Abstracts Presented at the 2025 Advancing Healthcare Innovation Summit: April 26, 2025 (Virtual) Open Access

Introduction: In the era of precision medicine and rapidly evolving healthcare technologies, digital pathology has emerged as a game-changing innovation in pathology and biobanking. By enabling high-resolution imaging, efficient data storage, and remote access, it has the potential to enhance diagnostic accuracy, streamline workflows, and unlock the hidden research value of biobanked specimens. This paradigm shift supports advance-ments in biomarker discovery and personalized therapies while fostering collaboration and innovation in healthcare. Methods: At the Medical University of Bialystok (MUB), Poland, the integration of digital pathology involved a carefully planned, multi-disciplinary approach. A comprehensive needs assessment identified gaps in technology, workflow, and training, guiding the acquisition of state-of-the-art whole-slide imaging scanners capable of high-resolution digitization of glass slides. Extensive training programs were implemented to prepare pathology and biobank staff for adopting digital workflows, ensuring competency in using the new technology. A secure digital pathology platform was established, fully integrated with routine pathology software to ensure compliant storage and efficient retrieval of digitized images. Telepathology capabil-ities were also developed, allowing for remote consultations and collaborative research initiatives. These efforts were underpinned by rigorous attention to data security, patient confidentiality, and ethical standards, ensuring compliance with national and international regulations. Results: The introduction of digital pathology significantly improved diagnostic precision in routine pathology, with faster turnaround times and streamlined remote consultations. Enhanced imaging and data storage capabilities enabled the Biobank to digitize and preserve high-value biospecimens sys-tematically. This transition has facilitated efficient data sharing and analysis, granting researchers access to a robust repository of digitized histopathological data. These advancements have expedited biomarker discovery and the identification of novel therapeutic targets. Furthermore, the incorporation of tele-pathology has strengthened collaboration across institutions and geographical boundaries, enhancing research outputs and patient care outcomes. Conclusion: The integration of digital pathology at MUB exemplifies a strategic and well-executed approach to leveraging advanced technology in precision medicine. By fostering collaboration, improving diagnostic capabil-ities, and optimizing the use of biobanked specimens, digital pathology has become a pivotal asset in achieving the institution’s mission to advance patient care, research, and education.

Introduction: Inter-patient variability in cancer treatment response poses a significant challenge in clinical oncology. Our research aims to enhance personalized medicine by offering clinicians a digital strategy that predicts treatment responsiveness based on a patient's inherited genetic profile. We test the hypothesis that the complexity of a patients inherited genetic landscape predisposes them to having therapeutic responsiveness or non-responsive phenotype. This study introduces an innovative approach to predicting individual patient responsive-ness to treatment that is protocol agnostic by analyzing genetic variations through a framework that mirrors the clinical decision-making process. Methods: Our methodology first identifies the most informative group of genetic variants across the patient population that allow disease outcome distinction and then refines these findings to recognize patient-specific genetic patterns that inform personalized clinical decisions. This layered approach reflects how clinicians combine their broad medical knowledge with individual patient observations when developing treatment plans. The study design ensures reliable selection of genetic variants across a cohort whilst avoiding misleading associations, leading to the characterization of the genetic landscapes that are genuinely linked to treatment response. Instance based assessment of each patient further clarifies which genetic variants best reveal the significant biological differences that influence their cancer-related processes and may be indicative of therapeutic options, thus providing meaningful insights that could guide clinician decisions. Results: We present results from a dual-stage genetic variant selection framework that implements an initial global feature reduction from 13,927 single nucleotide polymorphisms (SNP) to 106 influential markers through hierarchical feature selection. This global subset then undergoes instance-specific refinement via a neural network module, which employs weight management and feedback to dynamically include or exclude SNPs based on individual patient characteristics, facilitating personal-ized genetic variant set extraction. Conclusion: This experiment suggests that constructionist analysis of genetic information can provide valuable clinical insights, potentially providing further certainty around cancer treatment planning and improving patient outcomes.

Introduction: Biobanks play a crucial role in accelerating the development of precision medicine by facilitating biomarker discovery, advancing research on targeted therapies, and supporting disease prevention, prediction, monitoring, and treatment optimization. This presentation aims to highlight the significance of biobanks as essential structures in modern medical research. Despite the ongoing implementation of personalized medicine into clinical practice, numerous challenges still hinder the advancement of innovative therapeutic solutions. One of the key obstacles is large-scale validation, which requires extensive collections of prospectively acquired biospecimens and comprehensive datasets. The identification of single molecules in novel research areas strongly depends on the quality of primary biospecimens used for analysis. Multiple factors may influence molecular composition, making accurate measurements highly dependent on sample handling procedures. The standardization of bio-banking processes has been recognized as a critical necessity, addressed through international guidelines such as ISO 20387:2018, ISBER Best Practices, and IARC publications. These expert-driven standards ensure high-quality biological collections and associated data for research. Methods: To meet the demand for high-quality biospecimens and validated data, the Medical University of Bialystok (MUB) established an Oncological Biobank Collection in 2015 as part of the MOBIT Project. Standardization of sample collection, processing, and storage, along with legal and ethical considerations, led to the formal foundation of MUB Biobank. This facility is dedicated to providing high-quality, strictly monitored biospecimens and associated data for the advancement of personalized diagnostics, integrating genomics, transcriptomics, and metabol-omics. MUB Biobank adheres to stringent quality assurance (QA) and quality control (QC) measures, ensuring compliance with international biobanking standards. Long-term collaborations with inter-national partners, such as Indivumed GmbH, National Cancer Institute, and Pangea Biomed, have enabled MUB Biobank to join the Biobanking and BioMolecular Resources Research Infrastructure – European Research Infrastructure Consortium (BBMRI-ERIC) and the Polish Biobanking Network. Results: Currently, MUB Biobank supports multiple projects led by internal and external stakeholders. Its extensive collection exceeds 2,000.000 biospecimens, including liquid samples (whole blood, plasma, serum, urine supernatant, urine sediment) and tissue samples (fresh frozen tissue stored in the vapor phase of liquid nitrogen and FFPE blocks). Moreover, the digital pathology approach has been implemented to develop machine learning algorithms for automated image analysis, enabling more precise histo-pathological assessments, biomarker quantification, and diagnostic decision support in research and clinical practice. The implementation of standardized biobanking practices ensures biospecimen integrity, enhancing research reproducibility and enabling AI-driven applications in medical research. Conclusion: Biobanks are not only responsible for the collection, storage, and utilization of biospecimens but also for maintaining their scientific value through strict integration with clinical data. To support biospecimen research and precision medicine, biobanks must operate under high professional standards, ensuring adequate funding, training, and certification, such as ISO 20387:2018. The field of biobanking is evolving into an organized, multidisciplinary domain with a strong emphasis on quality management, which is essential for the successful implementation of translational science. As biobanks continue to play a crucial role in biotechnology, pharmaceuticals, and academic research, they should be recognized as fundamental resources for medical research, ensuring access to high-quality samples and associated data for basic, translational, clinical, and diagnostic studies.

Introduction: The global biobanking market and biosample-oriented research initiatives are vital for modern healthcare. However, traditional centralized models, faces significant challenges in biospecimens and data exchange, including data fragmentation, interoperability issues, security, ethical and regulatory barriers. AI-MetaBloQ is a transformative solution designed to address these challenges, offering a secure, transparent, and ethically compliant marketplace for biospecimen and related data transactions. This decentralized platform bridges the gap between diverse stakeholders with a Distributed Ledger Technology (DLT) Marketplace that ensures immutable records, trust-based interactions, and decentralized identity management, empowering patients with control over their data through the “Ledger of Me” concept. Meanwhile, its Reactive Artificial Intelligence (AI) Tool enables real-time biospecimen assessment and automated categ-orization, optimizing their selection for downstream applications. Compliant with global regulations like GDPR, AI-MetaBloQ showcase its role in transforming the biobanking ecosystem. Methods: This section outlines AI-MetaBloQ’s core functionalities and workflow roadmap, incorporating and describing functional and non-functional requirements per user type, data collection and regulatory requirements, incorporated methodologies, components and modules, overall architecture design, and deployment method. The development of AI-MetaBloQ followed a structured, data-driven, and user-centric approach, gathering users’ feedback at every step of the platform’s deployment. Its implementation is guided by four key principles: biospecimens data-driven sharing supported by DLT, tailored applications for end-users, interoperable customizable solutions for different organizations, regulatory compliance, and data auditability. Results: The platform was successfully developed with features enabling stakeholders to securely trade biospecimens and biomedical data, facilitating real-time transactions with transparency and data integrity. Each user type is enabled to interact with AI-MetaBloQ in ways that reflect their unique attributes and needs. Patients can track and control the use of their biospecimens and related data. Researchers can leverage patients’ and biosamples’ data integration features to enhance the accuracy and reliability of their studies. Hospitals can utilize secure data exchange and consent management tools to streamline their procedures for clinical trials and medical records dissemination. Biobanks can benefit from the platform's robust system for quality management and dissemination of biosamples and related data in a secure manner. The marketplace provides service exchange function-alities, fostering collaboration and promoting transactions through smart contracts and automated key actions. Conclusion: AI-MetaBloQ offers a transformative solution for the biobanking ecosystem, addressing key challenges in biospecimen and data exchange through secure, transparent, and compliant transactions. Customizable for diverse organizations, it streamlines operations, reduces costs, and improves interoperability, positioning itself as a key tool in advancing biomedical research.

Introduction: Despite extensive prevention efforts, nearly 2000 individuals die by suicide globally each day. Digital behavioral health (DBH) platforms provide a scalable means of collecting indicators of suicidal ideation (SI) outside clinical settings but converting these indicators to actionable risk intelligence remains challenging. Large language models (LLMs) are designed to understand the complexities of human language. Thus, they are well-positioned to enhance suicide prevention efforts by rapidly stratifying patient-entered text by SI severity, potentially enabling an asynchronous, near real-time safety net to aid clinicians and orchestrate lifesaving interventions. However, the effectiveness of LLMs in this context has not been systematically evaluated. We conducted a synthetic-data study to bridge this gap by comparing expert clinicians and LLMs in a suicide risk-stratification task. Methods: We developed and validated a corpus of 125 synthetic journal entries modeled after real-world DBH platform prompts. Entries were generated from a custom prompt with >1 trillion potential feature permutations and varied in SI severity, readability, length, and linguistic features. Five independent clinicians stratified entries into SI risk categories (no risk and low, moderate, or high risk). The same entries were risk-stratified using five tailored LLMs, plus an ensemble model. LLM results were evaluated against clinician consensus, providing a robust measure of ground truth. Ethical clearance was not required for our study, as all data were synthetic and experts participated in a professional capacity. Results: The ensemble LLM significantly out-performed chance-level exact agreement (30.38%) with clinical raters, achieving 65.60% agreement (χ² = 86.58) in its risk classifications. It also aligned with 92% of clinicians’ “do/do not intervene” decisions (Cohen’s Kappa = 0.84), demonstrating 94% sensitivity and 91% specificity. Precision-recall analyses revealed that 4^th generation LLMs offered an excellent balance between sensitivity and precision, with area under curve (AUC) values reaching 0.90. Time to decision analyses indicated that LLMs completed assessments significantly faster than clinicians, with 4^th generation LLMs reaching speeds of 0.35 seconds per assessment, nearly 60 times faster than the clinical average of 20.7 seconds. Cost comparisons revealed substantial efficiency gains: GPT-4o Mini completed 1 million assessments for $6, whereas the estimated cost of clinicians completing that workload was $100,000-$250,000. Conclusion: LLM-powered SI risk stratification demonstrated strong alignment with expert clinical judgment and offered significant gains in speed and cost-effectiveness. These findings support the integration of AI-assisted risk detection as a safety net in existing clinical workflows to enhance suicide prevention efforts, although continued validation and ethical considerations remain critical.

Introduction: Data in healthcare have been constantly evolving between the specializations and expertise emerging from novel fields of study and discoveries, allowing for great opportunities to utilize the multimodal data collected in the health sector, thereby actively advancing understanding of diseases, developing treatments, and enabling personalized medicine. However, the data can be complex and may contain sensitive data, posing a significant challenge in the storage, procession, and utilization of the data, especially concerning data sharing efforts. Research towards supporting frameworks that encourage and empower data sharing activities, specifically a Trusted Research Environment (TRE) and Federated Analytics Network (FAN), may enable researchers globally to securely share and access health data with acknowledgement of the barriers brought by ELSI, as well as administrative, security and privacy issues. Through this investigation, we aim to produce a potential evaluative maturation model for a sustainable FAN supported by TRE nodal entities. Methods: Through evaluating examples of maturity level models by established federated networks Beyond One Million Genomes, ELIXIR European and Federated European Genome-Phenome Archive, which allowed entities within their network to evaluate respectively their organizations within an international network, a foundation for the evaluative framework could be drafted. Moreover, supporting documents targeted towards TREs, including Health Data Research UK’s Capability Model, as well as the Standard Architecture for TRE (SATRE) document, were utilized to address federated application of such TREs, and revealed domains that need elucidation for successful application. Results: Comparative analysis of the maturity models’ domains, subdomains and action descriptors from the federated networks denoted eleven overarching domains pertained to achieving a mature node within its network. These domains were further consolidated as governance and policies, operation and performance, data management, physical/technical infrastructure, outreach and communication, as well as clinical/research infrastructure, which are all important to the functioning and sustainability of a federated network. Concerning to TRE application, descriptors found in the Capability Model and SATRE were aligned to the aforementioned domains, providing a preliminary evaluative framework to assess the maturity of a potential FAN for TREs. Remnant domains with limited alignment of descriptors were implicated as gaps that require further elucidation for their assessment on such a proof of concept, specifically on encouraging collaborations, and the clinical and research infrastructure required to support such a framework. Conclusion: A novel preliminary model recommendation has been made through analysis of existing documentation for the application to federated analytics supporting network of TREs, with acknowledgment of the challenges it may face including legal, ethical, and operational aspects.

Introduction: Innovative approaches are essential in pediatric cancer care to enhance prognostic prediction and enable personalized stratification. However, there is a lack of studies assessing the performance, composition, and limitations of contemporary prognostic models. This study aimed to compare the accuracy of traditional and advanced prognostic models. Methods: A systematic search for this systematic review and meta-analysis (CRTN 42022370251) was conducted in PubMed, Embase, Scopus, and the Cochrane Library databases on June 28, 2024. Studies evaluating the accuracy of prognostic markers or models in pediatric hematological malignancies, central nervous system (CNS) tumors, or non-CNS solid tumors (NCNSST) were included. Prognostic models were categorized into three groups: (category 1) traditionally used clinical parameters, (category 2) genomic-transcriptomic data, and (category 3) models harnessing artificial intelligence (AI). The primary outcomes assessed were the area under the receiver operating characteristic curve (AUC) with a 95% CI for various overall survival (OS) intervals. Study selection and data extraction were conducted independently by two groups, utilizing published data and publicly available databases. Results: Out of 12,982 studies, 358 were included in the meta-analysis, with heavily limited findings on AI-based approaches. The majority of data focused on NCNSST at 5-year OS, showing a statistically significant difference between category 1 (AUC: 0.75, CI: 0.72-0.79) and category 2 (AUC: 0.85, CI: 0.82-0.88) (p < 0.001), but not between category 2 and 3 (AUC: 0.82, CI: 0.77-0.88) (p = 0.2834). What is more, category 3 models overperformed both category 1 and 2 models for non-time-dependent OS (AUC: 0.85; p = 0.014 and 0.035). Studies using internal validation demonstrated significantly better performance than those using external validation, emphasizing the high risk of bias associated with internal validation. Conclusion: Implementing category 2 and 3 models in clinical practice is recommended, particularly for NCNSST prognostication to support risk stratification. Although the current literature on AI approaches remains heavily lacking, as being reliant on a small number of high-quality biobank datasets. Expanding pediatric oncology biobanks and data repositories, along with harmonizing existing resources, will best provide the necessary quality-assured training and validation datasets needed to advance prognostic models in this rare disease field.

Introduction: Nightmares are a common and distress-ing symptom of post-traumatic stress disorder (PTSD) in children and adolescents. This review examines both pharmacological and non-pharmacological approaches to treating nightmares in pediatric PTSD populations. Pharmacological drugs include prazosin, cypro-heptadine, and others, though evidence for their efficacy and safety in children is currently insufficient. Non-pharmacological treatments with stronger evidence include cognitive-behavioral therapy (CBT), imagery rehearsal therapy (IRT), and eye movement desensitization and reprocessing (EMDR). These psychotherapeutic approaches show promise in reducing nightmare frequency and distress. More randomized controlled trials are necessary, particularly for drug treatments. A multimodal approach combining psychotherapy and judicious use of medications may be most effective for treating nightmares in pediatric PTSD. Methods: We conducted a systematic literature search using PubMed, Google Scholar, PsycINFO, Medline, and PsycArticles. Each search used combinations of these key terms: sleep, nightmares, insomnia, trauma, traumatic events, PTSD, and child abuse. Results: We found mostly case reports on managing nightmares with medications, but few controlled trials have been published. A good amount of research including some RCT is already published about psychotherapeutic approach to treat pediatric nightmares. Conclusion: Research on treatments for PTSD-related nightmares in children is limited. The research that exists is based on case reports, clinical experience, and extrapolations from the adult study findings to help and guide the treatment for children. CBT has been the first choice of treatment of PTSD in children and adolescents. Medications are rarely first line treatment for pediatric sleep disorders. More training in sleep medicine for primary care physicians along with substantial amounts of research including RCT are needed to explore empirical treatment guidelines for pediatric nightmare due to PTSD.

Introduction: Short-form media platforms, such as TikTok, have grown in popularity among teenagers and young adults. The average duration of this media is 27 seconds and provides users with quick, engaging content. However, the increasing use of short-form media has raised concerns about potential cognitive effects, particularly regarding attention and memory. One such concern is the release of dopamine. When users scroll through content, it leads to quick and easy dopamine release, causing addiction and affecting an individual's attention span for less engaging tasks. Methods: This study explores the effects of short-form media exposure on memory and focus on people of ages 15-22. The study predicts that longer durations of short-form media exposure decrease attention spans and short-term memory. To test the hypothesis, participants completed a series of cognitive tasks designed to assess both memory and focus. Afterward, they watched short-form media content aligned with their interests for 5, 10, and 15 minutes. Between each interval, participants retook the cognitive tests to measure any changes in performance and observe any patterns in data changes. Results: Overall, results showed that short-term memory improved as the watch duration increased. Across all the memory tests conducted, scores increased by an average of 49.1%, with the highest score increase measuring 114%. This improvement could be due to the fast-paced, engaging nature of the tests, helping the already highly stimulated participants perform better in their continued state of stimulation. However, focus among participants decreased significantly as they continued to be exposed to this form of media. Across all focus tests conducted, there was an average score drop of 50%. These findings indicate that short-form media may enhance certain fast-paced types of memory; however, it also hinders sustained attention and focus on a task. Conclusion: This study offers insights into the short-term cognitive effects of short-form media, showing that while it may improve memory in certain tasks, it negatively impacts attention and focus. Although the study examines short-term effects, more research is needed to observe the long-term effects of repeated consumption of said media. If trends of decreasing ability to concentrate, retain information, and focus on complex tasks continue, they may impair cognitive functions essential for daily life. Understanding the long-term effects of short-form media consumption on brain development is necessary to see the potential harm as this media continues to shape future generations.

Introduction: In today’s society, children are often exposed to many different types of audio and sound stimuli such as nature sounds and music. These sounds are used in many ways, from calming relaxation techniques to creating engaging environ-ments in video games. Studies reveal that 75% of individuals experience stress and anxiety, which in recent years, has become increasingly common in children and adolescents. As relaxation plays a vital role in mental health, finding effective methods to assist people is essential. This study aims to measure alpha and beta brainwave activity to investigate the impact of nature sounds and video game music on the relaxation levels of adolescents ages 9-15. Methods: Three participants (control and independent variables) were included and randomly selected and were subjected to two trials of the tested variables. The procedure involved the use of an EEG cap for approximately 10 min, alternating from nature music (5 min) to video game music (5 min), then switching to the opposite sound pattern after a 10-min break and repeating the listening session for the same amount of time. Throughout the experiment, the EEG cap measured brain activity (primarily alpha waves) before and after implementing the independent variables to assess the levels of relaxation. Results: Before listening, the EEG showed moderate alpha waves with some beta activity, indicating mild alertness. After exposure to natural sounds, all participants’ alpha waves stabilized and decreased, while beta brain wave activity decreased, suggesting deeper relaxation. In contrast, after listening to video game music, two of the participants’ alpha waves showed some reduction compared to normal while beta waves remained active, and all brain waves were generally more engaged and excited, indicating sustained mental engagement. For the third participant (age 13), alpha waves were highly reduced and beta waves were stimulated, showing higher alertness than the other participants. All participants reported feeling more relaxed with nature sounds but more alert/excited with video game music. Conclusion: This study reveals the profound influence of sound on relaxation. These findings show that natural sounds can significantly enhance calmness in children which can open new possibilities for using sound-based interventions in schools, therapy, and mental well-being. Given the limited sample size, further research is needed to explore individual differences and long-term effects. By advancing this understanding, we can leverage sound to create new devices that promote relaxation and create a balanced environment.

Introduction: In today’s world, many food labels contain ingredients that are difficult to understand, making it challenging to assess the impact of certain foods on health. This study aims to evaluate and compare the ability of various artificial intelligence (AI) models to assess patient profiles and provide nutritional advice based on an individual’s medical history and the nutritional value in foods. Methods: Twenty different theoretical patient profiles were randomly generated, varying in age, gender, weight, height, and medical history, to assess the impact of certain foods on individual health. The profiles were inputted into ChatGPT, Claude, Gemini, and DeepSeek alongside a list of 10 healthy and unhealthy food items. The models were tasked with identifying the health impacts of each food, considering each patient’s medical history. Additionally, the models categorized each food as either green (best impact), yellow (moderate impact), or red (worst impact). Each model's ability to identify risky foods was compared to the expected output. Results: All four models found unhealthy and sugary foods as detrimental and healthy foods as beneficial. However, each model provided different assessments of the health impacts associated with each food. ChatGPT and Gemini provided clear responses, detailing the immediate effects of food consumption on the patient’s specific profile and citing nutritional facts that exacerbated their medical conditions. In contrast, Claude focused more on the long-term effects and was not able to provide specific immediate impacts based on patient profiles. Lastly, although DeepSeek’s responses were specific and straightforward, they lacked consistency and failed to justify the long-term and short-term impacts on patient health. Overall ChatGPT had an accuracy of 72.5%, and Gemini, Claude and DeepSeek had accuracy rates of 63.5%, 67% and 63% respectively. Conclusion: ChatGPT proved to be the most successful model in assessing food health risks based on medical history, out of the four tested, due to accuracy rate and overall response structure. The AI models demonstrated their significant potential in identifying foods that may pose health risks to certain individuals based on health and lifestyle data. When integrated with medical dietary advice, AI can be an influential tool in assisting patients in maintaining healthy diets that adhere to their personal medical history. Furthermore, for individuals with conditions that are impacted by diet, such as diabetes or obesity, AI can aid both healthcare providers and patients in managing these conditions by understanding the specific nutritional impacts of certain foods.

Introduction: Mental health disorders are often misinterpreted as temporary emotional phases, resulting in delayed diagnoses and insufficient support. Artificial intelligence (AI) has emerged as a promising tool to improve diagnostic accuracy of the onset of mental disorders and facilitate early intervention. This study evaluates the effectiveness of AI-driven tools, including Large Language Model (LLM) chatbots, speech analysis, and behavior tracking in identifying mental health disorders. The goal is to determine whether AI LLM models can improve diagnostic precision and enable the earlier identification of at-risk individuals. Methods: This study utilized the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition Text Revision (DSM-5-TR) as the primary reference for diagnostic criteria. Due to restricted access to public clinical records, 200 patient profiles were randomly generated to simulate real-world psychiatric case data. These profiles incorporated key information such as name, age, sex, health condition, lifestyle, and symptoms. Multiple AI LLM models, including ChatGPT, DeepSeek, Gemini, and Claude, were used to evaluate diagnostic performance. Each model’s ability to identify mental health disorders based on behavioral patterns and emotional distress was assessed. Diagnostic accuracy was measured by comparing AI-generated assess-ments to established DSM-5-TR criteria. Results: The four AI models were evaluated using a five-tier rating system: success (4/4), substantial success (3/4), moderate success (2/4), partial success (1/4), or failure (0/4). A rating of “substantial success” or “success” was classified as a correct diagnosis. Based on this criterion, the models achieved a combined success rate of 89.5% across 200 simulated profiles, with ChatGPT outperforming the others in terms of accuracy, speed, and organization. AI models often accurately identified the specific DSM-5-TR disorder, but in cases where the precise diagnosis was missed, AI still recognized the broader symptom category. Conclusion: Among the four AI models we tested, we found that ChatGPT delivered the best results due to its accuracy, speed, and well-organized responses. AI-based tools demonstrate significant potential in improving the accuracy and timeliness of mental health disorder diagnoses. Integrating these technologies into clinical practice could aid healthcare providers in enhancing early intervention strategies and improving patient outcomes. Future research should focus on enhancing AI models for clinical integration, particularly by improving accuracy through the consideration of additional factors such as comorbid conditions.

Introduction: Emotional regulation is crucial for healthy child development, yet many children struggle with managing their emotions effectively. Research indicates that up to 30% of neurotypical children experience difficulty with emotional regulation, affecting their mental health, academic performance, and social relationships. For children with autism spectrum disorder (ASD), this challenge is even greater. Studies show that nearly 40% of children with ASD have difficulty identifying and regulating emotions. Our research aims to develop an application using brain-computer interface (BCI) and artificial intelligence technology to detect a child’s emotions and provide personalized interventions to achieve their desired emotional balance. Methods: The Moodio application platform integrates BCI technology and non-invasive electroencephalography (EEG) sensors to analyze alpha, beta, and theta brain waves, providing real-time emotional support through personalized interventions. Two neurotypical participants, ages 12 and 9, were selected for initial testing. Each participant completed two trials while wearing an EEG cap, beginning with a BCI headset recording brainwave activity during emotion-triggering activities to establish baseline emotional responses. After a 5-min rest period, participants engaged with Moodio’s interventions, including parental voice support, adaptive music therapy, and guided breathing exercises, while EEG readings were continuously monitored. Brainwave data was wirelessly transmitted via Bluetooth to a machine-learning model, which analyzed the patterns, identified user emotions, and triggered recommended personalized interventions. Data collection focused on fluctuations in alpha wave activity, which correlate with relaxation and emotional regulation. Future analysis will compare neurotypical and ASD brainwave patterns to identify differences in emotional processing and further refine Moodio’s personalized intervention methods. Results: Moodio influenced brainwave patterns associated with emotional regulation, as recorded by the Unicorn BCI. EEG electrodes placed across multiple brain regions captured activity related to focus, auditory processing, and sensory integration. Pre-intervention data showed decreased alpha waves and increased beta activity, indicating heightened anxiety. During the intervention, participants engaged in calming exercises, music, parental voice recordings, and journaling. Post-intervention EEG revealed increased alpha waves (relaxation), normalized beta waves (improved focus and regulation), and decreased theta waves (reduced anxiety). These results highlight Moodio’s effectiveness in promoting emotional stability. Conclusion: Moodio effectively detected emotions and provided personalized interventions, proving itself a powerful tool for emotional support. Future development will include incorporating advanced machine learning techniques to enable more personalized emotional responses, exploring more calming activities to offer a variety of options, adapting to other neurodiverse conditions, and expanding usability to all age groups. Additionally, integrating Moodio with wearable headphone would broaden its accessibility and practical use in everyday life.