The application of artificial intelligence in the field of mental health: a systematic review

Abstract

Introduction

The integration of artificial intelligence in mental health care represents a transformative shift in the identification, treatment, and management of mental disorders. This systematic review explores the diverse applications of artificial intelligence, emphasizing both its benefits and associated challenges.

Methods

A comprehensive literature search was conducted across multiple databases based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses, including ProQuest, PubMed, Scopus, and Persian databases, resulting in 2,638 initial records. After removing duplicates and applying strict selection criteria, 15 articles were included for analysis.

Results

The findings indicate that AI enhances early detection and intervention for mental health conditions. Various studies highlighted the effectiveness of AI-driven tools, such as chatbots and predictive modeling, in improving patient engagement and tailoring interventions. Notably, tools like the Wysa app demonstrated significant improvements in user-reported mental health symptoms. However, ethical considerations regarding data privacy and algorithm transparency emerged as critical challenges.

Discussion

While the reviewed studies indicate a generally positive trend in AI applications, some methodologies exhibited moderate quality, suggesting room for improvement. Involving stakeholders in the creation of AI technologies is essential for building trust and tackling ethical issues. Future studies should aim to enhance AI methods and investigate their applicability across various populations.

Conclusion

This review underscores the potential of AI to revolutionize mental health care through enhanced accessibility and personalized interventions. However, careful consideration of ethical implications and methodological rigor is essential to ensure the responsible deployment of AI technologies in this sensitive field.

Introduction

The incorporation of artificial intelligence (AI) into mental health care has become a revolutionary influence, altering the ways in which mental health disorders are diagnosed, treated, and managed [1]. This systematic review seeks to explore the diverse applications of AI in mental health, highlighting both its potential benefits and the challenges it poses. Over the past few years, numerous studies have underscored the capacity of AI to enhance the early detection and diagnosis of mental health conditions, facilitating timely and appropriate interventions.

The application of AI in mental health care not only enhances early detection and diagnosis but also enables personalized treatment and improved patient engagement. AI analyzes patient data to tailor interventions and includes tools like chatbots for support, especially in areas with limited access to traditional services. However, challenges such as ethical issues, data privacy, and the need for effective implementation frameworks must be addressed. This review will explore these aspects to ensure responsible integration of AI into mental health systems [2].

Gaffney and colleagues worked on the AI in mental health care in their article. They discuss the significant role of conversational agents, which are AI-driven tools such as chatbots, in providing immediate support and intervention for individuals experiencing mental health issues. The review highlights that conversational agents can enhance accessibility to mental health resources, deliver personalized therapeutic interactions, and facilitate ongoing engagement with users. By leveraging natural language processing and machine learning, these agents can effectively assess user needs, provide tailored responses, and promote self-management strategies, ultimately contributing to improved mental health outcomes [3].

In an examination of the transformative potential of AI in mental health care, Olawade et al. (2022) emphasize the increasing integration of AI technologies into mental health services, which offer innovative solutions for diagnosis, treatment, and patient engagement. Current trends suggest that AI can facilitate the early detection of mental health disorders, optimize therapeutic interventions through personalized approaches, and enhance access to care, particularly for underserved populations. Additionally, the authors discuss future prospects, highlighting the importance of ethical considerations and robust implementation strategies to ensure the effective integration of AI applications into clinical practice. By addressing both the opportunities and challenges associated with AI in mental health, this review aims to provide a comprehensive understanding of how these technologies can improve mental health outcomes and reshape the landscape of mental health care [4].

Despite these promising applications, the integration of AI in mental health care is not without its challenges. Ethical considerations surrounding data privacy and the transparency of AI algorithms are critical issues that require careful examination. Tavory (2023) highlighted the necessity for ethical frameworks to guide the development and deployment of AI technologies in mental health, advocating for a balanced approach that prioritizes patient safety and informed consent [5].

Tornero-Costa et al. (2022) critically assess the methodological and quality flaws present in the use of AI within mental health research. The authors identify several key issues that undermine the reliability and validity of AI applications in this field, including inadequate study designs, insufficient sample sizes, and a lack of transparency in reporting methodologies. Furthermore, they emphasize the importance of rigorous methodological standards to enhance the credibility of AI-driven research outcomes. By highlighting these flaws, the review underscores the need for improved research practices to ensure that AI technologies can be effectively and ethically integrated into mental health care, ultimately advancing the field and benefiting patient outcomes [6].

Furthermore, the previous review noted critical issues such as inadequate transparency in reporting AI model features and data preprocessing techniques, which are essential for ensuring reproducibility and reliability in research. Many studies included in the earlier review did not sufficiently assess the quality of data or the appropriateness of AI methodologies for their specific applications, leading to potential biases and an overly optimistic view of AI performance. These shortcomings underscore the necessity for a new systematic review that not only addresses these methodological gaps but also provides a detailed overview of AI applications in mental health, including their benefits and challenges.

This article seeks to address this gap by consolidating recent research findings, assessing methodologies, and pinpointing areas for further exploration. Through this approach, we aim to enhance the understanding of how AI can be successfully incorporated into mental health care, ultimately improving the quality of care for those experiencing mental health difficulties.

Methods

Reviewing the literature

The systematic review on the application of AI in mental health involved a comprehensive search across multiple databases to gather a robust collection of relevant studies.

The search encompassed both Persian databases, including IranDoc and ISC, and international databases such as ProQuest, PubMed, Scopus, Web of Science, Cochrane, Wiley, and the American Psychiatric Association’s APA database.

For the initial search phase, we employed a search syntax focusing on the terms “artificial intelligence” and “mental health.” The results varied significantly, with international databases yielding a higher number of records. Specifically, ProQuest identified 944 records, PubMed found 829, and Scopus retrieved 353.In contrast, Persian databases like IranDoc and ISC produced no relevant records, indicating a limited volume of research in the application of AI in mental health. Notably, SID, a Persian database, contributed four records, while Web of Science retrieved 171 records, and E-journal platforms like Cochrane and Wiley yielded 65 and 7 records, respectively. Overall, the search across all databases generated 2,638 initial records.

To refine the dataset, we removed duplicate entries, reducing the total to 1,471 unique records. Following this, we conducted a title screening phase to further narrow down the records to those most relevant to our research question. After this rigorous screening, only 369 records were deemed suitable for the subsequent phases of the systematic review process.

This comprehensive approach highlights the diversity of resources and the necessity of detailed screening processes in conducting a systematic review, particularly when analyzing the intersection of AI and mental health research.

This comprehensive approach highlights the diversity of resources and the necessity of detailed screening processes in conducting a systematic review, particularly when analyzing the intersection of AI and mental health research.The search strategy, along with thorough deduplication and screening, allowed for a focused dataset essential for synthesizing accurate insights into AI’s role and effectiveness within the mental health domain.

Selection criteria and search strategy

The systematic review followed a multi-stage selection process to ensure a thorough and unbiased analysis. Initially, we retrieved a total of 2, 638 records from various databases, including ProQuest, PubMed, Scopus, Web of Science, Cochrane, and Wiley, along with Persian databases such as IranDoc, ISC, and SID. Searches were conducted using specific syntaxes tailored to each database, primarily combining the terms “artificial intelligence” and “mental health” to capture relevant literature.

From the initial 2,638 records, we identified and removed 1,167 duplicates, resulting in a pool of 1,471 articles as of July 12, 2024. In the first screening stage, we reviewed these articles based on title relevance, leading to the identification of 369 articles pertinent to the topic by July 15, 2024. We then examined the abstracts of the remaining articles, yielding 80 articles that met the initial criteria. However, 23 articles were excluded due to the lack of full text, reducing the number to 57 from July 16–29, 2024. A thorough review of the full-text articles resulted in the exclusion of 42 articles based on specific criteria, leaving 15 articles for final inclusion from August 2–15, 2024. The reasons for excluding articles at each stage of the systematic review process are as follows: Initially, from the 2,638 records identified, 1,167 duplicates were removed, resulting in 1,471 unique articles. During the title screening phase, 1,471 articles were assessed for relevance, leading to the exclusion of 1,102 articles that did not meet the criteria based on their titles. Subsequently, the remaining 369 articles were further evaluated through abstract screening, where 289 articles were excluded due to insufficient relevance to the research question or lack of focus on artificial intelligence applications in mental health. In the full-text review phase, 57 articles were considered, but 42 were excluded for reasons such as the absence of full-text availability, lack of methodological rigor, or failure to address the specific aims of the review. This left 15 articles that met all inclusion criteria. Each of these articles underwent a comprehensive risk of bias assessment, ensuring that the final selection was based on robust methodological standards and relevance to the topic of AI in mental health Table 1.

Table 1 Search strategy (search syntax: (artificial intelligence AND mental))

We conducted a comprehensive risk of bias assessment for the 15 final articles to evaluate their methodological rigor and reliability, concluding this assessment phase on August 29, 2024. Data extraction from the selected studies was completed between September 5–15, 2024, ensuring that each study’s findings and details were systematically recorded for further analysis. This rigorous multi-stage approach facilitated a focused and high-quality selection of studies that addressed the application of artificial intelligence in mental health (see Fig. 1).

Fig. 1

Diagram for the inclusion and exclusion of articles

Quality appraisal methods

The comparative analysis of the 15 studies reveals significant insights regarding the quality assessment tools utilized in AI applications within mental health interventions.

The studies, spanning from 2018 to 2023, illustrate a growing interest in the integration of AI technologies in mental health research. Various quality assessment tools were used across these studies, reflecting the researchers’ commitment to rigorous evaluation of their methodologies (see Table 1).

A notable number of studies utilized the Mixed Methods Appraisal Tool (MMAT), which was applied in four studies, including those by Inkster et al. (2018), Chin et al. (2021), Götzl et al. (2022), and Pei (2022). The MMAT is particularly effective for evaluating mixed-methods research, allowing for a nuanced understanding of complex mental health dynamics. Its application across multiple studies underscores its robustness in assessing research quality in this field [7].

The Newcastle–Ottawa Scale (NOS) was another prominent tool, used in several studies, including those by Xiao Li (2023), Pei (2022), and Tate (2020). The NOS is well-regarded for its application in observational studies, providing a systematic approach to evaluating the quality of non-randomized studies. The consistent use of the NOS across these studies indicates its reliability and relevance in assessing the quality of observational research in AI-driven mental health interventions [8].

Additionally, the Joanna Briggs Institute (JBI) tool was employed in various experimental studies, including those by Didarul Alam et al. (2021), Yanqi Guo (2023), Ilona Halim et al. (2023), and Naveen Kumari & Rekha Bhatia (2022). The JBI tool is designed to assess the methodological quality of experimental and quasi-experimental studies, contributing to a comprehensive understanding of the effectiveness of AI interventions in mental health [9].

Despite the predominance of established tools like the MMAT, NOS, and JBI, variability in the quality of studies was observed, particularly with the application of different methodologies. For instance, the study by Rohit Rastogi et al. (2022) utilized the JBI tool but received a moderate quality score, indicating that even established assessment tools can yield varied results based on the study design and execution.

In summary, the analysis of the quality assessment tools used in these studies highlights the importance of methodological rigor in AI-related mental health research. The diversity of tools employed—MMAT, NOS, and JBI—demonstrates the multifaceted nature of research methodologies in this area. The consistent application of these established tools enhances the credibility of findings and paves the way for future advancements in the integration of AI technologies in mental health interventions.

Results

Application of AI in mental health

The 15 reviewed studies illustrate various interpretations and applications of AI in mental health contexts. Inkster et al. (2018) introduced Wysa, an AI-driven chatbot aimed at improving mental well-being by enabling self-reflection and resilience through conversational support [10](Table 2).

Table 2 Quality assessment of articles

Lei (2023) and Xiao Li (2023) explored more advanced AI techniques, with Lei et al. using Long Short-Term Memory (LSTM) models for emotion recognition tasks and Li applying deep learning (CNN) to assess mental health states by analyzing social media text data. These approaches underscore the adaptability of AI in processing complex data patterns, such as emotional and mental health indicators [12, 27].

Götzl et al. (2022) summarized AI’s broader purpose as learning systems that evaluate large information sets, often applied in platforms like YouTube and Spotify, showcasing AI’s versatility in various interactive media [28].

These studies reveal the diversity and scope of AI applications in mental health, ranging from emotional recognition and chatbot support to large-scale predictive models, illustrating AI’s role in advancing personalized and accessible mental health solutions.

Tools and algorithms used in AI in mental health

The comparative analysis of the 15 articles concerning the application of AI in mental health reveals a diverse landscape of methodologies, participant demographics, and AI technologies. Each study contributes unique insights into the effectiveness and acceptability of AI-driven interventions (Table 2).

The first study by Inkster et al. (2018) employed a mixed-methods approach to evaluate the Wysa app, an AI-enabled chatbot designed for mental well-being. This study focused on a global user base of individuals self-reporting symptoms of depression. The findings indicated that users who engaged more frequently with the app experienced significant improvements in their depressive symptoms, with 67.7% of participants finding the app helpful. This highlights the potential effectiveness of text-based conversational AI in mental health support [10].

In contrast, Xiao Li (2023) conducted a quantitative study aimed at understanding depression among the elderly. Utilizing Long Short-Term Memory (LSTM) networks, the research focused on emotional recognition, demonstrating how AI can model temporal information to improve mental health interventions. The integration of AI attention mechanisms enhanced the accuracy of emotional feature recognition, suggesting that advanced AI techniques can significantly contribute to understanding and addressing mental health issues [12].

Liang et al. (2022) evaluated the mental health of college students using a Convolutional Neural Network (CNN) to analyze text data from online forums. This experimental study indicated that AI could effectively monitor and assess mental health states, providing personalized psychological support based on user-generated content. The study’s results affirmed the feasibility of AI in delivering tailored mental health interventions [13].

Chin et al. (2023) explored user interactions with the SimSimi chatbot across various cultural contexts, analyzing 96,197 conversations from Eastern and Western countries. This mixed-methods study revealed cultural differences in the expression of depressive moods, emphasizing the chatbot’s role in social science research and its high acceptability among users. The findings suggest that AI can facilitate large-scale data collection and cultural analysis of mental health expressions [29].

Dadi (2021) utilized machine learning to analyze multimodal data from the UK Biobank, demonstrating how population modeling can derive mental health measures from diverse inputs, including brain imaging. This approach indicates that AI can complement traditional psychometric assessments, enhancing the understanding of mental health determinants in large populations [15].

Rathnayaka (2022) focused on the design and development of a BA-based AI chatbot, confirming its effectiveness in providing emotional support and remote mental health monitoring. The participatory evaluation highlighted user feedback on the chatbot’s capabilities, reinforcing the importance of user experience in AI applications for mental health [30].

Pei (2022) utilized a neural network algorithm to forecast and assess the mental health status of college students, uncovering notable psychological stressors. This research highlights the effectiveness of AI in recognizing mental health risks and customizing interventions through predictive analytics [17].

Tate (2020) created a model employing machine learning techniques to forecast mental health issues in adolescents, attaining a satisfactory level of predictive accuracy. While the model was not appropriate for clinical application, it established a foundation for future investigations into predicting mental health in adolescents [31].

Didarul Alam et al. (2021) explored the adoption of mobile health (mHealth) technologies during the COVID-19 pandemic by utilizing Structural Equation Modeling and Artificial Neural Networks (ANN). The results demonstrated a positive correlation between mHealth usage and mental well-being, emphasizing the role of AI in improving user engagement during times of crisis [19].

Götzl et al. (2022) explored young people’s attitudes toward AI-informed mHealth apps, revealing a generally positive reception and willingness to use AI for personalized interventions. This study emphasizes the potential of AI to adapt interventions to individual needs, enhancing the effectiveness of mental health support [28].

Onuki et al. (2022) examined the use of wearable devices to estimate mental health conditions based on heart rate data, finding high acceptability among users. This study illustrates how AI can integrate with wearable technology to provide real-time health monitoring [21].

Alamgir et al. (2023) focused on facial expression recognition using a hybrid AI model, demonstrating its ability to categorize emotions effectively. This research highlights the potential of AI in enhancing emotional awareness and understanding in mental health contexts [32].

Yanqi Guo (2023) analyzed the impact of AI technology on family education, revealing significant improvements in mental health outcomes among students receiving AI-based support compared to traditional methods. This study showcases the versatility of AI applications beyond direct mental health interventions [23].

Finally, Halim et al. (2023) tested an individualized virtual reality (iVR) approach to improve self-compassion and reduce depressive symptoms. Although primarily focused on VR, the study emphasized the potential of integrating AI to personalize user experiences and enhance therapeutic outcomes [33].

In conclusion, the combined results from these studies reflect an increasing acceptance and effectiveness of AI in mental health interventions across different demographics and methodologies. The varied applications of AI, including chatbots, predictive models, and wearable technologies, underscore its potential to revolutionize mental health care by offering personalized, accessible, and effective support. As the field progresses, additional research is crucial to enhance these technologies and confirm their effectiveness across diverse populations (Table 3).

Table 3 Extracted data from articles

Quality appraisal methods

In examining the application of AI in mental health studies, various study types and quality assessment tools were employed, with each study exhibiting unique methodological strengths (Table 1).

Inkster et al. (2018) [10] conducted a mixed-methods study, receiving a total score of 7, assessed as “good” quality using the Mixed-Methods Appraisal Tool (MMAT), demonstrating effective integration of both qualitative and quantitative data. Similarly, Xiao Li (2023) [12] and Chin et al. (2023) [29] employed quantitative and mixed-methods approaches, respectively, with their studies scored at 7 by the Newcastle–Ottawa Scale (NOS), indicating high methodological quality and thorough evaluation of mental health indicators related to AI applications.

Dadi (2021) [15] focused on population modeling with AI, achieving a “good” quality rating and a score of 6 from NOS. In a similar context, Liang et al. (2022) [13] conducted an experimental study focused on AI model performance, assessed by NOS and rated “good” with a score of 5, reflecting methodological rigor but with room for further improvement.

A more complex study design was employed by Rathnayaka (2022) [30], who utilized the two-phase prediction and analysis approach to AI applications. This study received a score of 6 on NOS, indicating “moderate” quality, suggesting effective use of predictive techniques but with potential limitations in methodological consistency.

Conversely, Pei (2022) [17] and Tate (2020) [31] undertook model development and empirical study using SEMANN (Structural Equation Modeling – Artificial Neural Network), both achieving “good” quality with a score of 7 under NOS. Didarul Alam et al. (2021) [19] conducted a parallel mixed-method study, assessed by the Joanna Briggs Institute (JBI) checklist, scoring 10 and rated “good,” reflecting strong methodological coherence across study phases.

In randomized controlled trials, Götzl et al. (2022) [28] and Onuki et al. (2022) [21] both received “good” ratings with scores of 9 and 7, respectively, indicating robust experimental control and rigorous assessment in AI application to mental health. Similarly, Alamgir FM et al. (2022) [32] and Yanqi Guo et al. (2023) [23] conducted experimental studies, each with “good” quality ratings, showing consistency in applying AI within structured trials.

Finally, Kumari et al. (2023) [25] and Rohit Rastogi et al. (2022) [34] utilized a more experimental framework, rated as “moderate” quality with scores of 8 and 5 under JBI, illustrating potential for reliable AI-driven results but highlighting the need for additional methodological refinement.

In summary, the overall quality of studies indicates a promising trend in AI applications within mental health research, with studies rated as “good” by recognized assessment tools, although some studies show moderate methodological rigor, suggesting areas for improvement in future research designs.

Demographics of participants

In a comprehensive analysis of participant demographics across various studies examining the application of AI in mental health, several key insights emerge. Inkster et al. (2018) focused on anonymous global users of the Wysa app, specifically targeting individuals who self-reported symptoms of depression. This study provided insights into how AI-enabled conversational tools can engage users in mental health support [10].

Xiao Li (2023) examined elderly participants from two community groups, a control group (n = 46) and an intervention group (n = 47), to understand the factors influencing depression among older adults and to formulate psychological intervention plans. This demographic highlights the importance of addressing mental health in aging populations [12].

Liang et al. (2022) involved 185 college students in sports majors, divided into training and test sets, to evaluate the effectiveness of AI in assessing and improving mental health. The focus on college students underscores the relevance of mental health interventions in educational settings [13].

Chin et al. (2023) analyzed user interactions with the SimSimi chatbot, utilizing a dataset from 96,197 conversations in Eastern countries and 56,586 conversations in Western countries. This cross-cultural analysis provides valuable insights into how different populations express depressive moods through AI platforms [29].

Dadi (2021) utilized a large cohort from the UK Biobank, comprising 11,175 participants, to study cognitive tests and demographic characteristics, emphasizing the potential of AI in population modeling and mental health research [15].

Rathnayaka (2022) conducted a pilot study with 318 individuals globally to evaluate a BA-based AI chatbot designed for mental health support, demonstrating the growing interest in AI applications across diverse populations [30]. Pei (2022) focused on college students, although specific demographic details were not provided, emphasizing the need for targeted mental health interventions in this age group [17].

Tate (2020) studied 7,638 twins from a Swedish adolescent mental health study, providing a unique perspective on genetic and environmental influences on mental health outcomes [31].

Didarul Alam et al. (2021) focused on mHealth users in Bangladesh during the COVID-19 pandemic, emphasizing how global crises affect mental health and the role of technology in delivering support [19].

Götzl et al. (2022) included young people aged 12–25, alongside stakeholders in mHealth app development, to explore attitudes towards AI-informed mental health applications, indicating a collaborative approach to mental health solutions [28].

Onuki et al. (2022) focused on 97 women cancer survivors in Japan, providing insights into the mental health challenges faced by this specific demographic and the potential for AI to offer tailored support [21]. Alamgir et al. (2022) involved 10 Japanese female participants in facial expression recognition studies, emphasizing the intersection of AI and emotional recognition in mental health contexts [32].

Yanqi Guo (2023) studied 320 Japanese middle school students, comparing the effects of AI-based family education systems on mental health, which highlights the importance of early intervention in educational settings [23].

Halim et al. (2023) recruited 36 young adult participants from a university community to evaluate the effects of individualized virtual reality (VR) on self-compassion and depression, showcasing innovative approaches to mental health interventions [33].

Kumari and Bhatia (2022) did not specify participant demographics but focused on deep learning for emotion recognition, indicating the technical aspects of AI applications in mental health [25].

Finally, Rastogi et al. (2022) involved 20 participants in a trial related to ancient practices for mental fitness, reflecting a unique blend of traditional and modern approaches to mental health [34]. Overall, this diverse range of participant demographics across studies highlights the multifaceted nature of mental health research and the potential of AI to address the needs of various populations. The findings underscore the importance of tailoring AI applications to specific demographic groups to enhance their effectiveness and acceptability in mental health interventions.

Acceptancy of AI in mental health

In reviewing the acceptability of AI in various mental health and healthcare contexts, studies demonstrate a general positive trend across different demographics, applications, and intervention types. Inkster et al. (2018) show that the Wysa app, aimed at depression support, demonstrated high engagement and effectiveness, emphasizing the potential acceptability of AI among users with self-reported depressive symptoms [10] (Table 2).

In this view AI tools are especially beneficial in student populations, where they effectively address mental health challenges, particularly in sports education contexts. Moreover, advanced AI mechanisms, like attention-based LSTM models, have shown promise in improving the accuracy of emotion recognition, an essential component in mental health interventions [27].

In higher education, the feasibility of AI-driven mental health plans for college students has been confirmed with favorable results, underscoring its practicality in academic environments [13].

The global scale of AI adoption is notable, with platforms reaching millions of users across 111 languages, underscoring a high level of acceptability and engagement [29]. Similarly, openly accessible databases like the UKBB, with extensive health data, signify the potential for AI integration in broader health-related applications [15]. Rathnayaka (2022) emphasizes that user-driven data input, such as voluntary mental health surveys, aligns well with AI's non-invasive monitoring applications [30].

In youth-focused applications, Tate (2020) notes that AI-driven interventions in mental health must address ethical concerns such as inequality and population bias [31]. Similarly, Guo et al. (2024) suggest that wearable devices provide an accepted, practical approach to monitoring stress and anxiety levels in real-time [23].

These findings collectively highlight the high acceptability and potential efficacy of AI-based mental health interventions across various settings, populations, and applications.

Application of AI in mental health

The studies highlight the extensive applications of AI across various domains in mental health and healthcare, with each study exploring unique facets of AI’s functionality and its impact on patient outcomes (Table 2).

The effectiveness of AI in educational settings is further highlighted by Lei et al. (2023), where AI was used to reduce anxiety and depression symptoms among students. In another innovative application, Xiao Li (2023) optimized LSTM outputs through AI attention mechanisms, which enhanced emotion recognition accuracy, a significant step for personalized mental health care [12, 27].

Beyond individual assessments, AI is also used in population-level mental health monitoring, as demonstrated in study by Liang (2022) who used AI for psychiatric assessments, adherence tracking, and CBT delivery [13].

AI applications extend beyond therapy and assessments; they are increasingly relevant in preventive interventions and public health efforts. AI's potential for predictive modeling is also recognized, with Pei (2022) applying AI to analyze physiological and digital signals from wearable devices to detect stress and anxiety, enhancing real-time monitoring and intervention capabilities [17].

In healthcare systems, AI facilitates patient engagement and clinical decision-making. Guo et al. (2024) used AI to analyze wearable data for stress detection, while Götzl et al. (2022) incorporated AI into clinical documentation and patient interaction systems to improve clinician efficiency and patient satisfaction. Furthermore, the application of AI in personalized mHealth apps allows adaptation to user needs and preferences, enhancing intervention outcomes [23, 28].

Collectively, these studies highlight the diverse functions of AI in mental health, encompassing real-time monitoring, preventive care, personalized interventions, and predictive modeling. A common thread throughout these applications is AI’s capacity to bridge gaps in mental health services, enhance patient engagement, and provide customized support for various populations.

Effectiveness of AI in mental health

The reviewed studies demonstrate the varied effectiveness of AI in improving mental health assessment, monitoring, and treatment, leading to notable advancements across different mental health conditions (Table 2).

In the domain of digital mental health interventions (DMHIs), Inkster et al. (2018) reported that high engagement with the Wysa app led to significant improvements in self-reported depression scores among users, with a moderate effect size [10].

AI’s capacity to improve early mental health recognition and intervention is also evident in studies by Lei et al. (2023) and Xiao Li (2023), where deep learning and attention mechanisms were used to significantly enhance emotion recognition and classification accuracy in mental health monitoring [12, 27]. Furthermore, studies by Liang (2022) emphasize the effectiveness of AI in psychiatric assessment, including real-time symptom tracking and personalized therapy delivery [13].

At the same time, Tornero-Costa et al. (2023) emphasized the potential of AI in facilitating large-scale mental health screenings and interventions, which could allow for the early identification of mental health issues and provide tailored support for various populations [1].

In summary, these studies confirm the multifaceted effectiveness of AI in mental health across preventive, diagnostic, and therapeutic stages, with applications that include emotion recognition and personalized mental health interventions. This broad applicability underscores AI’s potential to enhance access to mental health resources, improve treatment delivery, and facilitate large-scale mental health screening and monitoring initiatives.

Discussion

The integration of AI into mental health care has emerged as a focal point of research, reflecting a wide array of applications and methodologies. A systematic review encompassing 15 studies revealed varying degrees of methodological rigor and quality, offering a nuanced perspective on AI’s role in enhancing mental health interventions. Notably, studies that employed robust methodologies and high-quality assessment tools, such as the MMAT and the JBI checklist, provided more credible insights into the effectiveness of AI applications in this field.

This review captures the evolution and current landscape of AI applications in mental health care, particularly focusing on advancements from 2009 to 2024. For instance, the research conducted by Didarul Alam et al. (2021) utilized a rigorous empirical framework, achieving a high-quality score of 10. This study analyzed mobile health (mHealth) technologies during the COVID-19 pandemic and demonstrated a positive correlation between mHealth usage and mental well-being, highlighting the potential of AI-driven tools to enhance user engagement during crises [19]. Similarly, studies by Inkster et al. (2018) and Chin et al. (2023), both of which received quality ratings of 7, showcased the effectiveness of AI chatbots like Wysa in improving mental health outcomes through increased user engagement [10, 29].

Conversely, studies with moderate quality scores, such as those by Dadi (2021) and Rathnayaka (2022), while still informative, indicated areas needing methodological improvement. Dadi’s population modeling approach garnered a score of 6, suggesting that although it provided valuable insights into the relationship between AI and mental health, its methodological limitations may hinder the generalizability of its findings. Similarly, Rathnayaka’s study on a behavioral activation-based AI chatbot also received a score of 6, underscoring the necessity for more robust evaluation frameworks to assess the efficacy of such interventions [15, 30].

The findings from high-quality studies emphasize the critical importance of methodological rigor in AI-related mental health research. For example, Xiao Li (2023) employed advanced AI techniques, including Long Short-Term Memory (LSTM) networks, to enhance emotion recognition, achieving a quality score of 7 [12]. This illustrates that higher-quality studies not only yield more reliable data but also contribute to the advancement of AI methodologies in mental health applications.

Furthermore, the diversity of participant demographics across the studies reflects the complex nature of mental health challenges and the potential for AI to address these varied needs. Research targeting college students, elderly populations, and individuals from diverse cultural backgrounds demonstrates the adaptability of AI tools in meeting the specific requirements of different groups. This adaptability is essential for enhancing the acceptability and effectiveness of AI interventions, as evidenced by positive user feedback in studies like that of Götzl et al. (2022), which explored young people’s attitudes toward AI-informed mHealth applications [28].

Nevertheless, despite the promising applications of AI in mental health, several ethical considerations and methodological limitations warrant attention. The reviewed studies highlighted concerns regarding data privacy, algorithm transparency, and the necessity for stakeholder involvement in the development of AI technologies. As Tavory (2023) noted, establishing ethical frameworks is crucial for guiding the responsible deployment of AI in mental health care. AI is increasingly becoming an integral component of digital medicine, with the potential to significantly influence mental health research and practice. To fully leverage AI’s capabilities, it is essential for a diverse community of stakeholders—including scientists, clinicians, regulators, and patients—to engage in open communication and collaboration [5].

As AI techniques continue to evolve, there is potential to redefine mental illnesses more objectively than the current DSM-5 classification system. Such advancements could enable earlier identification of mental illnesses, even at prodromal stages, when interventions are likely to be more effective. Moreover, AI can facilitate the customization of prescribed treatments based on individual characteristics, leading to more personalized and effective mental health care. It is also noteworthy that consumers recognize several advantages in using medical chatbots, such as anonymity and quicker access to pertinent information. Previous research indicates that consumers are often just as willing to share emotional and personal information with a chatbot as they would with a human friend. Studies suggest that interactions with chatbots and humans yield comparable levels of perceived understanding, disclosure closeness, and cognitive reappraisal, indicating that individuals engage psychologically with chatbots similarly to their interactions with people. This reinforces the potential of chatbots to provide effective support in mental health contexts. In this systematic review, we synthesized evidence regarding the effectiveness and user evaluation of AI-based conversational agents (CAs) in mental health care. Our findings suggest that these CAs can effectively alleviate psychological distress, with the most significant effects observed in studies utilizing generative AI, multimodal or voice-based CAs, and interventions delivered through mobile applications and instant messaging platforms. This aligns with the findings of Shimada (2023) and Alhuwaydi (2024), who highlight AI’s potential in addressing disparities in access to mental health services [35, 36].

These findings are further supported by Oladimeji et al. (2023), who emphasize AI’s role in the early detection and prevention of mental health issues [37]. However, it is crucial to recognize that these technologies could exacerbate existing inequalities if not implemented thoughtfully and equitably. Addressing potential disparities in access and ensuring that all populations benefit from AI advancements in mental health care is essential to prevent widening the gap in mental health services.

Conclusion

The paper provides a comprehensive review of the entire spectrum of AI in mental health, highlighting its positive contributions to the field. AI holds numerous promises for enhancing mental health care, and this paper explores various facets of its application. AI technologies are anticipated to introduce innovations to existing medical practices and future health care systems. Currently available AI-based health care technologies have demonstrated significant efficacy in accurately diagnosing and classifying patient conditions, as well as predicting disease trajectories by leveraging accumulated medical data.

Limitation

One significant limitation of this systematic review is the variability in methodological quality and reporting standards among the included studies. Although various quality assessment tools were employed, such as the Mixed Methods Appraisal Tool (MMAT) and the Newcastle–Ottawa Scale (NOS), discrepancies in study design, sample sizes, and data collection methods were observed. Many studies lacked transparency in reporting AI model features and data preprocessing techniques, which are crucial for ensuring reproducibility and reliability. This inconsistency may lead to biases in the interpretation of AI effectiveness in mental health applications and limit the generalizability of the findings across diverse populations.

Future research should prioritize the establishment of standardized methodologies and reporting guidelines for AI applications in mental health. This would enhance the robustness of the evidence base, facilitate comparative analyses, and ultimately improve the integration of AI technologies into clinical practice. Additionally, involving a broader range of stakeholders, including mental health professionals and patients, in the development and evaluation of AI tools may help address ethical concerns and ensure that these technologies meet the needs of diverse user groups.