AI in Medicine doi: 10.3390/aimed1030025

Authors: Dimitris Kalatzis Alkmini Nega

Biomedical spectroscopy, including Raman, surface-enhanced Raman spectroscopy (SERS), infrared spectroscopy, and hyperspectral imaging, is increasingly combined with machine learning for disease classification, sample characterization, and biomarker-oriented analysis. However, high predictive performance does not establish whether model-relevant spectral features are biologically meaningful or whether highlighted regions can support reliable biochemical interpretation. Explainable artificial intelligence (XAI) methods, particularly SHAP and LIME, are increasingly used to identify influential wavenumbers, spectral bands, and wavelength intervals; yet feature importance is often interpreted too directly as biochemical or clinical evidence. This focused narrative review synthesizes SHAP, LIME, and related XAI methods across biomedical spectroscopy applications in cancer diagnostics, microbial identification, pharmaceutical analysis, and tissue or biofluid characterization. We examine key challenges, including correlated variables, peak overlap, preprocessing dependence, background choice, model dependence, and explanation instability. Beyond spectral attribution, we propose a five-step validation framework linking valid model development, explanation stability, region-level interpretation, biochemical plausibility, and independent analytical, biological, or clinical validation. The framework is intended to distinguish candidate spectral evidence from unstable or technically confounded explanations and to support reproducible, transparent, and clinically meaningful use of XAI in biomedical spectroscopy.

AI in Medicine doi: 10.3390/aimed1030024

Authors: Qais Al-Azzam Wamadeva Balachandran Ziad Hunaiti

Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide, underscoring the importance of early and accurate diagnosis to improve patient outcomes. Magnetic resonance imaging (MRI) is a highly sensitive imaging modality for detecting breast malignancies, particularly in patients with dense breast tissue or those at high risk, where conventional imaging techniques may have limited sensitivity. Recent advances in deep learning (DL) have demonstrated considerable potential for improving the automated analysis of breast MRI, including tumour classification, prediction, and segmentation. This systematic review synthesises peer-reviewed studies published between 2014 and 2025 that exclusively applied DL techniques to breast MRI for cancer classification, prediction, or segmentation. The included studies were critically evaluated with respect to model architectures, dataset characteristics, image preprocessing methods, validation strategies, and reported performance metrics. The reviewed literature demonstrates that DL models consistently achieve high diagnostic performance and have the potential to enhance radiological workflows by supporting automated lesion detection and clinical decision-making. However, several challenges continue to limit their translation into routine clinical practice, including limited access to large, diverse, and well-annotated datasets, inadequate external validation, variability in MRI acquisition protocols, and concerns regarding model interpretability and generalisability. Future research should prioritise the development of robust, explainable, and clinically validated DL models trained on multicentre datasets using standardised evaluation frameworks. Addressing these challenges will be essential to improve the reliability, reproducibility, and clinical applicability of AI-assisted breast cancer diagnosis using MRI.

AI in Medicine doi: 10.3390/aimed1030023

Authors: Georgios Kritopoulos Georgios Neofotistos Georgios D. Barmparis Giorgos P. Tsironis

Class imbalance in clinical electrocardiogram (ECG) datasets limits the diagnostic sensitivity of automated arrhythmia classifiers, particularly for rare but clinically significant beat types. We propose a three-stage hybrid generative pipeline that combines a spectral-guided conditional variational autoencoder (cVAE), a class-conditional latent denoising diffusion probabilistic model (DDPM), and a Quantum Latent Refinement (QLR) module built on parameterized quantum circuits, implemented and evaluated using a classical quantum-circuit simulator, to augment minority arrhythmia classes, and present results based on the MIT-BIH Arrhythmia Database. The QLR module applies a bounded residual correction guided by Maximum Mean Discrepancy minimization to align synthetic latent distributions with real class-specific latent banks. A lightweight 1D MobileNetV2 classifier evaluated over ten independent random seeds and four augmentation ratios serves as the downstream benchmark. Our findings establish latent diffusion augmentation as an effective strategy for imbalanced ECG classification. To our knowledge, the proposed QLR module is the first use of a parameterized quantum circuit as a distributional refiner within a generative augmentation pipeline. While its performance is comparable to that of the classical latent diffusion framework under the present experimental conditions, the proposed approach demonstrates the feasibility of integrating quantum latent operators into generative medical AI pipelines and provides a foundation for future investigations on quantum-enhanced representation learning and data augmentation.

AI in Medicine doi: 10.3390/aimed1030022

Authors: Ahmad Ibrahim Alshdaifat Wamadeva Balachandran Ziad Hunaiti

Coronary artery disease (CAD) is the leading cause of death worldwide, highlighting the need for more reliable and efficient diagnostic tools beyond conventional methods. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), has shown strong potential for detecting obstructive CAD by learning complex patterns from electrocardiogram (ECG) and coronary computed tomography angiography (CCTA) data. This rapid systematic review assesses and compares the diagnostic performance and methodological quality of AI models built for CAD prediction using ECG and CCTA data. A systematic search following PRISMA-ScR guidelines was conducted for primary studies published between 2021 and 2025. Eleven studies were included, six using ECG data and five using CCTA data. Methodological quality was evaluated using the PROBAST+AI tool. ECG-based models achieved AUCs of 0.72–0.961 and CCTA-based models showed slightly stronger top-end performance, with AUCs of 0.77–0.97. External validation was uncommon in both groups, applied in only 40% of CCTA studies and 33% of ECG studies, so neither modality demonstrated clearly greater validation maturity. Despite these strong results, PROBAST+AI assessment revealed a high risk of bias in 90.9% of the included studies, largely due to weaknesses in the analysis domain, including poor handling of missing data and the absence of model calibration reporting. AI models show strong diagnostic accuracy for CAD across both modalities, although external validation was limited and applied in a minority of studies. However, the widespread methodological bias means these tools should currently support clinical decision-making rather than replace standard diagnostic methods. Future studies should focus on prospective multicentre validation and the use of multimodal data.

AI in Medicine doi: 10.3390/aimed1030021

Authors: Feiya Xiang Geet Khatri Alec Brewer Emily Garceau Kirstie M. K. Queener Mauro Caballero Victorio Parvez Ahmmed James Reynolds Vladimir Aleksandrovich Pozdin Michael Daniele Alper Bozkurt Edgar Lobaton

Wearable sleep monitoring systems provide a scalable and low-burden alternative to laboratory-based polysomnography, but overnight physiological recordings collected from wearable sensors are frequently corrupted by poor skin contact, sensor displacement, flatline behavior, saturation, abrupt autoscaling, outliers, and non-physiological noise. These low-quality segments can prevent reliable extraction of clinically relevant biomarkers, including heart rate, heart rate variability, pulse rate, oxygen saturation, and electrodermal activity, thereby limiting the robustness of downstream sleep stage classification and sleep apnea prediction. In this work, we present a multistage hybrid anomaly detection framework for signal quality assessment in a custom-designed multimodal wearable sleep monitoring platform. The proposed framework is shared across modalities: each sensor stream is segmented into windows, screened using signal processing algorithms for obvious sensor failures, and then analyzed using an autoencoder trained on normal windows to detect subtler morphology-level deviations from normal physiological patterns. On chest ECG recordings collected over 10 overnight sessions, the hybrid fusion detector achieves an accuracy of 0.9067, AUROC of 0.9068, F1 score of 0.8739, precision of 0.9470, and recall of 0.8202, outperforming rule-based detection and autoencoder-based detection alone. Additional experiments on PPG, EDA and ExG recordings show consistently high F1 scores of 0.9431, 0.9567 and 0.8728, respectively. These results demonstrate that hybrid anomaly detection can serve as an effective quality-control layer for multimodal wearable sleep monitoring and provide a foundation for future work on robust sleep stage and apnea prediction under real-world sensing conditions.

AI in Medicine doi: 10.3390/aimed1030020

Authors: Patrick J. Silva Jian Tao Sara L. Rogers Qiang He Joshua D. Robert Lance Black Scott A. Bruce Paula K. Shireman Kenneth S. Ramos

Digital twins (DTs) are dynamic, virtual representations of individual patients that could support predictive diagnostics and personalized therapeutic optimization. Their development depends on patient-level data from real-world data (RWD) sources and electronic medical record (EMR) data, but major barriers in data-stewardship, interoperability, provenance, and governance persist. This review examines critical bottlenecks within the current healthcare ecosystem, with particular attention to fragmented EMRs, limited longitudinal data continuity, and missing, incomplete, or inaccurate information. We address data availability, stewardship, and provenance rather than model construction itself. We also explore ethical imperatives of mitigating representational bias, where over-representation of European ancestry amplifies existing health inequalities. We evaluate blockchain technology as a decentralized trust anchor to ensure provenance, automate consent, and incentivize longitudinal data stewardship. In parallel, we acknowledge that clinically useful DTs also depend on substantial advances in model specification, calibration, and validation, especially for biologically complex diseases. By synthesizing computational, regulatory, and ethical challenges, we provide a roadmap for developing robust, equitable, and interoperable ecosystems for precision medicine.

AI in Medicine doi: 10.3390/aimed1030019

Authors: Tadesse M. Abegaz Gabriel Frietze Anindya Bijoy Das

Gastrointestinal (GI) adverse drug reactions (ADRs) are common among glucagon-like peptide-1 receptor agonist (GLP-1 RA) users and frequently contribute to treatment discontinuation and reduced therapeutic benefit. This retrospective study aimed to develop and validate an explainable artificial intelligence (XAI) model to predict GI ADR risk among GLP-1 RA users using real-world clinical data from the NIH All of Us Research Program. Adults prescribed GLP-1 RAs were identified and classified according to the occurrence of GI ADRs following treatment initiation. Multiple supervised machine learning models, including logistic regression, random forest, extreme gradient boosting (XGBoost), support vector machine, neural network, LightGBM, and CatBoost, were evaluated using demographic, socioeconomic, clinical, medication, and laboratory variables. Model performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1-score. A total of 8697 participants were included, of whom 59.1% experienced GI ADRs. All models demonstrated reasonable predictive performance, with AUC values ranging from 0.82 to 0.84. The XGBoost achieved discrimination of (AUC: 0.84 ± 0.01). SHapley Additive exPlanations (SHAP) identified gastroesophageal reflux disease, hemorrhoids, and elevated HbA1c as important predictors of GI ADR risk. These findings demonstrate the potential utility of explainable machine learning approaches for predicting the safety of GLP-1 RA therapy.

AI in Medicine doi: 10.3390/aimed1030018

Authors: Mizanu Zelalem Degu Midhila Madhusoodanan Medha Chippa Abhilash Hareendranathan

(1) Background: Ultrasound (US) imaging is widely used in clinical diagnosis but is often degraded by speckle noise, which reduces image quality and can hinder interpretation. Deep learning (DL) has emerged as a promising approach for US denoising, yet its clinical applicability remains unclear. (2) Methods: A scoping review of studies published in the last four years on DL-based US denoising was conducted following PRISMA-ScR guidelines. Searches were performed in IEEE-Xplore, PubMed, ScienceDirect, Scopus, Web of Science, and Google Scholar. Data was extracted on anatomy, noise type, learning paradigm, network architecture, datasets, evaluation metrics, and performance outcomes. (3) Results: From 951 records retrieved, 36 studies were included. Most focused on breast, fetal, cardiac, and abdominal US. Convolutional neural networks (CNNs), particularly U-Net, were the most common approach, while generative adversarial network, vision transformers, and variational autoencoders were less explored. Reported peak signal-to-noise ratio ranged from 30 to 45 dB and structural similarity index measure from 0.85 to 0.97. Most studies (34 out of 36) relied on synthetic noise, 2D images and paired datasets, with limited evaluation on real clinical images. (4) Conclusion: Supervised CNN-based methods dominate US image denoising, but clinical translation is limited by reliance on synthetic data. Non-paired and no-ground-truth learning approaches remain underexplored despite their suitability for US imaging. Progress is further hindered by inconsistent evaluation protocols, limited robustness assessment on clinical tasks, and restricted dataset access. Future work should focus on standardized clinically meaningful evaluation, openly available datasets, and clinical validation to improve reliability and generalizability of DL-based US denoising methods.

AI in Medicine doi: 10.3390/aimed1030017

Authors: Brian A. Zaboski Gregory N. Muller

The rapid integration of artificial intelligence (AI) into mental healthcare presents opportunities and ethical challenges, particularly for complex conditions like obsessive–compulsive disorder (OCD). In this perspective, we argue for a Dual Imperative: establishing safety architectures for AI-powered therapeutic tools to prevent algorithmic sycophancy (symptom accommodation), while mandating explainable AI (XAI) in prognostic models to ensure clinical auditability. In therapeutics, we propose a Guardian Angel architecture that utilizes patient-specific fear hierarchies and linguistic stance detection to distinguish compulsive reassurance-seeking from legitimate patient questions. This approach transforms potential therapeutic ruptures into opportunities for distress tolerance via the Digital Ulysses Pact, a patient-authorized, algorithmically enforced response prevention protocol. In diagnostics, we address the black box problem in precision psychiatry. We argue that as AI evolves from detection to high-stakes treatment selection, safety and accountability become a prerequisite for clinical application. Although distinct in implementation, these architectures form an integrated framework for aligning therapeutic and diagnostic AI. These architectures are not parallel tracks but a unified ecosystem: A patient’s XAI-audited profile can inform the Guardian Angel’s configuration, while the longitudinal data gathered during therapy enriches diagnostic precision. Grounded in ethical principles and best practices in OCD, this suggests a path toward AI that is auditable in its diagnostic logic, firm in its therapeutic boundaries, and enforceable through emerging regulatory frameworks.

AI in Medicine doi: 10.3390/aimed1020016

Authors: Kaaviyashri Saraboji Keerthy Gopalakrishnan Divyanshi Sood Anmolpreet Kaur Suganti Shivaram Scott A. Helgeson Shivaram P. Arunachalam Dipankar Mitra

Artificial intelligence (AI) is rapidly transforming healthcare through applications in clinical decision support, diagnostic imaging, population health management, and workflow optimization. Despite these advances, real-world deployment continues to expose critical challenges related to safety, bias, transparency, and integration into clinical workflows. Algorithmic bias can exacerbate health disparities, limited explainability may undermine clinician trust, and insufficient validation and post-deployment monitoring can compromise patient safety. Although the World Health Organization (WHO) has established six ethical principles for AI in health, including autonomy, well-being and safety, transparency, accountability, equity, and sustainability, translating these high-level principles into practical and enforceable governance mechanisms remains a persistent challenge. This narrative review synthesizes insights from bioethics, health policy, computer science, and clinical medicine to identify gaps in current AI governance approaches and proposes a lifecycle-aligned governance-by-design framework that operationalizes WHO ethical principles across key stages of the healthcare AI lifecycle, including data collection, model development, validation, deployment, and post-deployment monitoring. The framework integrates concrete governance mechanisms such as consent governance, fairness evaluation, external validation, explainability, clinician oversight, and continuous performance monitoring. Overall, this work advances a practical, lifecycle-integrated approach to AI governance and provides a structured foundation for developing safe, equitable, and trustworthy AI systems in healthcare.

AI in Medicine doi: 10.3390/aimed1020015

Authors: Luis Ramalhete Vitor Oliveira Rui Quintas Rúben Araújo

Background: Chest radiography is widely used in clinical workflows; however, exploratory image-level classification across multiple public-dataset categories remains less studied than single-disease classification tasks. We aimed to develop and internally evaluate a compact SqueezeNet-based pipeline for nine-class chest radiograph classification within a public dataset. Low-computational-footprint approaches may be relevant for future research prototypes in resource-constrained settings, particularly when offline operation is desirable; however, no real-world clinical deployment or triage validation was assessed in the present study. Methods: Using a public dataset of 6743 frontal radiographs spanning normal anatomy and eight pathology categories, we extracted 512-dimensional embeddings from a pre-trained SqueezeNet-1.0 (features module with global average pooling) and trained a scikit-learn MLP with a single hidden layer. Performance was assessed with stratified 5-fold cross-validation using accuracy and class-wise precision, recall, and F1; interpretability was examined via confusion matrices and dimensionality reduction techniques (t-SNE, and MDS). Results: The model achieved a mean accuracy of 98.83% across folds, with per-class precision, recall, and F1 generally ≥0.96 and a weighted F1 of 0.99; confusion matrices showed minimal off-diagonal errors, and embedding visualizations revealed well-separated, class-consistent clusters. Conclusions: Compact CNN features coupled with a simple MLP demonstrated strong internal performance for multi-class CXR classification within the evaluated dataset. However, the absence of external validation, the use of synthetically augmented data, and the lack of patient-level provenance metadata substantially limit conclusions regarding generalizability and clinical applicability.

AI in Medicine doi: 10.3390/aimed1020014

Authors: Joshua L. Ebbert Dennis Della Corte

Artificial intelligence foundation models are increasingly deployed for prostate cancer Gleason grading, where GP3/GP4 distinction directly impacts treatment decisions (active surveillance vs. intervention). However, these models may achieve high validation accuracy by learning specimen-specific artifacts rather than generalizable biological features, limiting real-world clinical utility. We introduce PANDA-PLUS-Bench, a curated benchmark dataset derived from expertly annotated prostate biopsies designed specifically to quantify this failure mode. The benchmark comprises nine carefully selected whole slide images from nine unique patients containing diverse Gleason patterns, with non-overlapping tissue patches extracted at both 512 × 512 and 224 × 224-pixel resolutions across eight augmentation conditions. Using this benchmark, we evaluate seven foundation models (Virchow, Virchow2, UNI, UNI2, Phikon, Phikon-v2, and HistoEncoder) on their ability to separate biological signals from slide-level confounders. Our results reveal substantial variation in robustness across models: the Virchow models achieved the lowest slide-level encoding among large-scale models (slide ID accuracy: 80.7–81.0%), yet Virchow2 exhibited the lowest cross-slide accuracy (47.2%). HistoEncoder, trained specifically on prostate tissue, demonstrated the highest cross-slide accuracy (59.7%) and the strongest slide-level encoding (slide ID accuracy: 90.3%), suggesting tissue-specific training may enhance both biological feature capture and slide-specific signatures. All models exhibited measurable within-slide vs. cross-slide accuracy gaps, though the magnitude varied from 19.9 percentage points (HistoEncoder) to 26.9 percentage points (Phikon). We provide an open-source Google Colab notebook enabling researchers to evaluate additional foundation models against our benchmark using standardized metrics. PANDA-PLUS-Bench addresses a critical gap in foundation model evaluation by providing a purpose-built resource for robustness assessment in the clinically important context of Gleason grading.

AI in Medicine doi: 10.3390/aimed1020013

Authors: Karim Shater Catharina Gerhards Osman Evliyaoglu Stefanie Nittka Andreas Fischer

Background: Complete blood count (CBC) phenotypes are routinely assigned using deterministic rule-based thresholds. While operationally efficient, such rules may lead to unstable phenotype assignments for results close to clinical cutoffs in the presence of analytical variability. Methods: We analyzed routine CBC data from a tertiary care hospital laboratory. Rule-based phenotypes for anemia subtype, white blood cell (WBC) status, and platelet (PLT) status were assigned using established laboratory thresholds. A patient-independent development and holdout split was applied. A multi-output gradient boosting model was trained to reproduce rule-based labels and provide probabilistic outputs. Phenotype stability was assessed by perturbing CBC parameters under realistic analytical noise. Instability was defined as any change in phenotype assignment across perturbations. Distances to decision boundaries were grouped into quantile-based bins. Model uncertainty was evaluated for the triage of unstable cases. Results: Phenotype instability was strongly concentrated near decision boundaries. Under medium analytical variability, samples closest to hemoglobin cutoffs exhibited the highest instability, with the highest instability in the bin closest to the cutoff, a sharp decrease in the adjacent bin, and lower instability across more distant bins. Model uncertainty was enriched among unstable cases, enabling prioritization of borderline samples while reviewing only a subset of all cases. Conclusions: Rule-based CBC phenotyping exhibits intrinsic instability near decision thresholds. Uncertainty-aware machine learning supports a practical framework to identify and prioritize borderline cases without replacing existing laboratory rules, supporting workload-controlled post-analytical decision support.

AI in Medicine doi: 10.3390/aimed1020012

Authors: A. H. M. Fahad Masahiko Nakatsui Takeshi Abe Takahide Hayano M. H. Mahbub Ryosuke Hase Natsu Yamaguchi Yoshihiro Hayakawa Yusuke Inohana Yutaka Umakoshi Ryo Yamaguchi Ren Kimura Hisashi Tsujimura Mitsuharu Matsumoto Fumiaki Higashijima Takuya Yoshimoto Kazuhiro Kimura Tsunahiko Hirano Keiji Ohishi Keiko Doi Kazuto Matsunaga Tsuyoshi Tanabe Yoshiyuki Asai

This study developed a two-stage, explainable machine learning framework to predict 18-month MMSE-based cognitive status from baseline multimodal data in community-dwelling older adults in Japan. A hierarchical design was used in which Stage 1 distinguished cognitively Normal participants from those with any abnormality (Possible Mild Cognitive Impairment (MCI) or Impaired), and Stage 2 further separated Possible MCI from Impaired within the abnormal subgroup. Both an Imbalanced-Learn Random Forest and a penalized logistic regression baseline were trained under Leave-One-Out Cross-Validation, yielding fair discrimination in Stage 1 (Random Forest AUC = 0.72, accuracy = 0.71; logistic regression AUC = 0.71, accuracy = 0.76) and apparently strong separability in Stage 2 (Random Forest AUC = 0.95, accuracy = 0.96; logistic regression AUC = 0.82, accuracy = 0.92) in a small sample size with high class imbalance. SHapley Additive exPlanations (SHAP) with TreeExplainer for Random Forest and LinearExplainer for logistic regression were used to identify interpretable biomarkers at each stage though feature attribution. In Stage 1, both models highlighted renal and systemic metabolic markers (e.g., creatinine, uric acid, blood urea nitrogen), amino acid and redox-related metabolites (including D-serine, D-amino acid proportions, L-asparagine, alanine, L-glutamic acid, cysteine, methionine sulfoxide), and wearable-derived activity variability (e.g., fluctuation coefficients and steps per minute), with the Simpson index of gut microbiome diversity also contributing in the logistic model. In Stage 2, the models converged on a distinct signature involving glucose and albumin, uric acid and uridine, choline and carnitine, multiple amino acids (such as phenylalanine, proline, ornithine, tryptophan, threonine, and short-chain amino acids), oxidative/energy markers (niacinamide, methionine, methionine sulfoxide, ergothioneine), hematologic indices, and high-MET activity fluctuation metrics. Collectively, these results support a stage-dependent, multisystem view of cognitive aging in which broad renal–metabolic, amino acid, and behavioral vulnerabilities characterize early abnormality, whereas more pronounced alterations in energy metabolism, nucleotide and choline pathways, oxidative stress, and activity irregularity accompany progression from Possible MCI to Impaired status. By combining routine clinical chemistry, targeted metabolomics, gut microbiome diversity, and wearable-derived behavioral measures within an explainable AI framework, this two-stage approach illustrates a scalable, biologically grounded strategy for stage-aware risk stratification and monitoring of cognitive decline in community settings.

AI in Medicine doi: 10.3390/aimed1020011

Authors: Sorin Liviu Jurj

Glucocorticoid-induced hypertension affects over 30% of treated patients, yet its underlying mechanisms remain unclear, particularly how glucocorticoids regulate renin within the renin-angiotensin-aldosterone system (RAAS). Modeling these dynamics is difficult because only four dose-response measurements are available at a single 24-h timepoint (36 observations total), while the system depends on roughly eleven biochemical parameters spanning minutes-long receptor interactions to days-long protein secretion. Classical parameter estimation becomes unreliable in this extremely underdetermined setting, and purely data-driven methods offer limited biological interpretability. In this paper, we introduce a physics-informed neural network (PINN) framework that integrates ELISA measurements from As4.1 juxtaglomerular cells, ordinary differential equations describing glucocorticoid receptor signaling and renin transcription, and automatic differentiation to enforce mechanistic constraints. By systematically tuning synthetic-data weights (SW in {0.2, 0.3, 0.5}), we identify an intermediate value of SW = 0.3 that provides the best overall balance between predictive accuracy, accepted ensemble size, and biologically plausible parameter estimates among the tested configurations. The framework uses adaptive constraint scheduling with a plateau ramp to reduce premature convergence and introduces calibrated plausibility thresholds reflecting experimental noise. The accepted PINN ensemble (n = 5, 50% success rate) achieved R2 = 0.803, compared with 0.759 for the SW = 0.5 baseline and −0.220 for the ODE-only baseline, with RMSE = 0.024. Key learned parameters (IC50 = 2.925 ± 0.012 mg/dL, Hill = 1.950 ± 0.009) are biologically plausible within the model assumptions, and the best single accepted model attained R2 = 0.891. Information criteria favored the PINN over the ODE model, with improvements of approximately 77× (AIC) and 5.9× (BIC). Despite training on a single timepoint, the PINN also infers full 48-h trajectories and reproduces non-monotonic dose-response behavior. This work presents, to our knowledge, the first PINN framework for glucocorticoid-mediated renin regulation and should be interpreted as a proof-of-concept approach for integrating sparse biomedical data with mechanistic constraints. The inferred parameters and temporal dynamics are best viewed as model-dependent, hypothesis-generating estimates rather than validated biological quantities.

AI in Medicine doi: 10.3390/aimed1020010

Authors: Jeremie Biringanine Ruvunangiza Carlos Alberto Valderrama Sakuyama

The exponential growth of unstructured medical data, particularly clinical notes and diagnostic reports, presents mounting challenges for healthcare knowledge extraction and utilization. This study introduces the Digital Medical Catalog (DMC), a framework that automates the conversion of clinical narratives into an auditable, semantically structured knowledge base. The framework combines BioClinicalBERT embeddings, c-TF-IDF statistical grounding, and semantic clustering, enabling high-fidelity classification (Macro F1 = 0.877 ± 0.012), traceable topic labeling, and temporal trend analysis. By demonstrating that semantic representation methods, reinforced with statistical grounding, are essential for large-scale medical text processing, this work establishes a foundation for privacy-preserving data governance and real-time intelligence within modern healthcare infrastructures.

AI in Medicine doi: 10.3390/aimed1010009

Authors: Pierre Boulanger

Distinguished colleagues in the global medical and AI communities, [...]

AI in Medicine doi: 10.3390/aimed1010008

Authors: Christos Raptis Nikolaos Bouzianis Efstratios Karavasilis Athanasios Zissimopoulos Pipitsa Valsamaki Athanasia Kotini Georgios Anastassopoulos Adam Adamopoulos

This study introduces a novel approach for optimizing models that reconstruct high-quality full-dose bone scintigraphy images from their 40% low-dose counterparts using optimized attention-based U-Net architectures. We utilized Genetic and Epigenetic Algorithms (epiGA) hyperparameter optimization of two distinct models: a standard Attention U-Net and an Attention U-Net modified with ResNet blocks. Models were trained using a hybrid Mean Squared Error and Structural Similarity (MSE/SSIM) loss function. Obtained results demonstrated superior performance, achieving an average SSIM of 0.9197 and an average Peak Signal-to-Noise ratio (PSNR) of 34.1516 dB, significantly surpassing the baseline low-dose image quality, by gaining ΔSSIM = 0.0333 and ΔPSNR = 3.0729 dB, due to hyperparameter optimization. Comparative benchmarks against Bayesian optimization revealed that epiGA offers superior search efficiency—exploring twice the architecture space in comparable wall-clock time—while consistently identifying more compact, hardware-efficient solutions. These results highlight the effectiveness of integrating epigenetic mechanisms for robust, scalable hyperparameter tuning in medical image reconstruction.

AI in Medicine doi: 10.3390/aimed1010007

Authors: Bhavya Gandhi Leo Morjaria Imeth Illamperuma Praveen Nadesan Aidan Arora Matthew Sibbald

Large language model-based virtual patients (LLM-VPs) are an emerging simulation tool for health professions education, but their design and integration into curricula are not well characterized. This scoping review mapped how LLM-VPs are being used for simulated clinical learning across health professions. Following a protocol registered on OSF, we searched MEDLINE, EMBASE, CENTRAL, Scopus, and Web of Science to 11 April 2025, per PRISMA-ScR guidelines, and included 21 studies that used LLMs to generate virtual patients for simulated clinical encounters. Data were extracted on technical design, fidelity domains, curricular integration, human factors, and Technology Acceptance Model constructs, and synthesized narratively. Most studies (n = 11) were pilot or feasibility evaluations with small samples (median 21) and used GPT-based models with dynamic text chat. Integration was limited to 10 studies that operated as pilots, 7 as electives, and 3 as core curricular components. The outcomes focused on Level 2 learning (clinical reasoning and preclinical OSCE performance), with predominantly self-report assessments. No studies reported Level 3 or 4 outcomes. Fidelity was strongest in cognitive, socio-cultural, and emotional domains, and 11 studies reported hallucinations or inaccurate outputs. LLM-VPs appear feasible and well-received but remain early-stage, underscoring the need for fidelity-aligned design and more rigorous, longitudinal evaluations.

AI in Medicine doi: 10.3390/aimed1010006

Authors: Athanasios Kranas Vassilios S. Verykios

Internet Gaming Disorder (IGD), recognized in the International Classification of Diseases (ICD-11), affects millions—especially adolescents and young adults—and poses challenges that invite scalable innovations in care. This narrative review examined how Large Language Models (LLMs) could support IGD prevention, assessment, treatment, and research. We conducted targeted searches of PubMed, Scopus, Google Scholar, and IEEE Xplore for 2010–October 2025, supplemented by backward/forward citation chasing. English, peer-reviewed clinical, methodological, and review work was prioritized. As a narrative review, we did not apply PRISMA or perform quantitative synthesis. In total, we synthesized over 50 sources. We analyzed peer-reviewed, IGD-specific AI/ML studies with explicit reporting of training approaches, validation/performance, dataset size, and model openness. While preliminary improvements have been observed in digital-health trials for depression, anxiety, and substance use, we emphasize that no IGD-specific LLM therapeutic trials exist to date; thus, evidence regarding treatment efficacy discussed here is extrapolated from these adjacent disorders. Evidence spans transformer-embedding text screening (r ≈ 0.48) and multimodal classification using EEG or fNIRS (accuracy ≈ 71–88%). Sample sizes ranged from 40 to 417 participants. Notably, most implementations remain research-only, lacking public code or data. Principal concerns include privacy and data governance, algorithmic bias, inconsistent crisis-escalation performance, and a nascent clinical evidence base. We conclude that LLMs may augment—but should not replace—human clinicians; future potential lies in hybrid human–AI pathways, multimodal integrations with wearables and gaming APIs, and rigorous prospective trials to establish safety, effectiveness, and equity in IGD care.

AI in Medicine doi: 10.3390/aimed1010005

Authors: Donald C. Wunsch Daniel B. Hier Donald C. Wunsch

Neuromuscular diseases are biologically diverse, clinically heterogeneous, and often difficult to diagnose and treat, highlighting the need for computational tools that can help resolve overlapping phenotypes and support timely, mechanism-informed interventions. This narrative review synthesizes recent advances in artificial intelligence (AI) and machine learning applied to neuromuscular diseases across diagnosis, outcome modeling, biomarker development, and therapeutics. AI-based approaches may assist clinical and genetic diagnosis from phenotypic data; however, early phenotype-driven tools have seen limited clinician adoption due to modest accuracy, usability challenges, and poor workflow integration. Electrophysiological studies remain central to diagnosing neuromuscular diseases, and AI shows promise for accurate classification of electrophysiological signals. Predictive models for disease outcome and progression—particularly in amyotrophic lateral sclerosis—are under active investigation, but most remain at an early stage of development and are not yet ready for routine clinical use. Digital biomarkers derived from imaging, gait, voice, and wearable sensors are emerging, with MRI-based quantification of muscle fat replacement representing the most mature and widely accepted application to date. Efforts to apply AI to therapeutic discovery, including drug repurposing and optimization of gene-based therapies, are ongoing but have thus far yielded limited clinical translation. Persistent barriers to broader adoption include disease rarity, data scarcity, heterogeneous acquisition protocols, inconsistent terminology, limited external validation, insufficient model explainability, and lack of seamless integration into clinical workflows. Addressing these challenges is essential to moving AI tools from the laboratory into clinical practice. We conclude with a practical checklist of considerations intended to guide the development and adoption of AI tools in neuromuscular disease care.

AI in Medicine doi: 10.3390/aimed1010004

Authors: Pierre Boulanger

Despite widespread adoption, Electronic Medical Record (EMR) systems remain limited in providing intelligent clinical decision support, particularly for early detection of patient deterioration. We present MedROAD V2 (Medical Records Organization, Analysis, and Display), an open-source EMR that integrates AI-driven physiological analysis with comprehensive patient management. The system combines continuous vital sign monitoring and laboratory data using an ensemble of the following four complementary machine learning models: gradient boosting for supervised prediction, isolation forests for anomaly detection, autoencoders for pattern recognition, and Long Short-Term Memory networks for temporal modeling. A novel framework couples these predictions with a large language model (Claude AI) to generate explainable differential diagnoses grounded in medical literature. Validation on the MIMIC-IV database demonstrated excellent 12 h deterioration prediction. MedROAD demonstrates that combining quantitative prediction with natural language explanation can enhance clinical decision support while extending quality care to populations that would otherwise lack access.

AI in Medicine doi: 10.3390/aimed1010003

Authors: Sotiria Vrouva Christos Raptis George A. Koumantakis George Anastasopoulos Efstratios Karavasilis Adam Adamopoulos

Within the framework of the ongoing development of application of Machine Learning models in Medicine and Physical Therapy, the development of accurate prognosis algorithms for postoperative patients during the rehabilitation phase remains an area requiring further refinement. This paper examines hybrid Deep Learning models that integrate Convolution Neural Networks, Long Short-Term Memory and Gated Recurrent Unit networks, as well as genetic algorithm optimization for feature selection for predicting the time needed for a patient to rehabilitate. Patient data included features like age, passive range of available movements (preoperative and postoperative) and total rehabilitation time. Genetic Algorithm optimization for feature selection indicated that 4 out of the 16 available features are adequate for predicting rehabilitation time. Hybrid Deep Learning models achieved a Root Mean Squared Error (RMSE) of 12 days (less than 0.4 months) in rehabilitation time prediction, demonstrating good performance on a relatively small dataset of 120 patients.

AI in Medicine doi: 10.3390/aimed1010002

Authors: Agostino Marengo Vito Santamato Massimo Iacoviello

The increasing challenges of healthcare systems demand innovative approaches to resource optimization, particularly for hospitals operating under economic and operational constraints. This study investigates the organizational and managerial factors influencing scale efficiency in 127 Italian hospitals, leveraging advanced machine learning (ML) techniques to identify key determinants of efficiency. A multi-level framework was developed, integrating Principal Component Analysis (PCA), Data Envelopment Analysis (DEA), and K-Means clustering to assess the interplay between energy costs, staff composition, and medical equipment across hospital levels. To enhance predictive capabilities, a classification model based on K-Nearest Neighbors (K-NN) was implemented, demonstrating high performance in distinguishing efficiency classes and confirming the importance of targeted resource management strategies. Additionally, the use of LIME (Local Interpretable Model-agnostic Explanations) provided actionable insights into the contribution of individual variables, enabling a deeper understanding of their impact on operational efficiency. In conclusion, this research highlights the importance of an integrated approach to support decision-makers in managing hospital resources, offering innovative tools to optimize efficiency and ensure the economic and operational sustainability of the healthcare system.

AI in Medicine doi: 10.3390/aimed1010001

Authors: Luis Ramalhete Rúben Araújo Cristiana Teixeira Isaias Pedro Isabel Silva Anibal Ferreira

Background: Mapping anti–HLA class I cross-reactivity from single-antigen bead (SAB) mean fluorescence intensity (MFI) data supports donor selection. However, interpretation is complicated by analytical choices and assay variability. Methods: A total of 4327 SAB assays were analyzed (antigen × test matrix) using an interpretable, distance-based workflow. Antigen profiles were z-scored across tests. Multidimensional scaling (MDS) was used for visualization and hierarchical clustering analysis (HCA) for grouping, and complemented these with a common PCA space for model selection (K-Means via Silhouette; Gaussian Mixture Models via BIC), agglomerative (Ward and average-link on 1–correlation), spectral clustering on correlation-derived affinities, and a graph approach (k-NN ∪ minimum-spanning-tree with modularity-based communities). Non-linear embeddings (t-SNE/UMAP) and density-based HDBSCAN were used only for visual analytics, not for primary inference. Results: The pipeline revealed coherent reactivity neighborhoods that partially overlapped known cross-reactive antigen groups (CREGs) and eplet-based expectations while also highlighting less-documented relationships. Robustness was confirmed through bootstrap resampling, graph modularity, and consensus clustering across methods. Conclusions: This unified, auditable workflow converts descriptive maps into method-robust summaries of antibody reactivity and cross-reactivity. While exploratory and performed on a single dataset without linked outcomes, the approach provides a reproducible structure for comparing cohorts and prioritizing hypotheses that could be prospectively validated for clinical decision support in transplantation.