveness of AI models in the diagnosis of mental health issues
In 2021, research focused on utilizing machine learning (ML) techniques, particularly deep learning (DL) strategies, within computer-aided diagnosis (CAD) models to effectively diagnose and classify patients with stable average scores, Alzheimer’s disease (AD), and mild cognitive impairment (MCI) This study analyzed imaging data from 675 participants in Iran.
Disorders use ML algorithms to shorten observation-based screening and diagnosis of autism, specifically focusing on the Autism Diagnostic Observation Schedule-Generic (ADOS) Module 1 data
[52] 2024 Alzheimer's Disease a comparative study of Graph NN (GNN) and Multi-Layer
Perceptron (MLP) based ML for the diagnosis of Alzheimer’s Disease involving data synthesis
[53] 2022 Alzheimer's Disease to propose a new method for early detection of Alzheimer's disease by combining Surface-Enhanced Raman Scattering (SERS) and
Convolutional NN (CNN) USA N Imaging Data and
[54] 2021 Schizophrenia diagnosis of schizophrenia based on DL using functional magnetic resonance imaging (fMRI) USA 147 Imaging Data
[55] 2023 Alzheimer's Disease early detection of associated cognitive impairment improve the automatic detection of dementia in MRI brain data USA 416 Imaging Data
Using strategies that obtain biomarkers where early symptoms coincide, the early detection of Alzheimer's disease and its complications is essential Iran N Imaging Data and
[57] 2014 Semantic Dementia classify transcribed speech samples according to clinical aspects, using only lexical data UK 32 Imaging Data
[58] 2024 Alzheimer's Disease propose an ensemble method based on DL for the early diagnosis of AD using MRI images Iran 747 Imaging Data
Early detection of AD and its prodromal stage, mild cognitive impairment (MCI) which can be either stable (sMCI) or progressive (pMCI
ID Year Disease Name Objective Country Sample
[60] 2015 Alzheimer's Disease diagnosis and prognostication of Alzheimer's disease Canada 822 Imaging Data
[61] 2017 Alzheimer's Disease Diagnosis of Alzheimer's Disease Based on Structural MRI Images South
[62] 2011 Alzheimer's Disease diagnosis of Alzheimer disease using data base independent estimation and fluorodeoxyglucose positron-emission tomography and 3D-stereotactic surface projection
[63] 2019 Alzheimer's Disease predict a diagnosis of Alzheimer's disease UK 1737 Imaging Data
[64] 2022 Alzheimer's Disease diagnosis of Alzheimer's disease China 1340 Imaging Data
[65] 2016 Alzheimer's Disease propose a novel deep sparse multi-task learning method for feature selection in Alzheimer's disease (AD) and mild cognitive impairment (MCI) diagnosis using neuroimaging data
In 2014, researchers developed and validated an automated diagnostic tool for Alzheimer's Disease, which enhances early diagnosis accuracy by integrating Positron Emission Tomography (PET) images with neuropsychological test data.
[67] 2016 Alzheimer's Disease to propose an inherent structure-based multi-view learning (ISML) method using multiple templates for the classification of
Alzheimer's Disease (AD) and its prodromal stage, mild cognitive impairment (MCI)
This article proposes an automated method for diagnosing Alzheimer’s Disease by utilizing dual-tree complex wavelet transform (DTCWT) for feature extraction, principal component analysis (PCA) for dimensionality reduction, and a feed-forward neural network (FNN) for the classification of MR images.
[69] 2022 Fibromyalgia to identify objective biomarkers of fibromyalgia by applying AI algorithms to structural data on the neuroretina obtained using swept-source optical coherence tomography (SS-OCT) Spain 61 Biomarker data
ID Year Disease Name Objective Country Sample
[70] 2022 Schizophrenia to develop a robust and interpretable DL model for the diagnosis of schizophrenia using psychiatric observation data based on DSM-5 criteria
In 2024, the RLKT-MDD (Representation Learning and Knowledge Transfer for Multimodal Depression Diagnosis) model framework will be proposed and validated to improve the accuracy of depression diagnosis This innovative approach combines representation learning and knowledge transfer techniques, aiming to enhance diagnostic precision in multimodal contexts.
In 2023, AlzDiagnostics was introduced as a mobile application aimed at facilitating the early diagnosis of Alzheimer's Disease This user-friendly tool combines established clinical methodologies with advanced machine learning techniques and various diagnostic approaches to enhance detection and diagnosis.
There are 6 articles (accounted for 19% of total research papers) published in
Between 2021 and 2024, there has been a notable increase in research focused on utilizing AI techniques for diagnosing mental illnesses, with five studies conducted each year, each contributing 15% to the overall research output This surge can be linked to advancements in AI technologies and a heightened awareness of mental health issues spurred by the COVID-19 pandemic In contrast, the years 2019 and 2023 saw only three studies each, accounting for 9% of the total research Earlier years, specifically from 2011 onwards, reflect a gradual evolution in this field.
2017), there were minimal research efforts (from 3% - 6%)
Figure 3.2 Publication Year of articles
Alzheimer's disease dominates research efforts, comprising 47.2% of total studies, largely due to its complexity and the availability of extensive datasets like the Alzheimer's Disease Neuroimaging Initiative (ADNI), which facilitate data collection for researchers In contrast, artificial intelligence applications in diagnosing bipolar disorder, autism spectrum disorder, depression, and schizophrenia each represent 8.3% of the research landscape While there is some investigation into AI-based diagnosis for other mental illnesses, these studies are minimal, ranging from 2.8% to 5.6%.
Figure 3.3 Types of Mental Illness
The United States leads in research on AI techniques for diagnosing mental illness, contributing 7 papers, which represents 21.2% of the total China follows closely with 6 publications, accounting for 18.2% This dominance is largely due to significant funding from government bodies like the National Institutes of Health (NIH) and private sector investments, alongside advanced research infrastructures in both countries South Korea and the UK are tied for third place, each with 4 articles (12.1%), while Iran contributes notably with 3 publications (9.1%) Other countries, including Romania, Spain, Belgium, Japan, France, Denmark, the Netherlands, and Brazil, collectively account for 3% of the total research output in this field.
Figure 3.4 Articles published by countries
Imaging data plays a crucial role in AI model diagnostics, accounting for 48.8% of the data utilized, as it offers in-depth insights into brain structure and function, revealing abnormalities and connectivity patterns linked to mental illnesses Clinical data follows with a contribution of 14%, while genetic data accounts for 11.6% Additionally, biomarker data, categorical data, and questionnaire responses each contribute 7% Other data types, including multimodal and EEG, contribute the least at 2.3%.
Figure 3.6 illustrates notable discrepancies among the studies analyzed, with one research paper featuring an exceptionally large sample size of 63,535 participants, attributed to the use of country data for analysis In contrast, most other studies have sample sizes ranging from a few hundred to a few thousand, with many falling below 500 participants This variation underscores the coexistence of large-scale and focused research within the field, highlighting how differing sample sizes affect the generalizability of findings and reinforcing the comprehensive nature of the review.
Figure 3.6 Sample size of selected articles
Application of AI in data analysis to diagnose mental illness
This section explores the role of AI in diagnosing mental illnesses through data analysis, providing an overview of techniques tailored to specific disorders It categorizes AI methodologies into three main approaches: Neural Networks (NN) and Deep Learning (DL), Traditional Machine Learning models, and Ensemble Methods, evaluating their effectiveness in enhancing diagnostic accuracy and efficiency The discussion includes the types of data utilized and the performance of these AI models in delivering early and precise diagnoses Furthermore, it addresses the strengths and limitations of AI applications, offering insights into their practical implications and the potential for future advancements in mental health diagnostics.
3.2.1 AI Techniques apply in specific Mental Health Disease
Table 3.2 AI Techniques apply in specific Mental Health Disease
Disorder Category AI Techniques Data Types
- deep weighted subclass-based sparse multi-task learning (DW-S2MTL) [65, 66]
Imaging Data Genetic Data Clinical data Biomarker Data
- the least absolute shrinkage and selection operator (LASSO) [41]
Clinical data Categorical data Questionnaire
- Random Forest (RF) [45] Genetic data
Imaging Data Genetic Data EEG Data Categorical data Clinical data
- generalized linear/logistic regression model (GLM) [44]
- Light Gradient Boosting Model (LGBM [49]
Depression - Gradients boosting model (GBM) [46]
- Light Gradient-Boosting Machine (LGBM) [49]
Multimodal Data schizophrenia - VGG16 [54] Imaging Data
Categorical semantic dementia - naive Bayes (NB) [57] Imaging data
Table 3.2 illustrates the extensive use of AI techniques in diagnosing mental health disorders, with Alzheimer's Disease employing the widest range of methods, including CNNs, GNNs, decision trees, and regression models, utilizing diverse data types such as imaging, genetic, clinical, and biomarker information Depression diagnosis incorporates gradient-boosting methods, deep neural networks, and classifiers, leveraging imaging, questionnaires, and multimodal data Bipolar Disorder utilizes Random Forest and SVR techniques with clinical, categorical, and questionnaire data, while Autism Spectrum Disorders rely on ADTree and Random Forests with genetic and questionnaire inputs ADHD research employs CNNs, decision trees, SVMs, and logistic regression, focusing on imaging, genetic, and EEG data Anxiety Disorders are assessed using generalized linear models and CNNs, emphasizing biomarkers and imaging data Schizophrenia diagnoses utilize the VGG16 CNN model with imaging and categorical data, whereas semantic dementia is analyzed using Naive Bayes, and Fibromyalgia employs LASSO, SVM, and MLP with biomarker and clinical data This comprehensive application of AI enhances diagnostic accuracy and treatment strategies across various mental health conditions.
3.2.2 Analysis of AI application to diagnose Mental Illness based on NN and DL approaches
The objectives of this section are to evaluate various AI approaches based on
Neural networks (NN) and deep learning (DL) are powerful techniques that excel in managing complex patterns and large datasets These architectures are particularly effective in applications such as image recognition and natural language processing, where deep feature learning is essential However, they also come with certain limitations that should be considered when implementing these models.
3.2.2.1 Evaluate AI application to diagnose Mental Illness based on NN and DL approaches
Table 3.3 Analysis of the AI application using NN and DL approaches
Ref Approach Targets for Application Evaluation criteria
AUC ACC SEN PRE SPE
CNN Improve accuracy in diagnosis, classify and early detection x 82.3%-
DW-S2MTL feature selection and classification, improve the accuracy of diagnostic classification x 47.83%-
Explaination: x: no information, AUC: Area Under the Curve , ACC: Accuracy Rate , SEN:Sensitivity , PRE: Precision , SPE: Specitivity
Table 3.3 provides a detailed comparison of various NN and DL approaches applied to diagnosing mental illnesses
Convolutional Neural Networks (CNNs) are increasingly utilized for enhancing diagnostic accuracy, classification, and early detection of mental illnesses, demonstrating impressive performance with accuracy rates between 82.3% and 99.83% Their sensitivity ranges from 75% to 98.1%, while precision scores vary from 92.4% to 100% This robustness in performance solidifies their status as a preferred tool for diagnosing mental health conditions.
Graph Neural Networks (GNNs) and Multi-Layer Perceptrons (MLPs) are utilized for diagnostic applications, achieving impressive accuracies of 90.18% and 88.39%, respectively While both models demonstrate high accuracy, they may lack the versatility and performance of Convolutional Neural Networks (CNNs) across a wider array of diagnostic tasks.
The DW-S2MTL method demonstrates significant potential in feature selection and diagnostic classification, achieving accuracy rates between 47.83% and 90.27% However, its variability indicates a need for further refinement to ensure consistent results In classification and early detection tasks, ResNet, DenseNet, and EfficientNet show varying accuracy levels, with ResNet ranging from 50.6% to 72.12%, DenseNet from 52.4% to 73.05%, and EfficientNet from 48% to 69.53% While these methods are effective in certain contexts, they may require additional optimization for broader applicability.
FNNs are utilized mainly for classification, demonstrating strong performance with 90.06% accuracy, 92% sensitivity, 87.78% precision, and 89.6% specificity This makes FNNs another reliable choice for classification tasks in mental illness diagnosis
VGG16 achieves an impressive AUC of 0.83 and an accuracy of 85.27%, along with a precision rate of 86.33% While it demonstrates strong performance, it may not offer the same level of robustness as Convolutional Neural Networks (CNNs) and Feedforward Neural Networks (FNNs).
CNNs and FNNs are highly effective for various diagnostic applications, demonstrating superior performance across multiple metrics Meanwhile, GNNs, MLPs, and DW-S2MTL exhibit significant potential in niche areas like feature selection and optimization Additionally, models such as ResNet, DenseNet, and EfficientNet indicate the necessity for further enhancements in their capabilities.
3.2.2.2 Advantages and Disadvantages of NN and DL approaches
Table 3.4 Advantages and Disadvantages of NN and DL approaches
CNN improve accuracy, diagnosis, classify, early detection
- Efficient feature extraction due to parameter sharing
- High generalizability across diverse datasets
- Effectively distinguishes between normal and Alzheimer's disease (AD) individuals with high reproducibility and sensitivity
- Classifies various stages of Alzheimer's disease
- Highly dependent on the quality and quantity of input data
- Susceptible to overfitting, especially with limited data
-Achieves high accuracy for AD/MCI/NC classification with effective graph representation and robust performance
Requires additional genetic information and higher memory load due to complex graph structures
-Performs well with straightforward implementation and benefits from data fusion strategies
Lower accuracy compared to GNN and limited in handling complex relationships between data points
[65, 66] DW-S2MTL feature selection and classification, improve the accuracy of diagnostic classification
-Improved diagnostic accuracy through iterative feature selection and adaptive weighting, effectively handling complex data distributions
Computationally intensive and may be sensitive to initial clustering results
- Strong performance on diverse datasets with identity connections enhancing feature extraction
- Capable of handling complex data with deep layers
- Limited representation of information across different datasets
- High computational cost and parameter count, with limited ability to fully capture data representation
- falls short when compared to traditional
Application Advantages Disadvantages gold standard methods such as amyloid PET scans or CSF analysis
- Efficient use of parameters and strong gradient propagation through dense connections
- Efficient feature reuse through dense connections, leading to improved gradient flow and reduced vanishing gradient issues
- More complex architecture leading to higher computational demands
- Can become computationally intensive and memory-consuming due to the dense connections
- Optimized scaling leading to high efficiency
- Scalable with better accuracy and efficiency using compound scaling
- Poor performance on complex problems like brain MRI classification due to limited channel information flow
- Suboptimal for complex tasks with information transfer limitations through inverted residual blocks
[68] FNN Classify strong performance in detecting Alzheimer's disease - issues with generalizability
- solving the classification problem of small samples and high-dimensional data - Computationally intensive and potential loss of information due to high dimensionality and convolutional operations
Table 3.4 provides an overview of various neural network (NN) and deep learning (DL) approaches for diagnosing mental illnesses, highlighting their advantages and disadvantages Convolutional Neural Networks (CNNs) are effective in feature extraction and generalization across datasets but are computationally demanding and susceptible to overfitting Graph Neural Networks (GNNs) excel in differentiating between normal individuals and those with Alzheimer's disease, achieving high accuracy but requiring additional genetic data and increased memory Multi-Layer Perceptrons (MLPs) are easy to implement and benefit from data fusion but have lower accuracy than GNNs The DW-S2MTL enhances diagnostic precision through feature selection and adaptive weighting, although it is computationally intensive and sensitive to initial clustering ResNet manages complex data well with its deep architecture but faces limitations in data representation and incurs high computational costs DenseNet improves gradient flow and mitigates vanishing gradient issues while using parameters efficiently, yet it demands significant computational resources EfficientNet offers high efficiency and scalability through compound scaling, but struggles with complex tasks like brain MRI classification due to restricted channel information flow Feedforward Neural Networks (FNNs) perform well in detecting Alzheimer's disease but have generalizability challenges Finally, VGG16 is adept at addressing classification issues in small, high-dimensional datasets but is computationally intensive and may lead to information loss during convolutional operations.
3.2.3 Analysis of AI application to diagnose Mental Illness based on Traditional ML models
This section evaluates the application of AI techniques using traditional machine learning (ML) models for diagnosing mental illnesses, highlighting their advantages and disadvantages A comprehensive analysis of various models is presented, focusing on performance metrics and clinical effectiveness Traditional ML models are generally more interpretable and demand less computational power than deep learning models, making them suitable for scenarios with less complex data or when interpretability is essential.
Table 3.5 Analysis of AI application to diagnose Mental Illness based on Traditional ML models
Ref Approach Targets for Application
AUC ACC SEN PRE SPE
[50, 56] KNN diagnosis, classify, early detection x 71.4%-
SVM improve accuracy, diagnostic support, diagnose, classify, early detection, optimizing
[48, 50] DT diagnostic support, diagnose, classify x 73.2%-
[41, 50, 56] LDA shorten, diagnose, classify, early detection x 43.8%-
[41, 48, 63] LR Shorten, diagnostic support, predict 0.93 - 0.96 73.2%-90% x x x
Explaination: x: no information, AUC: Area Under the Curve , ACC: Accuracy Rate , SEN:Sensitivity , PRE: Precision , SPE: Specitivity
Table 3.5 evaluates traditional machine learning models for diagnosing mental illnesses, detailing their performance metrics The KNN algorithm demonstrates a diagnosis accuracy of 71.4% to 77.5% and sensitivity ranging from 62.5% to 79.7%, though it has a lower precision of 63.7% In contrast, SVM models show a broader range of accuracy from 41.1% to 93.83%, with an AUC between 0.87 and 0.96, sensitivity from 9.4% to 92.31%, and precision between 31.6% and 47.7% This variability highlights the adaptability of SVMs across different diagnostic scenarios.
Decision Trees (DT) provide dependable diagnostic support and classification with an accuracy ranging from 73.2% to 75.3%, sensitivity between 78.5% and 79.3%, and specificity from 58.3% to 72.1%, though they may lack precision compared to other methods Linear Discriminant Analysis (LDA) excels in classification and early detection, achieving an accuracy of 84.3%, sensitivity of 70.5%, and a precision rate of 75.2%, making it a trustworthy option for diagnosing mental illnesses Logistic Regression (LR) aims to enhance diagnostic support and predict outcomes, showcasing an AUC between 0.93 and 0.96, accuracy from 73.2% to 90%, and sensitivity ranging from 56% to 90.9%, indicating high reliability across multiple metrics.
LASSO excels in prediction and optimization, achieving an impressive AUC of 0.92 to 0.96 and an accuracy of 90% SVR focuses on reducing diagnostic times, with an AUC between 0.905 and 0.943, making it suitable for specific diagnostic tasks, though less versatile than other methods Naive Bayes (NB) stands out in classification, boasting an accuracy range of 93% to 98%, positioning it as one of the leading classification techniques Generalized Linear Models (GLM) aim to enhance accuracy, but with an AUC of 0.5865 to 0.6344, they may fall short compared to other approaches.
In summary, SVM and LR models exhibit versatile performance across various diagnostic scenarios, while KNN, DT, and LDA maintain reliable accuracy and sensitivity Additionally, models such as LASSO and NB showcase robust classification abilities.
3.2.3.2 Advantages and Disadvantages of Traditional ML models
The models have proven to be highly effective in assessing mental illness diagnoses for targeted purposes, fitting seamlessly within the broader system Nevertheless, it is essential to acknowledge that each approach comes with distinct advantages and disadvantages Understanding these strengths and limitations is vital for their proper use in clinical environments.
Table 3.6 Advantages and Disadvantages of Traditional ML models
Ref Approach Targets for Application Advantages Disadvantages
56] KNN diagnosis, classify, early detection - simplicity and effectiveness in classification tasks - computationally intensive with large datasets
SVM improve accuracy, diagnostic support, diagnose, classify, early detection, optimizing
- effective for high-dimensional spaces and maintain high accuracy
- High accuracy in classification, as demonstrated by high AUC values across multiple disorders
- handles high-dimensional data and constructs a clear margin of separation between different classes, leading to high classification accuracy
- effectively handles high-dimensional data and finds the optimal hyperplane that separates classes with maximum margin
- effectively combines neuropsychological scores and imaging data to improve diagnostic accuracy
- leverages multi-view feature representations from multiple templates to enhance classification accuracy
- high classification accuracy and effective handling of high-dimensional data in distinguishing between Alzheimer's disease and healthy controls
- Computationally intensive and less effective with larger datasets, requiring careful tuning of parameters
- Relatively lower precision and recall in certain classifications compared to other models, indicating potential issues with false positives and false negatives
- complex and may lack interpretability needed for clinical use
- computationally intensive and challenging to interpret compared to simpler models like logistic regression
- computationally intensive and less effective when dealing with non-separable patterns without appropriate kernel functions
- computationally intensive and requires careful tuning of parameters to handle complex data integrations
- computationally intensive and require careful parameter tuning to optimize performance
50] DT diagnostic support, diagnose, classify
- Simple to interpret and implement, requires less data preprocessing
- High interpretability allowing clear decision- making processes
- Prone to overfitting, lower accuracy compared to more complex models
- Potential for overfitting and lower
Ref Approach Targets for Application Advantages Disadvantages generalization compared to more complex models
50, 56] LDA shorten, diagnose, classify, early detection - Simple and efficient for binary classification, providing good performance with relatively low computational cost
- computationally efficient and works well with linearly separable data, providing straightforward interpretation of class boundaries
- Assumes linear relationships between features, which may not capture complex data patterns effectively
- low overall accuracy and sensitivity, making it less effective for distinguishing between the categories
- assumes normality and equal covariance matrices across classes, which may not hold for all datasets
- provides a straightforward interpretation of coefficients
- straightforward to implement and provides interpretable results with high diagnostic accuracy
- underperform in handling non-linear relationships within the data
- prone to overfitting with excessive features, requiring careful feature selection to maintain performance
- not capture complex nonlinear relationships within the data, potentially limiting its predictive performance compared to more sophisticated models
- reduces the number of features while maintaining high predictive accuracy, making it useful for feature selection and model simplification
- effectively performs feature selection and reduces overfitting by shrinking less important feature coefficients to zero
- reduces overfitting by selecting only the most relevant features, leading to more interpretable models
- exclude important features if they are highly correlated, leading to a loss of relevant information
- struggle with correlated features, often selecting only one feature from a group of highly correlated features, which may result in loss of important information
- exclude correlated features, potentially losing important information in the process
Ref Approach Targets for Application Advantages Disadvantages
[41] SVR shorten - handles high-dimensional data and provides robust performance metrics in diagnostic classifications
- computationally intensive and requires careful parameter tuning to prevent overfitting and optimize performance
Potential research gaps of AI techniques in the diagnosis of mental illnesses
From all the analyze above, we summarize some potential research gaps in Figure
The small sample sizes utilized in recent research can significantly impact the generalizability of AI models, leading to overfitting, where models excel on training data but struggle with new, unseen data For example, Andrew's research indicates that training on small, non-representative datasets can cause models to develop biases that mirror the specific traits of the training data, rather than capturing genuine underlying patterns This bias can adversely affect predictive performance across different populations or settings To enhance the accuracy and reliability of AI models, researchers should prioritize larger application sample sizes.
Despite the availability of highly accurate and reliable models, their application in diagnosing mental diseases remains limited Research by Ehiabhi indicates that models such as Logistic Regression demonstrate significant accuracy and robustness across various applications, including mental health diagnostics, yet they are underutilized in clinical settings This highlights a promising research direction for employing these models in the diagnosis of mental diseases, particularly for conditions that have not been extensively studied.
There is a significant gap in real-world validation studies to evaluate the performance of AI models in everyday clinical settings Jiang [76] emphasizes that without such validation, these models may struggle with the complexities and variabilities of actual clinical environments Conducting studies to assess their effectiveness, accuracy, and adaptability in diverse scenarios is essential for ensuring their reliability and utility in mental health diagnostics and treatment Future research should prioritize clinical trials and pilot studies to validate AI models in real-world contexts, confirming their practicality and effectiveness in routine clinical practice.
Future research should focus on integrating multimodal data to enhance AI diagnostic capabilities in mental health By combining neuroimaging, genetic, clinical, and behavioral data, a more comprehensive understanding of mental health conditions can be achieved This holistic approach promises more accurate and personalized diagnoses, ultimately improving patient outcomes Notably, the integration of neuroimaging and genetic data has demonstrated potential in better identifying mood disorders and enhancing diagnostic precision.
CONCLUSIONS AND RECOMMENDATIONS
Conclusion
AI techniques, including machine learning (ML) and deep learning (DL), have been widely utilized in the diagnosis and early detection of various mental health conditions, demonstrating encouraging outcomes A detailed evaluation of 33 studies on the application of AI in analyzing data for mental health diagnoses reveals significant findings that highlight the effectiveness of these technologies in improving mental health assessments.
Machine learning models, including support vector machines and random forests, effectively predict mental health conditions by analyzing electronic health records and patient data, facilitating early interventions In parallel, deep learning models, especially convolutional neural networks (CNNs), excel in analyzing neuroimaging data, providing accurate diagnoses for conditions such as Alzheimer’s disease and schizophrenia.
Despite advancements in AI for mental health diagnostics, challenges persist, particularly concerning dataset quality and diversity Many AI models rely on limited datasets and small sample sizes, which hinders their ability to generalize across wider populations Additionally, the opaque nature of many AI algorithms complicates clinicians' understanding and trust in the results Thus, it is essential to tackle these issues to improve the reliability and clinical relevance of AI-driven diagnostic tools in mental health.
Integrating diverse data types, including neuroimaging, genetic, clinical, and behavioral information, enhances our understanding of mental health conditions Additionally, AI models streamline data analysis and diagnosis, enabling faster identification of mental health issues while alleviating the workload for healthcare professionals.
Overall, the integration of AI in mental health diagnostics shows great promise in transforming the field Despite all the mentioned challenges, the potential benefits of
AI will ultimately lead to better patient outcomes and more efficient healthcare delivery.
Recommendations
Future research should focus on gathering and integrating high-quality, diverse datasets, including neuroimaging, genetic, clinical, and behavioral data, to offer a comprehensive understanding of patient mental health Furthermore, implementing real-world clinical trials using these integrated models will be essential for validating their effectiveness and practicality.
Researchers are working on developing AI models for real-time diagnostics that seamlessly integrate into telehealth platforms, including wearable devices and mobile health apps These innovations aim to deliver immediate support to patients with minimal latency, enhancing the effectiveness of remote healthcare solutions.
Researchers must focus on creating explainable AI models that offer clear insights into their reasoning and decision-making processes This transparency is essential for clinicians to comprehend how AI systems arrive at their conclusions, fostering trust and enabling effective utilization in medical settings.
1 Sun, L., et al., Artificial intelligence for healthcare and medical education: a systematic review Am J Transl Res, 2023 15(7): p 4820-4828
2 Hashimoto, D.A., et al., Artificial Intelligence in Anesthesiology: Current Techniques,
Clinical Applications, and Limitations Anesthesiology, 2020 132(2): p 379-394
3 Bernauer, S.A., N.U Zitzmann, and T Joda, The Use and Performance of Artificial
Intelligence in Prosthodontics: A Systematic Review Sensors (Basel), 2021 21(19)
4 Ossowska, A., A Kusiak, and D Świetlik, Artificial Intelligence in Dentistry-Narrative
Review Int J Environ Res Public Health, 2022 19(6)
5 Graham, S., et al., Artificial Intelligence for Mental Health and Mental Illnesses: an
6 Pai, V.V and R.B Pai, Artificial intelligence in dermatology and healthcare: An overview
7 Noorbakhsh-Sabet, N., et al., Artificial Intelligence Transforms the Future of Health Care
8 Wiederhold, G and J McCarthy, Arthur Samuel: Pioneer in Machine Learning 1992 36: p 329-331
9 Patel, J., et al., Predicting stock market index using fusion of machine learning techniques
10 Esteva, A., et al., Dermatologist-level classification of skin cancer with deep neural networks 2017 542(7639): p 115-118
11 Jurafsky, D and H Martin, JH (2000) Speech and Language processing 2012, New
12 Koren, Y., R Bell, and C Volinsky, Matrix factorization techniques for recommender systems Computer, 2009 42(8): p 30-37
13 LeCun, Y., Y Bengio, and G Hinton, Deep learning 2015 521(7553): p 436-444
14 Schroff, F., D Kalenichenko, and J Philbin FaceNet: A unified embedding for face recognition and clustering in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2015
15 Hinton, G., et al., Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups IEEE Signal processing magazine, 2012 29(6): p
16 Bojarski, M., et al., End to end learning for self-driving cars arXiv preprint arXiv:1604.07316, 2016
17 Alowais, S.A., et al., Revolutionizing healthcare: the role of artificial intelligence in clinical practice BMC Med Educ, 2023 23(1): p 689
18 Awasthi, R., et al., Artificial Intelligence in Healthcare: 2023 Year in Review medRxiv,
19 Bhattamisra, S.K., et al., Artificial intelligence in pharmaceutical and healthcare research
Big Data and Cognitive Computing, 2023 7(1): p 10
20 Sun, G and Y.H Zhou, AI in healthcare: navigating opportunities and challenges in digital communication Front Digit Health, 2023 5: p 1291132
21 van Genk, C., et al., Current insights of community mental healthcare for people with severe mental illness: A scoping review Front Psychiatry, 2023 14: p 1156235
22 Kessler, R.C., et al., Design and field procedures in the US National Comorbidity Survey
Replication Adolescent Supplement (NCS-A) Int J Methods Psychiatr Res, 2009 18(2): p
23 Singh, B., et al., Effectiveness of physical activity interventions for improving depression, anxiety and distress: an overview of systematic reviews Br J Sports Med, 2023 57(18): p
24 Bannatyne, A.J., et al., A systematic review of mental health interventions to reduce self- stigma in medical students and doctors Front Med (Lausanne), 2023 10: p 1204274
25 White, T., Hadoop: The definitive guide 2012: " O'Reilly Media, Inc."
26 Zaharia, M., et al., Apache spark: a unified engine for big data processing
27 Bradshaw, S., E Brazil, and K Chodorow, MongoDB: the definitive guide: powerful and scalable data storage 2019: O'Reilly Media
28 Szeliski, R., Computer vision: algorithms and applications 2022: Springer Nature
29 Babaji, P., et al., Evaluation of child preference for dentist attire and usage of camouflage syringe in reduction of anxiety Eur J Dent, 2017 11(4): p 531-536
30 De Choudhury, M and E Kıcıman, The Language of Social Support in Social Media and its Effect on Suicidal Ideation Risk Proc Int AAAI Conf Weblogs Soc Media, 2017 2017: p 32-41
31 McCarter, R., et al., Physical health disparities and severe mental illness: A longitudinal comparative cohort study using hospital data in Northern Ireland Eur Psychiatry, 2023
32 Pizzol, D., et al., Relationship between severe mental illness and physical multimorbidity: a meta-analysis and call for action BMJ Ment Health, 2023 26(1)
33 Goh, Y.S.S., J.Q.Y Ow Yong, and A.Z Li, Effectiveness of Illness Management and
Recovery program on people with severe mental illnesses: a systematic review and meta- analysis Front Psychiatry, 2023 14: p 1162288
34 Holt, J., et al., Protocol for a systematic review and meta-analysis of the prevalence of mental illness among nursing home residents Syst Rev, 2024 13(1): p 109
35 Feng, W., et al., Automated MRI-Based Deep Learning Model for Detection of Alzheimer's
Disease Process Int J Neural Syst, 2020 30(6): p 2050032
36 Miner, A.S., et al., Key Considerations for Incorporating Conversational AI in
37 Santos, R.M.S., et al., The associations between screen time and mental health in adolescents: a systematic review BMC Psychology, 2023 11(1): p 127
38 Ogilvie, D., et al., Systematic reviews of health effects of social interventions: 2 Best available evidence: how low should you go? J Epidemiol Community Health, 2005
39 Ahmed, N., et al., Artificial Intelligence Techniques: Analysis, Application, and Outcome in Dentistry-A Systematic Review Biomed Res Int, 2021 2021: p 9751564
40 Allesứe, R.L., et al., Deep Learning for Cross-Diagnostic Prediction of Mental Disorder
Diagnosis and Prognosis Using Danish Nationwide Register and Genetic Data JAMA
41 Ma, Y., et al., Implementing machine learning in bipolar diagnosis in China Transl
42 Wall, D.P., et al., Use of artificial intelligence to shorten the behavioral diagnosis of autism PLoS One, 2012 7(8): p e43855
43 Zhou, D., Z Liao, and R Chen, Deep Learning Enabled Diagnosis of Children's ADHD
Based on the Big Data of Video Screen Long-Range EEG J Healthc Eng, 2022 2022: p
44 Sharma, A and W Verbeke, Understanding importance of clinical biomarkers for diagnosis of anxiety disorders using machine learning models PLoS One, 2021 16(5): p e0251365
45 Andrade, E., et al., A Protocol for the Diagnosis of Autism Spectrum Disorder Structured in Machine Learning and Verbal Decision Analysis Comput Math Methods Med, 2021
46 Benacek, J., et al., Identification of Predictors of Mood Disorder Misdiagnosis and
Subsequent Help-Seeking Behavior in Individuals With Depressive Symptoms: Gradient- Boosted Tree Machine Learning Approach JMIR Ment Health, 2024 11: p e50738
47 Fernandes, B.S., et al., Precision psychiatry with immunological and cognitive biomarkers: a multi-domain prediction for the diagnosis of bipolar disorder or schizophrenia using machine learning Transl Psychiatry, 2020 10(1): p 162
48 Itani, S., et al., Towards interpretable machine learning models for diagnosis aid: A case study on attention deficit/hyperactivity disorder PLoS One, 2019 14(4): p e0215720
49 Yoon, J.M., et al., Enhancing foveal avascular zone analysis for Alzheimer's diagnosis with AI segmentation and machine learning using multiple radiomic features Sci Rep,
50 Amini, M., et al., Diagnosis of Alzheimer's Disease Severity with fMRI Images Using
Robust Multitask Feature Extraction Method and Convolutional Neural Network (CNN)
51 Wall, D.P., et al., Use of machine learning to shorten observation-based screening and diagnosis of autism Transl Psychiatry, 2012 2(4): p e100
52 Chen, K., et al., A comparative study of GNN and MLP based machine learning for the diagnosis of Alzheimer's Disease involving data synthesis Neural Netw, 2024 169: p 442-
53 Yu, X., et al., The Feasibility of Early Alzheimer's Disease Diagnosis Using a Neural
Network Hybrid Platform Biosensors (Basel), 2022 12(9)
54 Zheng, J., et al., Diagnosis of Schizophrenia Based on Deep Learning Using fMRI Comput
55 Carcagnì, P., et al., Convolution Neural Networks and Self-Attention Learners for
Alzheimer Dementia Diagnosis from Brain MRI Sensors (Basel), 2023 23(3)
56 Amini, M., et al., Diagnosis of Alzheimer's Disease by Time-Dependent Power Spectrum
Descriptors and Convolutional Neural Network Using EEG Signal Comput Math Methods
57 Garrard, P., et al., Machine learning approaches to diagnosis and laterality effects in semantic dementia discourse Cortex, 2014 55: p 122-9
58 Fathi, S., et al., A Deep Learning-Based Ensemble Method for Early Diagnosis of
Alzheimer's Disease using MRI Images Neuroinformatics, 2024 22(1): p 89-105
59 Abdelaziz, M., T Wang, and A Elazab, Alzheimer's disease diagnosis framework from incomplete multimodal data using convolutional neural networks J Biomed Inform, 2021
60 Dai, P., et al., A hybrid manifold learning algorithm for the diagnosis and prognostication of Alzheimer's disease AMIA Annu Symp Proc, 2015 2015: p 475-83
61 Lama, R.K., et al., Diagnosis of Alzheimer's Disease Based on Structural MRI Images
Using a Regularized Extreme Learning Machine and PCA Features J Healthc Eng, 2017
62 Uemura, T., et al., Computer-assisted system for diagnosis of Alzheimer disease using data base- independent estimation and fluorodeoxyglucose- positron-emission tomography and 3D-stereotactic surface projection AJNR Am J Neuroradiol, 2011 32(3): p 556-9
63 Moore, P.J., T.J Lyons, and J Gallacher, Using path signatures to predict a diagnosis of
64 Guan, H., et al., A parallel attention-augmented bilinear network for early magnetic resonance imaging-based diagnosis of Alzheimer's disease Hum Brain Mapp, 2022 43(2): p 760-772.