Artificial intelligence (AI) is reshaping healthcare, but without safeguards it can entrench disparities and erode trust. We advance an Equity-by-Design governance approach that embeds fairness audits, ethical oversight, and transparency across the AI lifecycle. Drawing on literature and regulation (e.g., EU AI Act, FDA GMLP) and two case studies, we show how continuous auditing, lifecycle checkpoints, institutional review, and post-deployment monitoring surface and mitigate demographic bias. We outline a practical architecture (metrics, dashboards, documentation) aligned with clinical workflows and regulatory duties, and conclude with policy and research actions to operationalize equity, accountability, and trustworthy deployment in diverse healthcare settings.
Air Quality Index (AQI) has been a viral issue since the Coronavirus Disease 2019 (COVID-19) outbreak worldwide. Much research has been published on the topic of COVID-19 and its correlation with the air quality parameters such as Ozone (O3), Nitrogen Dioxide (NO2), Particulate Matter (PM) with microns size of 2.5, 10, and other parameters related to the measurement of the air quality. This paper research proposed and continued the issue of the correlation between the air quality parameters with the pandemic of COVID-19 and compared it to the improvement of the traffic location in Jakarta. We used several air quality parameters to measure the AQI in Jakarta during COVID-19. The air quality parameters are NO2, SO2, CO, O3, and PM10. The results showed a significant improvement in the air quality parameters when compared pre and during COVID-19. Some locations of research that are Kebon Jeruk and Kelapa Gading, have the better air quality since the pandemic COVID-19 meanwhile, other research locations that are Lubang Buaya, Jagakarsa, and Thamrin, did not have a significant impact on the COVID-19.
Heart disease is one of the diseases known to have highest mortality rate, with estimated that one in third of death per year is caused by heart disease. Heart disease comes randomly and can’t be predicted on whether a patient has a heart attack. Potential of heart attack can be predicted with analysing the health situation of the patient. This research is used to predict the heart disease potential in a patient. This research uses statistical methods, namely Logistic Regression and uses deep learning, namely Neural Network. Comparison between the two models is expected to produce the best model for predicting heart disease. In this experiment the neural network has a very high accuracy reaching from 80
Recent advances in neuroimaging, genomics, and other technology driven data acquisition methods have greatly increased the complexity and volume of medical data. Traditional machine learning (ML) approaches are becoming increasingly difficult to apply in this context, particularly in neuroscience, where datasets are often high-dimensional but contain a limited number of samples because of the difficulty of collecting data from human participants. Although ML techniques are powerful tools for analyzing large datasets, they typically require substantial training sets containing balanced data and accurate labels. In real-world medical research, such data is rather rare. Consequently, small sample sizes can introduce bias in model performance estimates, thereby limiting the feasibility of predictive modeling. Nevertheless, such datasets are essential for identifying potential biomarkers and for conducting pilot or feasibility studies within the framework of personalized medicine. However, the limited sample size can lead to biased machine learning performance estimates, which makes it impossible to apply ML methods to predictive modeling. Therefore, artificial intelligence-based data mining tools are being developed to process large volumes of data and explore hidden features and correlations. This narrative review provides an overview of ML strategies tailored to neurological datasets with limited sample sizes, to better understand recent trends in this area and identify opportunities for future research. Particular attention is given to dimensionality reduction in complex data with few instances, as well as the integration of data mining and statistical learning techniques to improve the analysis and interpretation of small-scale but information-rich datasets.
The detection of somatic mutations and Copy number alterations (CNAs) in cancer cells is crucial for diagnosis and careful observation; however, existing traditional methods are both insufficient and inefficient. Examining the latest advances in Machine learning (ML) techniques, such as the Random forest algorithm, and ensemble models such as SomaticSeq, which turns out to have good mutation detection accuracy and efficiency. This method also proves capable of handling a variety of sample purity and sequencing strategies. Therefore, this method can offer good results and efficiency levels when compared to conventional approaches. Although there are still challenges, such as the need for a capable training dataset and high computational requirements, this ML model promises to make significant progress in cancer diagnosis, early detection, and personalized treatment. This review paper aims to review the ML methods that involve the detection of somatic mutations and (CNA). The results of the review showed that the ML method was promising in both implications, as proven by the minimum accuracy being above 70
Physician disengagement from telemedicine remains a critical yet underexplored challenge to the long-term viability of digital health systems. While adoption has been widely promoted, the reasons clinicians reduce or discontinue its use after initial uptake are less understood. This study explores the sociotechnological factors influencing Filipino physicians’ decisions to scale back telemedicine use in post-adoption settings. Using a qualitative design, in-depth interviews were conducted with physicians across various specialties and practice locations. Thematic analysis revealed that discontinuance is shaped by a combination of diagnostic limitations, poor system usability, lack of interoperability, medico-legal concerns, emotional exhaustion, and financial burden. Physicians reported that virtual consultations often hindered accurate assessments, strained communication, and disrupted professional boundaries. This research offers a novel contribution by reframing telemedicine disengagement as a professional and clinical recalibration rather than a mere reaction to technical issues. It extends discontinuance models by integrating clinical judgment, ethical accountability, and work-life balance considerations. The study provides practical insights for designing telemedicine systems that are not only technologically reliable but also aligned with physicians’ workflow realities and care standards. These findings are particularly relevant for health systems in low- and middle-income countries seeking sustainable digital health integration.
Mobile health applications are increasingly used in Indonesia, yet little is known about how users evaluate and experience these tools. This study aims to identify user needs, challenges, and expectations from popular Indonesian mHealth applications by analysing publicly available app reviews. We applied a mixed-methods approach, combining automated preprocessing with qualitative thematic coding. The dataset included 100,000+ user reviews from leading mHealth apps, with analysis supported by inter-rater reliability checks and AI-assisted keyword identification. The findings reveal recurring issues of usability, reliability, and trust, alongside user expectations for affordable services, responsive customer support, and integration with the national healthcare system. Based on these insights, we propose prioritised design and policy recommendations, considering implementation complexity, success metrics, and cultural context. This study contributes to the design of inclusive, trustworthy mHealth platforms in low- and middle-income countries, and informs both academic research and practitioner development.
Sexually transmitted diseases (STDs) and other sexual disorders continue to pose significant global health challenges, with cases steadily increasing, including in Indonesia. In Bandung, the rising incidence and low public awareness underscore the need for facilities that combine medical rehabilitation with comprehensive sexual education. Current rehabilitation centers often lack psychological support and user oriented spatial design, while stigma and shame hinder individuals from seeking help. This study proposes an interior design concept for a Sexual Rehabilitation and Education Center in Bandung based on the Positive Environment Design approach. The design seeks to reduce stigma and promote emotional healing through human-centered strategies. Its conceptual framework Revival, Encourage, Sanctuary, and Privacy is manifested through zoning, materials, lighting, and circulation. Using a qualitative method involving literature reviews, field observations, interviews, and fishbone diagram analysis, the research identifies user needs that inform the spatial arrangement of medical, educational, and support areas. The resulting prototype demonstrates how interior design can integrate functionality and psychological comfort, serving as a model for future sensitive healthcare environments in Indonesia.
Telemedicine has emerged as a transformative approach to delivering healthcare services, particularly during the COVID-19 pandemic, by enabling remote access to medical consultations and monitoring. However, many healthcare workers remain inadequately prepared for this digital shift. This study aimed to evaluate healthcare workers’ readiness for telemedicine implementation across four key dimensions: core readiness, e-learning readiness, clinical readiness, and technology readiness. A cross-sectional descriptive study was conducted using convenience sampling. Data were collected through an online questionnaire adapted from the Core, Clinical, and E-Learning Readiness Assessment and the Technology Readiness Assessment tools. A total of 60 healthcare workers participated voluntarily. Descriptive statistics were used to analyze readiness levels in each domain. Among respondents, 30
Traditional manual review methods are plagued by issues such as being time-consuming, labor-intensive, and prone to inconsistent standards. To overcome these limitations, an intelligent medical record review assistant was developed in conjunction with the cardiac surgery department of a tertiary Grade A hospital, leveraging the advanced text comprehension and reasoning of large language models (LLMs). This assistant, built on the Dify platform, employs the divide-and-conquer strategy that structures the review into a three-stage visual workflow: text structuring, multi-dimensional parallel analysis, and results aggregation. During the parallel analysis stage, prompt engineering techniques, including Expert Mimicry and Chain-of-Thought, direct multiple LLM nodes to independently assess records across four key dimensions: medical terminology, content completeness, diagnostic rationale, and the appropriateness of the treatment plan. Experimental results demonstrate that LLMs outperform both traditional machine learning and deep learning methods in medical record reviewing, and that the divide-and-conquer strategy yields better outcomes than using a single LLM.
The detection of focus and unfocus is critical in education, healthcare, and human–computer interaction. Electroencephalography (EEG) offers a non-invasive and real-time approach to assessing brain activity related to attention, yet challenges persist due to individual variability, non-stationary signals, noise, and limited labeled datasets. This paper reviews current trends in EEG-based focus detection, with an emphasis on deep learning (e.g., CNN, LSTM), meta-learning (MAML), and self-supervised learning (SSL). Publication analysis shows a significant rise in interest, with deep learning studies increasing from 10 in 2015 to more than 3,000 in 2025, while SSL and meta-learning have rapidly emerged since 2020. Our contributions are threefold: (1) identification of major challenges in EEG-based attention detection, including data scarcity and adaptability across subjects; (2) comparative evaluation of learning strategies in terms of data requirements, adaptability, and computational complexity; and (3) discussion of implementation pathways and application areas spanning brain–computer interfaces, neurofeedback, education, mental health, and autonomous systems. This review highlights promising methodologies for improving accuracy, generalizability, and efficiency, underscoring the potential of adaptive AI-driven EEG systems to advance both research and real-world applications.
The issue of preserving sensitive information in medical datasets while maintaining their utility is a significant concern in the implementation of Differential Privacy (DP). This paper presents a framework for allocating privacy budgets, designed to optimize the total budget for medical datasets and thereby enhance data utility. Previous strategies for allocating privacy budgets have primarily relied on fixed mathematical rules, and excessive or insufficient noise addition can impact data utility. Therefore, we propose a Genetic Algorithm (GA)-based framework that generates a privacy budget sequence through selection, crossover, and mutation operations to arrive at an attribute-wise optimal privacy budget. After that, the same individual optimal budget is utilized for each record in publishing, ensuring individual privacy guarantees. Experimental findings on two medical datasets reveal enhanced data utility when compared to heuristic budget allocation methods. This framework presents a straightforward and efficacious strategy for allocating privacy budgets within the context of privacy-preserving medical data publication. The source code used in this study is publicly available at https://github.com/Wayne-on-the-road/OPBA-MDP .
In the current landscape of economic uncertainty and escalating healthcare costs, health insurance plays a pivotal role as a vital financial tool. The projected surge in healthcare expenditures in Indonesia, as outlined by the Mercer Marsh Benefits Trend Health 2023 report, underscores the urgency for robust health insurance solutions to mitigate the financial risks associated with unpredictable health events. This paper explores the evolving dynamics of health insurance in Indonesia against the backdrop of rising healthcare costs and the growing awareness among the populace regarding the significance of financial protection. We also delve into the burgeoning field of Industry 4.0 and its implications for health insurance claim processing, emphasizing the role of Optical Character Recognition (OCR) technology in automating and streamlining claim administration processes. The study underscores the criticality of optimizing service quality within health insurance companies and delineates the claim decision process, elucidating the pivotal role of technological advancements such as OCR in enhancing efficiency and accuracy. Furthermore, we offer insights into the challenges and considerations associated with OCR implementation and present a comparative analysis of claim decision processes with and without OCR. Overall, the research underscores the importance of leveraging technology and expert human oversight to navigate the complexities of health insurance in Indonesia and ensure swift, equitable, and cost-effective claim resolutions.
Depression presents a significant public health challenge, with recent advancements highlighting the need to investigate potential neuroimaging biomarkers underlying its complex neurobiological effects. This paper introduces Diffusion with Transformers for Depression (DiT-Dep), an innovative framework that utilizes DiT for effective depression detection and biomarkers identification. By modeling neuroimaging data as graphs, DiT-Dep optimizes the classification performance through a dual-objective training strategy. Moreover, a novel entropy-based attention refinement mechanism is introduced to enhance the model’s ability to learn discriminative features, coupled with perturbation-based post hoc explanation methods that clarify the relationships between functional brain networks and depression. Evaluations across multiple datasets reveal that DiT-Dep significantly outperforms leading baselines, achieving superior detection performance while also providing meaningful insights into the replicable and verifiable neuroimaging biomarkers associated with depression, thereby underscoring the promise of using advanced AI methodologies for scientific research in psychiatry. The code is available at: https://github.com/RosalindFok/DiT-Dep.git .
Diabetes constitutes an escalating public health issue, with Indonesia positioned sixth worldwide in prevalence, recording 19.5 million cases in 2021. Comprehending public perception around diabetes on social media can guide health communication tactics and policy formulation. This study evaluates two machine learning algorithms—Support Vector Machine (SVM) and K-Nearest Neighbors (KNN)—for the classification of diabetes-related emotions on X (previously Twitter). We gathered 12,847 Indonesian tweets regarding diabetes from January to March 2024 and utilized the CRISP-DM framework for methodical data processing. Following preprocessing (cleaning, normalization, tokenization, stopword elimination, and stemming), we employed TF-IDF vectorization and lexicon-based sentiment classification. The dataset was divided in a 70:30 ratio for training and testing purposes. SVM substantially surpassed KNN, attaining an accuracy of 81.96
Artificial Intelligence in the healthcare sector has been highlighted as a key technology to tackle the major challenges facing healthcare systems today. ChatGPT, a large language model developed by OpenAI, shows promise as a support tool for healthcare workers. This study investigates the adoption and usage of ChatGPT among medical doctors in the Philippines by utilizing an integrated theoretical framework combining Task-Technology Fit (TTF) and Unified Theory of Acceptance and Use of Technology (UTAUT). A total of 279 licensed medical doctors in the Philippines participated in an online survey through convenience sampling. Structural equation modeling was used to analyze the relationships among constructs. Performance Expectancy and Social Influence emerged as positive significant factors of Behavioral Intention to Use. Furthermore, Task Technology Fit and Behavioral Intention to Use significantly influenced Actual Use of ChatGPT by medical doctors. The findings emphasize the importance of understanding the influencing factors in usage intention and the task-technology alignment of AI tools in driving adoption. The study provides empirical insights into exploring the potential role of ChatGPT in healthcare and discusses theoretical and practical implications.
Deep learning and large language models (LLMs) are being investigated to detect depression from social media data. However, traditional deep learning methods such as Long Short-Term Memory (LSTM) are limited by their reliance on feature engineering, resulting in time consuming and hard to capture complex patterns arising from media data. To address this issue, we propose a novel deep-learning approach that is based on Bidirectional Encoder Representations from Transformers (BERT) framework with Transformer Regression (called BERT-TR) for depression severity prediction from media data. In addition, we employ different techniques such as under sampling, oversampling and weighted loss function to ensure robust and generalizable predictions for depression. The data from X, derived from clinical assessments aligned with the fifth edition of the (DSM-5-TR) criteria, is used for investigations. We theoretically and empirically show that BERT-TR can predict depression severity from media data effectively and efficiently.
Releasing electronic health records (EHRs) typically requires trialing multiple, complementary privacy models—k-anonymity, l-diversity, and t-closeness—before publication, with model and parameter choices selected to fit the release context by balancing each model’s privacy protection level against task-specific data utility. However, most methods optimize these models in isolation, limiting effectiveness and efficiency. We formalize their joint execution as multi-task optimization (MTO) over a shared anonymization plan spanning per-attribute generalization and record suppression. Further, we present multi – task anonymization by differential evolution (MTADE), a differential evolution (DE) framework that coevolves three privacy model related populations with distributed evaluation and an elite-migration/weak-replacement knowledge transfer policy, enabling cross-task reuse while curbing negative transfer. Across 16 healthcare datasets and six DE backbones, MTADE attains higher utility under equal privacy thresholds and converges faster than per-model single-task optimizers, yielding robust anonymization plans. This formulation and algorithm provide a principled route to simultaneously satisfying k-anonymity, l-diversity, and t-closeness for EHR release within a unified optimization pipeline, avoiding fragmented per-model tuning.
Electroencephalography (EEG)-based Brain – Computer Interface (BCI) systems enable direct communication between the brain and external devices, facilitating neurorehabilitation. Motor imagery (MI) classification and event-related potential (ERP) detection are two critical paradigms for developing efficient EEG-based BCIs. While deep learning enhances decoding accuracy, centralized training poses significant risks to user privacy, data ownership, and regulatory compliance. Especially the following three challenges remain unsolved in the existing research work: (1) lack of mechanisms for privacy-preserving feature extraction, (2) poor handling of inter-subject heterogeneity, and (3) minimal evaluation of privacy risks alongside model performance. To address these challenges, we propose FedDeepAutoCloAk a novel Federated learning framework that integrates local Deep Autoencoder-based unsupervised feature extraction and KMeans Cluster Optimization with Adaptive Knowledge for improving MI and ERP classification while minimizing information loss and strengthening privacy protection for stroke patient data. In this framework raw EEG data stay local, with cluster centroids homomorphically encrypted before secure server-side aggregation, ensuring confidentiality. The framework was evaluated on two post-stroke publicly available EEG datasets, achieved better classification performance using three deep learning models. This work provides a scalable, secure, and personalized solution for decentralized EEG-based BCIs, advancing both technical robustness and ethical integrity in neurotechnology.
Electroencephalogram (EEG) signal analysis plays a significant role in recognizing brain function and supporting the diagnosis of Epilepsy. Existing graph approaches are binary or use endpoint-only (slope/correlation) weights that do not capture the interior fluctuations and trend departures, which limit robustness for epilepsy detection. Many rely heavily on preprocessing and lack shift/scale robustness, which compromises generalization and reliability across subjects and recording conditions. The research aims to propose a novel SXTD-Weighted Visibility Graph framework with an information-rich edge weighting scheme to enhance interpretability and diagnostic accuracy in epilepsy EEG analysis. In addition, new EEG graph features such as Transitive Amplification Index (TAI), MedianWeightEps are developed that capturing the interior fluctuations and trend deviations, remaining shift-invariant and scale-equivariant, providing tunable noise–structure control, and requiring no additional assumptions. The proposed framework achieved 100
