Causability and explainability of artificial intelligence in medicine

Explainability and uncertainty: Two sides of the same coin for enhancing the interpretability of deep learning models in healthcare

BACKGROUND

The increasing use of Deep Learning (DL) in healthcare has highlighted the critical need for improved transparency and interpretability. While Explainable Artificial Intelligence (XAI) methods provide insights into model predictions, reliability cannot be guaranteed by simply relying on explanations.

OBJECTIVES

This position paper proposes the integration of Uncertainty Quantification (UQ) with XAI methods to improve model reliability and trustworthiness in healthcare applications.

METHODS

We examine state-of-the-art XAI and UQ techniques, discuss implementation challenges, and suggest solutions to combine UQ with XAI methods. We propose a framework for estimating both aleatoric and epistemic uncertainty in the XAI context, providing illustrative examples of their potential application.

RESULTS

Our analysis indicates that integrating UQ with XAI could significantly enhance the reliability of DL models in practice. This approach has the potential to reduce interpretation biases and over-reliance, leading to more cautious and conscious use of AI in healthcare.

Explainable Depression Classification Based on EEG Feature Selection From Audio Stimuli

With the development of affective computing and Artificial Intelligence (AI) technologies, Electroencephalogram (EEG)-based depression detection methods have been widely proposed. However, existing studies have mostly focused on the accuracy of depression recognition, ignoring the association between features and models. Additionally, there is a lack of research on the contribution of different features to depression recognition. To this end, this study introduces an innovative approach to depression detection using EEG data, integrating Ant-Lion Optimization (ALO) and Multi-Agent Reinforcement Learning (MARL) for feature fusion analysis. The inclusion of Explainable Artificial Intelligence (XAI) methods enhances the explainability of the model's features. The Time-Delay Embedded Hidden Markov Model (TDE-HMM) is employed to infer internal brain states during depression, triggered by audio stimulation. The ALO-MARL algorithm, combined with hyper-parameter optimization of the XGBoost classifier, achieves high accuracy (93.69%), sensitivity (88.60%), specificity (97.08%), and F1-score (91.82%) on a auditory stimulus-evoked three-channel EEG dataset. The results suggest that this approach outperforms state-of-the-art feature selection methods for depression recognition on this dataset, and XAI elucidates the critical impact of the minimum value of Power Spectral Density (PSD), Sample Entropy (SampEn), and Rényi Entropy (Ren) on depression recognition. The study also explores dynamic brain state transitions revealed by audio stimuli, providing insights for the clinical application of AI algorithms in depression recognition.

Influence of artificial intelligence on ophthalmologists' judgments in glaucoma

PURPOSE

To examine the influence of artificial intelligence (AI) on physicians' judgments regarding the presence and severity of glaucoma on fundus photographs in an online simulation system.

METHODS

Forty-five trainee and expert ophthalmologists independently evaluated 120 fundus photographs, including 30 photographs each from patients with no glaucoma, mild glaucoma, moderate glaucoma, and severe glaucoma. A second trial was conducted at least one week after the initial trial in which photograph presentation order was randomized. During the second trial, 30% of the glaucoma judgments made by the AI system were intentionally incorrect. The evaluators were asked about their thoughts on AI in ophthalmology via a 3-item questionnaire.

RESULTS

The percentage of correct responses for all images significantly improved (P < 0.001) from 48.4 ± 24.8% in the initial trial to 59.6 ± 20.3% in the second trial. The improvement in the correct response rate was significantly greater for trainees (14.2 ± 19.0%) than for experts (8.6 ± 11.4%) (P = 0.04). The correct response rate was 63.9 ± 20.6% when the AI response was correct, significantly greater than the 47.9 ± 26.6% when the AI response was incorrect (P < 0.0001). For trainees, the correct response rate was significantly greater when the AI's response was correct than when it was incorrect. However, for experts, the effect was less pronounced. The decision time was significantly longer when the AI response was incorrect than when it was correct (P = 0.003).

CONCLUSION

In fundus photography-based glaucoma detection, the results of AI systems can influence physicians' judgments, particularly those of physicians with less experience.

Using artificial intelligence to predict patient outcomes from patient-reported outcome measures: a scoping review

PURPOSE

This scoping review aims to identify and summarise artificial intelligence (AI) methods applied to patient-reported outcome measures (PROMs) for prediction of patient outcomes, such as survival, quality of life, or treatment decisions.

INTRODUCTION

AI models have been successfully applied to predict outcomes for patients using mainly clinically focused data. However, systematic guidance for utilising AI and PROMs for patient outcome predictions is lacking. This leads to inconsistency of model development and evaluation, limited practical implications, and poor translation to clinical practice.

MATERIALS AND METHODS

This review was conducted across Web of Science, IEEE Xplore, ACM, Digital Library, Cochrane Central Register of Controlled Trials, Medline and Embase databases. Adapted search terms identified published research using AI models with patient-reported data for outcome predictions. Papers using PROMs data as input variables in AI models for prediction of patient outcomes were included.

RESULTS

Three thousand and seventy-seven records were screened, 94 of which were included in the analysis. AI models applied to PROMs data for outcome predictions are most commonly used in orthopaedics and oncology. Poor reporting of model hyperparameters and inconsistent techniques of handling class imbalance and missingness in data were found. The absence of external model validation, participants' ethnicity information and stakeholders involvement was common.

CONCLUSION

The results highlight inconsistencies in conducting and reporting of AI research involving PROMs in patients' outcomes predictions, which reduces the reproducibility of the studies. Recommendations for external validation and stakeholders' involvement are given to increase the opportunities for applying AI models in clinical practice.

Remote Vital Sensing in Clinical Veterinary Medicine: A Comprehensive Review of Recent Advances, Accomplishments, Challenges, and Future Perspectives

Remote vital sensing in veterinary medicine is a relatively new area of practice, which involves the acquisition of data without invasion of the body cavities of live animals. This paper aims to review several technologies in remote vital sensing: infrared thermography, remote photoplethysmography (rPPG), radar, wearable sensors, and computer vision and machine learning. In each of these technologies, we outline its concepts, uses, strengths, and limitations in multiple animal species, and its potential to reshape health surveillance, welfare evaluation, and clinical medicine in animals. The review also provides information about the problems associated with applying these technologies, including species differences, external conditions, and the question of the reliability and classification of these technologies. Additional topics discussed in this review include future developments such as the use of artificial intelligence, combining different sensing methods, and creating monitoring solutions tailored to specific animal species. This contribution gives a clear understanding of the status and future possibilities of remote vital sensing in veterinary applications and stresses the importance of that technology for the development of the veterinary field in terms of animal health and science.

Causality and scientific explanation of artificial intelligence systems in biomedicine

With rapid advances of deep neural networks over the past decade, artificial intelligence (AI) systems are now commonplace in many applications in biomedicine. These systems often achieve high predictive accuracy in clinical studies, and increasingly in clinical practice. Yet, despite their commonly high predictive accuracy, the trustworthiness of AI systems needs to be questioned when it comes to decision-making that affects the well-being of patients or the fairness towards patients or other stakeholders affected by AI-based decisions. To address this, the field of explainable artificial intelligence, or XAI for short, has emerged, seeking to provide means by which AI-based decisions can be explained to experts, users, or other stakeholders. While it is commonly claimed that explanations of artificial intelligence (AI) establish the trustworthiness of AI-based decisions, it remains unclear what traits of explanations cause them to foster trustworthiness. Building on historical cases of scientific explanation in medicine, we here propagate our perspective that, in order to foster trustworthiness, explanations in biomedical AI should meet the criteria of being scientific explanations. To further undermine our approach, we discuss its relation to the concepts of causality and randomized intervention. In our perspective, we combine aspects from the three disciplines of biomedicine, machine learning, and philosophy. From this interdisciplinary angle, we shed light on how the explanation and trustworthiness of artificial intelligence relate to the concepts of causality and robustness. To connect our perspective with AI research practice, we review recent cases of AI-based studies in pathology and, finally, provide guidelines on how to connect AI in biomedicine with scientific explanation.

ViT-based quantification of intratumoral heterogeneity for predicting the early recurrence in HCC following multiple ablation

OBJECTIVES

This study aimed to develop a quantitative intratumoral heterogeneity (ITH) model for assessing the risk of early recurrence (ER) in pre-treatment multimodal imaging for hepatocellular carcinoma (HCC) patients undergoing ablation treatments.

METHODS

This multi-centre study enrolled 633 HCC patients who underwent ultrasound-guided local ablation between January 2015 and September 2022. Among them, 422, 85, 57 and 69 patients underwent radiofrequency ablation (RFA), microwave ablation (MWA), laser ablation (LA) and irreversible electroporation (IRE) ablation, respectively. Vision-Transformer-based quantitative ITH (ViT-Q-ITH) features were extracted from the US and MRI sequences. Multivariable logistic regression analysis was used to identify variables associated with ER. A combined model integrated clinic-radiologic and ViT-Q-ITH scores. The prediction performance was evaluated concerning calibration, clinical usefulness and discrimination.

RESULTS

The final training cohort and internal validation cohort included 318 patients and 83 patients, respectively, who underwent RFA and MWA. The three external testing cohorts comprised of 106 patients treated with RFA, 57 patients treated with LA and 69 patients who underwent IRE ablation. The combined model showed excellent predictive performance for ER in the training (AUC: .99, 95% CI: .99-1.00), internal validation (AUC: .86, 95% CI: .78-.94), external testing (AUC: .83, 95% CI: .73-.92), LA (AUC: .84, 95% CI: .73-.95) and IRE (AUC: .82, 95% CI: .72-.93) cohorts, respectively. Decision curve analysis further affirmed the clinical utility of the combined model.

CONCLUSIONS

The multimodal-based model, incorporating clinic-radiologic factors and ITH features, demonstrated superior performance in predicting ER among early-stage HCC patients undergoing different ablation modalities.

Advancing non-target analysis of emerging environmental contaminants with machine learning: Current status and future implications

Emerging environmental contaminants (EECs) such as pharmaceuticals, pesticides, and industrial chemicals pose significant challenges for detection and identification due to their structural diversity and lack of analytical standards. Traditional targeted screening methods often fail to detect these compounds, making non-target analysis (NTA) using high-resolution mass spectrometry (HRMS) essential for identifying unknown or suspected contaminants. However, interpreting the vast datasets generated by HRMS is complex and requires advanced data processing techniques. Recent advancements in machine learning (ML) models offer great potential for enhancing NTA applications. As such, we reviewed key developments, including optimizing workflows using computational tools, improved chemical structure identification, advanced quantification methods, and enhanced toxicity prediction capabilities. It also discusses challenges and future perspectives in the field, such as refining ML tools for complex mixtures, improving inter-laboratory validation, and further integrating computational models into environmental risk assessment frameworks. By addressing these challenges, ML-assisted NTA can significantly enhance the detection, quantification, and evaluation of EECs, ultimately contributing to more effective environmental monitoring and public health protection.

Artificial intelligence in dentistry: Exploring emerging applications and future prospects

OBJECTIVES

This narrative review aimed to explore the evolution and advancements of artificial intelligence technologies, highlighting their transformative impact on healthcare, education, and specific aspects within dentistry as a field.

DATA AND SOURCES

Subtopics within artificial intelligence technologies in dentistry were identified and divided among four reviewers. Electronic searches were performed with search terms that included: artificial intelligence, technologies, healthcare, education, dentistry, restorative, prosthodontics, periodontics, endodontics, oral surgery, oral pathology, oral medicine, implant dentistry, dental education, dental patient care, dental practice management, and combinations of the keywords.

STUDY

selection: A total of 120 articles were included for review that evaluated the use of artificial intelligence technologies within the healthcare and dental field. No formal evidence-based quality assessment was performed due to the narrative nature of this review. The conducted search was limited to the English language with no other further restrictions.

RESULTS

The significance and applications of artificial intelligence technologies on the areas of dental education, dental patient care, and dental practice management were reviewed, along with the existing limitations and future directions of artificial intelligence in dentistry as whole. Current artificial intelligence technologies have shown promising efforts to bridge the gap between theoretical knowledge and clinical practice in dental education, as well as improved diagnostic information gathering and clinical decision-making abilities in patient care throughout various dental specialties. The integration of artificial intelligence into patient administration aspects have enabled practices to develop more efficient management workflows.

CONCLUSIONS

Despite the limitations that exist, the integration of artificial intelligence into the dental profession comes with numerous benefits that will continue to evolve each day. While the challenges and ethical considerations, mainly concerns about data privacy, are areas that need to be further addressed, the future of artificial intelligence in dentistry looks promising, with ongoing research aimed at overcoming current limitations and expanding artificial intelligence technologies.

XGate: Explainable Reinforcement Learning for Transparent and Trustworthy API Traffic Management in IoT Sensor Networks

The rapid proliferation of Internet of Things (IoT) devices and their associated application programming interfaces (APIs) has significantly increased the complexity of sensor network traffic management, necessitating more sophisticated and transparent control mechanisms. In this paper, we introduce XGate, a novel explainable reinforcement learning framework designed specifically for API traffic management in sensor networks. XGate addresses the critical challenge of balancing optimal routing decisions with the interpretability demands of network administrators operating large-scale IoT deployments. Our approach integrates transformer-based attention mechanisms with counterfactual reasoning to provide human-comprehensible explanations for each traffic management decision across distributed sensor data streams. Through extensive experimentation on three large-scale sensor API traffic datasets, we demonstrate that XGate achieves 23.7% lower latency and 18.5% higher throughput compared to state-of-the-art black-box reinforcement learning approaches. More importantly, our user studies with sensor network administrators (n=42) reveal that XGate's explanation capabilities improve operator trust by 67% and reduce intervention time by 41% during anomalous sensor traffic events. The theoretical analysis further establishes probabilistic guarantees on explanation fidelity while maintaining computational efficiency suitable for real-time sensor data management. XGate represents a significant advancement toward trustworthy AI systems for critical IoT infrastructure, providing transparent decision making without sacrificing performance in dynamic sensor network environments.

AI-driven innovations in smart multifunctional nanocarriers for drug and gene delivery: A mini-review

The convergence of artificial intelligence (AI) and nanomedicine has revolutionized the design of smart multifunctional nanocarriers (SMNs) for drug and gene delivery, offering unprecedented precision, efficiency, and personalization in therapeutic applications. AI-driven approaches enhance the development of these nanocarriers by accelerating their design, optimizing drug loading and release kinetics, improving biocompatibility, and predicting interactions with biological barriers. This review explores the transformative role of AI in the fabrication and functionalization of SMNs, emphasizing its impact on overcoming challenges in targeted drug delivery, controlled release, and theranostics. We discuss the integration of AI with advanced nanomaterials-such as polymeric, lipidic, and inorganic nanoparticles-highlighting their potential in oncology and hematology. Furthermore, we examine recent clinical and preclinical case studies demonstrating AI-assisted nanocarrier development for personalized medicine. The synergy between AI and nanotechnology paves the way for next-generation precision therapeutics, addressing critical limitations in traditional drug delivery systems. However, data standardization, regulatory compliance, and translational scalability challenges remain. This review underscores the need for interdisciplinary collaboration to unlock AI's potential in nanomedicine fully, ultimately advancing the clinical application of SMNs for more effective and safer patient care.

Colon Cancer Screening, Surveillance, and Treatment: Novel Artificial Intelligence Driving Strategies in the Management of Colon Lesions

Colonoscopy, a crucial procedure for detecting and removing colorectal polyps, has seen transformative advancements through the integration of artificial intelligence, specifically in computer-aided detection (CADe) and diagnosis (CADx). These tools enhance real-time detection and characterization of lesions, potentially reducing human error, and standardizing the quality of colonoscopy across endoscopists. CADe has proven effective in increasing adenoma detection rate, potentially reducing long-term colorectal cancer incidence. However, CADe's benefits are accompanied by challenges, such as potentially longer procedure times, increased non-neoplastic polyp resections, and a higher surveillance burden. CADx, although promising in differentiating neoplastic and non-neoplastic diminutive polyps, encounters limitations in accuracy, particularly in the proximal colon. Real-world data also revealed gaps between trial efficacy and practical outcomes, emphasizing the need for further research in uncontrolled settings. Moreover, CADx limited specificity and binary output underscore the necessity for explainable artificial intelligence to gain endoscopists' trust. This review aimed to explore the benefits, harms, and limitations of artificial intelligence for colon cancer screening, surveillance, and treatment focusing on CADe and CADx systems for lesion detection and characterization, respectively, while addressing challenges in integrating these technologies into clinical practice.

On the practical, ethical, and legal necessity of clinical Artificial Intelligence explainability: an examination of key arguments

The necessity for explainability of artificial intelligence technologies in medical applications has been widely discussed and heavily debated within the literature. This paper comprises a systematized review of the arguments supporting and opposing this purported necessity. Both sides of the debate within the literature are quoted to synthesize discourse on common recurring themes and subsequently critically analyze and respond to it. While the use of autonomous black box algorithms is compellingly discouraged, the same cannot be said for the whole of medical artificial intelligence technologies that lack explainability. We contribute novel comparisons of unexplainable clinical artificial intelligence tools, diagnosis of idiopathy, and diagnoses by exclusion, to analyze implications on patient autonomy and informed consent. Applying a novel approach using comparisons with clinical practice guidelines, we contest the claim that lack of explainability compromises clinician due diligence and undermines epistemological responsibility. We find it problematic that many arguments in favour of the practical, ethical, or legal necessity of clinical artificial intelligence explainability conflate the use of unexplainable AI with automated decision making, or equate the use of clinical artificial intelligence with the exclusive use of clinical artificial intelligence.

Explainable Artificial Intelligence in Neuroimaging of Alzheimer's Disease

Alzheimer's disease (AD) remains a significant global health challenge, affecting millions worldwide and imposing substantial burdens on healthcare systems. Advances in artificial intelligence (AI), particularly in deep learning and machine learning, have revolutionized neuroimaging-based AD diagnosis. However, the complexity and lack of interpretability of these models limit their clinical applicability. Explainable Artificial Intelligence (XAI) addresses this challenge by providing insights into model decision-making, enhancing transparency, and fostering trust in AI-driven diagnostics. This review explores the role of XAI in AD neuroimaging, highlighting key techniques such as SHAP, LIME, Grad-CAM, and Layer-wise Relevance Propagation (LRP). We examine their applications in identifying critical biomarkers, tracking disease progression, and distinguishing AD stages using various imaging modalities, including MRI and PET. Additionally, we discuss current challenges, including dataset limitations, regulatory concerns, and standardization issues, and propose future research directions to improve XAI's integration into clinical practice. By bridging the gap between AI and clinical interpretability, XAI holds the potential to refine AD diagnostics, personalize treatment strategies, and advance neuroimaging-based research.

Do explainable AI (XAI) methods improve the acceptance of AI in clinical practice? An evaluation of XAI methods on Gleason grading

This work aimed to evaluate both the usefulness and user acceptance of five gradient-based explainable artificial intelligence (XAI) methods in the use case of a prostate carcinoma clinical decision support system environment. In addition, we aimed to determine whether XAI helps to increase the acceptance of artificial intelligence (AI) and recommend a particular method for this use case. The evaluation was conducted on a tool developed in-house with different visualization approaches to the AI-generated Gleason grade and the corresponding XAI explanations on top of the original slide. The study was a heuristic evaluation of five XAI methods. The participants were 15 pathologists from the University Hospital of Augsburg with a wide range of experience in Gleason grading and AI. The evaluation consisted of a user information form, short questionnaires on each XAI method, a ranking of the methods, and a general questionnaire to evaluate the performance and usefulness of the AI. There were significant differences between the ratings of the methods, with Grad-CAM++ performing best. Both AI decision support and XAI explanations were seen as helpful by the majority of participants. In conclusion, our pilot study suggests that the evaluated XAI methods can indeed improve the usefulness and acceptance of AI. The results obtained are a good indicator, but further studies involving larger sample sizes are warranted to draw more definitive conclusions.

Detecting IDH and TERTp mutations in diffuse gliomas using 1H-MRS with attention deep-shallow networks

BACKGROUND

Preoperative and noninvasive detection of isocitrate dehydrogenase (IDH) and telomerase reverse transcriptase gene promoter (TERTp) mutations in glioma is critical for prognosis and treatment planning. This study aims to develop deep learning classifiers to identify IDH and TERTp mutations using proton magnetic resonance spectroscopy (1H-MRS) and a one-dimensional convolutional neural network (1D-CNN) architecture.

METHODS

This study included 1H-MRS data from 225 adult patients with hemispheric diffuse glioma (117 IDH mutants and 108 IDH wild-type; 99 TERTp mutants and 100 TERTp wild-type). The spectra were processed using the LCModel, and multiple deep learning models, including a baseline, a deep-shallow network, and an attention deep-shallow network (ADSN), were trained to classify mutational subgroups of gliomas. The Gradient-weighted Class Activation Mapping (Grad-CAM) technique was used to interpret the models' decision-making process.

RESULTS

The ADSN model was the most effective for IDH mutation detection, achieving F1-scores of 93 % on the validation set and 88 % on the test set. For TERTp mutation detection, the ADSN model achieved F1-scores of 80 % in the validation set and 81 % in the test set, whereas TERTp-only gliomas were detected with F1-scores of 88 % in the validation set and 86 % in the test set using the same architecture.

CONCLUSION

Deep learning models accurately predicted the IDH and TERTp mutational subgroups of hemispheric diffuse gliomas by extracting relevant information from 1H-MRS spectra without the need for manual feature extraction.

Retrieval Augmented Medical Diagnosis System

Subjective variability in human interpretation of diagnostic imaging presents significant clinical limitations, potentially resulting in diagnostic errors and increased healthcare costs. While artificial intelligence (AI) algorithms offer promising solutions to reduce interpreter subjectivity, they frequently demonstrate poor generalizability across different healthcare settings. To address these issues, we introduce Retrieval Augmented Medical Diagnosis System (RAMDS), which integrates an AI classification model with a similar image model. This approach retrieves historical cases and their diagnoses to provide context for the AI predictions. By weighing similar image diagnoses alongside AI predictions, RAMDS produces a final weighted prediction, aiding physicians in understanding the diagnosis process. Moreover, RAMDS does not require complete retraining when applied to new datasets; rather, it simply necessitates re-calibration of the weighing system. When RAMDS fine-tuned for negative predictive value was evaluated on breast ultrasounds for cancer classification, RAMDS improved sensitivity by 21% and negative predictive value by 9% compared to ResNet-34. Offering enhanced metrics, explainability, and adaptability, RAMDS represents a notable advancement in medical AI. RAMDS is a new approach in medical AI that has the potential for pan-pathological uses, though further research is needed to optimize its performance and integrate multimodal data.

AI applications in HIV research: advances and future directions

With the increasing application of artificial intelligence (AI) in medical research, studies on the human immunodeficiency virus type 1(HIV-1) and acquired immunodeficiency syndrome (AIDS) have become more in-depth. Integrating AI with technologies like single-cell sequencing enables precise biomarker identification and improved therapeutic targeting. This review aims to explore the advancements in AI technologies and their applications across various facets of HIV research, including viral mechanisms, diagnostic innovations, therapeutic strategies, and prevention efforts. Despite challenges like data limitations and model interpretability, AI holds significant potential in advancing HIV-1 management and contributing to global health goals.

Survival machine learning model of T1 colorectal postoperative recurrence after endoscopic resection and surgical operation: a retrospective cohort study

OBJECTIVE

To construct a postoperative recurrence prediction model for patients with T1 colorectal cancer after endoscopic resection and surgical operation via survival machine learning algorithms.

METHODS

Based on two tertiary-level affiliated hospitals, case data of 580 patients with T1 colorectal cancer treated by endoscopic resection and surgery were obtained, and patients' personal information, treatment modalities, and pathology-related information were extracted. After Boruta's algorithmic feature selection, predictors with significant contributions were identified. The patients were divided into a train set and a test set at a ratio of 7:3, and five survival machine learning models were subsequently built, namely, Randomized Survival Forest (RSF), Gradient Boosting (GB), Survival Tree (ST), CoxPH and Coxnet. Interpretability analysis of the model is based on the SHAP algorithm.

RESULTS

Patients at high risk of lymph node metastasis have a poor prognosis, but different treatment modalities do not significantly affect the prognosis of patients with recurrence. The Random Survival Forest model shows better performance, with a C-index and Integrated Brier Score of 0.848 and 0.098 in the test set, respectively, and its time-dependent AUC is 0.918. The interpretability analysis of the model revealed that submucosal invasion depth < 1000 μm, tumor budding grade of BD1, lymphovascular invasion and perineural invasion are absent, well differentiated cancer cells, and tumor size < 20 mm have positive effects on the model, lts negative gain characteristics are a contributing factor to patient relapse.

CONCLUSIONS

The prognostic model constructed via survival machine learning for patients with T1 colorectal cancer has good performance, and can provide accurate individualized predictions.

Machine Learning in the Management of Patients Undergoing Catheter Ablation for Atrial Fibrillation: Scoping Review

BACKGROUND

Although catheter ablation (CA) is currently the most effective clinical treatment for atrial fibrillation, its variable therapeutic effects among different patients present numerous problems. Machine learning (ML) shows promising potential in optimizing the management and clinical outcomes of patients undergoing atrial fibrillation CA (AFCA).

OBJECTIVE

This scoping review aimed to evaluate the current scientific evidence on the application of ML for managing patients undergoing AFCA, compare the performance of various models across specific clinical tasks within AFCA, and summarize the strengths and limitations of ML in this field.

METHODS

Adhering to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines, relevant studies published up to October 7, 2023, were searched from PubMed, Web of Science, Embase, the Cochrane Library, and ScienceDirect. The final included studies were confirmed based on inclusion and exclusion criteria and manual review. The PROBAST (Prediction model Risk Of Bias Assessment Tool) and QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) methodological quality assessment tools were used to review the included studies, and narrative data synthesis was performed on the modeled results provided by these studies.

RESULTS

The analysis of 23 included studies showcased the contributions of ML in identifying potential ablation targets, improving ablation strategies, and predicting patient prognosis. The patient data used in these studies comprised demographics, clinical characteristics, various types of imaging (9/23, 39%), and electrophysiological signals (7/23, 30%). In terms of model type, deep learning, represented by convolutional neural networks, was most frequently applied (14/23, 61%). Compared with traditional clinical scoring models or human clinicians, the model performance reported in the included studies was generally satisfactory, but most models (14/23, 61%) showed a high risk of bias due to lack of external validation.

CONCLUSIONS

Our evidence-based findings suggest that ML is a promising tool for improving the effectiveness and efficiency of managing patients undergoing AFCA. While guiding data preparation and model selection for future studies, this review highlights the need to address prevalent limitations, including lack of external validation, and to further explore model generalization and interpretability.

Optimizing Cancer Treatment: Exploring the Role of AI in Radioimmunotherapy

Radioimmunotherapy (RIT) is a novel cancer treatment that combines radiotherapy and immunotherapy to precisely target tumor antigens using monoclonal antibodies conjugated with radioactive isotopes. This approach offers personalized, systemic, and durable treatment, making it effective in cancers resistant to conventional therapies. Advances in artificial intelligence (AI) present opportunities to enhance RIT by improving precision, efficiency, and personalization. AI plays a critical role in patient selection, treatment planning, dosimetry, and response assessment, while also contributing to drug design and tumor classification. This review explores the integration of AI into RIT, emphasizing its potential to optimize the entire treatment process and advance personalized cancer care.

Decoding pan-cancer treatment outcomes using multimodal real-world data and explainable artificial intelligence

Despite advances in precision oncology, clinical decision-making still relies on limited variables and expert knowledge. To address this limitation, we combined multimodal real-world data and explainable artificial intelligence (xAI) to introduce AI-derived (AID) markers for clinical decision support. We used xAI to decode the outcome of 15,726 patients across 38 solid cancer entities based on 350 markers, including clinical records, image-derived body compositions, and mutational tumor profiles. xAI determined the prognostic contribution of each clinical marker at the patient level and identified 114 key markers that accounted for 90% of the neural network's decision process. Moreover, xAI enabled us to uncover 1,373 prognostic interactions between markers. Our approach was validated in an independent cohort of 3,288 patients with lung cancer from a US nationwide electronic health record-derived database. These results show the potential of xAI to transform the assessment of clinical variables and enable personalized, data-driven cancer care.

Deep Learning Approach Predicts Longitudinal Retinal Nerve Fiber Layer Thickness Changes

This study aims to develop deep learning (DL) models to predict the retinal nerve fiber layer (RNFL) thickness changes in glaucoma, facilitating the early diagnosis and monitoring of disease progression. Using the longitudinal data from two glaucoma studies (Diagnostic Innovations in Glaucoma Study (DIGS) and African Descent and Glaucoma Evaluation Study (ADAGES)), we constructed models using optical coherence tomography (OCT) scans from 251 participants (437 eyes). The models were trained to predict the RNFL thickness at a future visit based on previous scans. We evaluated four models: linear regression (LR), support vector regression (SVR), gradient boosting regression (GBR), and a custom 1D convolutional neural network (CNN). The GBR model achieved the best performance in predicting pointwise RNFL thickness changes (MAE = 5.2 μm, R2 = 0.91), while the custom 1D CNN excelled in predicting changes to average global and sectoral RNFL thickness, providing greater resolution and outperforming the traditional models (MAEs from 2.0-4.2 μm, R2 from 0.94-0.98). Our custom models used a novel approach that incorporated longitudinal OCT imaging to achieve consistent performance across different demographics and disease severities, offering potential clinical decision support for glaucoma diagnosis. Patient-level data splitting enhances the evaluation robustness, while predicting detailed RNFL thickness provides a comprehensive understanding of the structural changes over time.

A systematic review on the integration of explainable artificial intelligence in intrusion detection systems to enhancing transparency and interpretability in cybersecurity

The rise of sophisticated cyber threats has spurred advancements in Intrusion Detection Systems (IDS), which are crucial for identifying and mitigating security breaches in real-time. Traditional IDS often rely on complex machine learning algorithms that lack transparency despite their high accuracy, creating a "black box" effect that can hinder the analysts' understanding of their decision-making processes. Explainable Artificial Intelligence (XAI) offers a promising solution by providing interpretability and transparency, enabling security professionals to understand better, trust, and optimize IDS models. This paper presents a systematic review of the integration of XAI in IDS, focusing on enhancing transparency and interpretability in cybersecurity. Through a comprehensive analysis of recent studies, this review identifies commonly used XAI techniques, evaluates their effectiveness within IDS frameworks, and examines their benefits and limitations. Findings indicate that rule-based and tree-based XAI models are preferred for their interpretability, though trade-offs with detection accuracy remain challenging. Furthermore, the review highlights critical gaps in standardization and scalability, emphasizing the need for hybrid models and real-time explainability. The paper concludes with recommendations for future research directions, suggesting improvements in XAI techniques tailored for IDS, standardized evaluation metrics, and ethical frameworks prioritizing security and transparency. This review aims to inform researchers and practitioners about current trends and future opportunities in leveraging XAI to enhance IDS effectiveness, fostering a more transparent and resilient cybersecurity landscape.

A systematic review of the impact of artificial intelligence on educational outcomes in health professions education

BACKGROUND

Artificial intelligence (AI) has a variety of potential applications in health professions education and assessment; however, measurable educational impacts of AI-based educational strategies on learning outcomes have not been systematically evaluated.

METHODS

A systematic literature search was conducted using electronic databases (CINAHL Plus, EMBASE, Proquest, Pubmed, Cochrane Library, and Web of Science) to identify studies published until October 1st 2024, analyzing the impact of AI-based tools/interventions in health profession assessment and/or training on educational outcomes. The present analysis follows the PRISMA 2020 statement for systematic reviews and the structured approach to reporting in health care education for evidence synthesis.

RESULTS

The final analysis included twelve studies. All were single centers with sample sizes ranging from 4 to 180 participants. Three studies were randomized controlled trials, and seven had a quasi-experimental design. Two studies were observational. The studies had a heterogenous design. Confounding variables were not controlled. None of the studies provided learning objectives or descriptions of the competencies to be achieved. Three studies applied learning theories in the development of AI-powered educational strategies. One study reported the analysis of the authenticity of the learning environment. No study provided information on the impact of feedback activities on learning outcomes. All studies corresponded to Kirkpatrick's second level evaluating technical skills or quantifiable knowledge. No study evaluated more complex tasks, such as the behavior of learners in the workplace. There was insufficient information on training datasets and copyright issues.

CONCLUSIONS

The results of the analysis show that the current evidence regarding measurable educational outcomes of AI-powered interventions in health professions education is poor. Further studies with a rigorous methodological approach are needed. The present work also highlights that there is no straightforward guide for evaluating the quality of research in AI-based education and suggests a series of criteria that should be considered.

TRIAL REGISTRATION

Methods and inclusion criteria were defined in advance, specified in a protocol and registered in the OSF registries ( https://osf.io/v5cgp/ ).

CLINICAL TRIAL NUMBER

not applicable.

Assessing the knowledge of ChatGPT and Google Gemini in answering peripheral artery disease-related questions

INTRODUCTION

To assess and compare the knowledge of ChatGPT and Google Gemini in answering public-based and scientific questions about peripheral artery disease (PAD).

METHODS

Frequently asked questions (FAQs) about PAD were generated by evaluating posts on social media, and the latest edition of the European Society of Cardiology (ESC) guideline was evaluated and recommendations about PAD were translated into questions. All questions were prepared in English and were asked to ChatGPT 4 and Google Gemini (formerly Google Bard) applications. The specialists assigned a Global Quality Score (GQS) for each response.

RESULTS

Finally, 72 FAQs and 63 ESC guideline-based questions were identified. In total, 51 (70.8%) answers by ChatGPT for FAQs were categorized as GQS 5. Moreover, 44 (69.8%) ChatGPT answers to ESC guideline-based questions about PAD scored GQS 5. A total of 40 (55.6%) answers by Google Gemini for FAQs related with PAD obtained GQS 5. In addition, 50.8% (32 of 63) Google Gemini answers to ESC guideline-based questions were classified as GQS 5. Comparison of ChatGPT and Google Gemini with regards to GQS score revealed that both for FAQs about PAD, and ESC guideline-based scientific questions about PAD, ChatGPT gave more accurate and satisfactory answers (p = 0.031 and p = 0.026). In contrast, response time was significantly shorter for Google Gemini for both FAQs and scientific questions about PAD (p = 0.008 and p = 0.001).

CONCLUSION

Our findings revealed that both ChatGPT and Google Gemini had limited capacity to answer FAQs and scientific questions related with PDA, but accuracy and satisfactory rate of answers for both FAQs and scientific questions about PAD were significantly higher in favor of ChatGPT.

Sequencing and characterizing human mitochondrial genomes in the biobank-based genomic research paradigm

Human mitochondrial DNA (mtDNA) harbors essential mutations linked to aging, neurodegenerative diseases, and complex muscle disorders. Due to its uniparental and haploid inheritance, mtDNA captures matrilineal evolutionary trajectories, playing a crucial role in population and medical genetics. However, critical questions about the genomic diversity patterns, inheritance models, and evolutionary and medical functions of mtDNA remain unresolved or underexplored, particularly in the transition from traditional genotyping to large-scale genomic analyses. This review summarizes recent advancements in data-driven genomic research and technological innovations that address these questions and clarify the biological impact of nuclear-mitochondrial segments (NUMTs) and mtDNA variants on human health, disease, and evolution. We propose a streamlined pipeline to comprehensively identify mtDNA and NUMT genomic diversity using advanced sequencing and computational technologies. Haplotype-resolved mtDNA sequencing and assembly can distinguish authentic mtDNA variants from NUMTs, reduce diagnostic inaccuracies, and provide clearer insights into heteroplasmy patterns and the authenticity of paternal inheritance. This review emphasizes the need for integrative multi-omics approaches and emerging long-read sequencing technologies to gain new insights into mutation mechanisms, the influence of heteroplasmy and paternal inheritance on mtDNA diversity and disease susceptibility, and the detailed functions of NUMTs.

Machine Learning for the Early Prediction of Delayed Cerebral Ischemia in Patients With Subarachnoid Hemorrhage: Systematic Review and Meta-Analysis

BACKGROUND

Delayed cerebral ischemia (DCI) is a primary contributor to death after subarachnoid hemorrhage (SAH), with significant incidence. Therefore, early determination of the risk of DCI is an urgent need. Machine learning (ML) has received much attention in clinical practice. Recently, some studies have attempted to apply ML models for early noninvasive prediction of DCI. However, systematic evidence for its predictive accuracy is still lacking.

OBJECTIVE

The aim of this study was to synthesize the prediction accuracy of ML models for DCI to provide evidence for the development or updating of intelligent detection tools.

METHODS

PubMed, Cochrane, Embase, and Web of Science databases were systematically searched up to May 18, 2023. The risk of bias in the included studies was assessed using PROBAST (Prediction Model Risk of Bias Assessment Tool). During the analysis, we discussed the performance of different models in the training and validation sets.

RESULTS

We finally included 48 studies containing 16,294 patients with SAH and 71 ML models with logistic regression as the main model type. In the training set, the pooled concordance index (C index), sensitivity, and specificity of all the models were 0.786 (95% CI 0.737-0.835), 0.77 (95% CI 0.69-0.84), and 0.83 (95% CI 0.75-0.89), respectively, while those of the logistic regression models were 0.770 (95% CI 0.724-0.817), 0.75 (95% CI 0.67-0.82), and 0.71 (95% CI 0.63-0.78), respectively. In the validation set, the pooled C index, sensitivity, and specificity of all the models were 0.767 (95% CI 0.741-0.793), 0.66 (95% CI 0.53-0.77), and 0.78 (95% CI 0.71-0.84), respectively, while those of the logistic regression models were 0.757 (95% CI 0.715-0.800), 0.59 (95% CI 0.57-0.80), and 0.80 (95% CI 0.71-0.87), respectively.

CONCLUSIONS

ML models appear to have relatively desirable power for early noninvasive prediction of DCI after SAH. However, enhancing the prediction sensitivity of these models is challenging. Therefore, efficient, noninvasive, or minimally invasive low-cost predictors should be further explored in future studies to improve the prediction accuracy of ML models.

TRIAL REGISTRATION

PROSPERO (CRD42023438399); https://tinyurl.com/yfuuudde.

Differences in technical and clinical perspectives on AI validation in cancer imaging: mind the gap!

Good practices in artificial intelligence (AI) model validation are key for achieving trustworthy AI. Within the cancer imaging domain, attracting the attention of clinical and technical AI enthusiasts, this work discusses current gaps in AI validation strategies, examining existing practices that are common or variable across technical groups (TGs) and clinical groups (CGs). The work is based on a set of structured questions encompassing several AI validation topics, addressed to professionals working in AI for medical imaging. A total of 49 responses were obtained and analysed to identify trends and patterns. While TGs valued transparency and traceability the most, CGs pointed out the importance of explainability. Among the topics where TGs may benefit from further exposure are stability and robustness checks, and mitigation of fairness issues. On the other hand, CGs seemed more reluctant towards synthetic data for validation and would benefit from exposure to cross-validation techniques, or segmentation metrics. Topics emerging from the open questions were utility, capability, adoption and trustworthiness. These findings on current trends in AI validation strategies may guide the creation of guidelines necessary for training the next generation of professionals working with AI in healthcare and contribute to bridging any technical-clinical gap in AI validation. RELEVANCE STATEMENT: This study recognised current gaps in understanding and applying AI validation strategies in cancer imaging and helped promote trust and adoption for interdisciplinary teams of technical and clinical researchers. KEY POINTS: Clinical and technical researchers emphasise interpretability, external validation with diverse data, and bias awareness in AI validation for cancer imaging. In cancer imaging AI research, clinical researchers prioritise explainability, while technical researchers focus on transparency and traceability, and see potential in synthetic datasets. Researchers advocate for greater homogenisation of AI validation practices in cancer imaging.

A latent diffusion approach to visual attribution in medical imaging

Visual attribution in medical imaging seeks to make evident the diagnostically-relevant components of a medical image, in contrast to the more common detection of diseased tissue deployed in standard machine vision pipelines (which are less straightforwardly interpretable/explainable to clinicians). We here present a novel generative visual attribution technique, one that leverages latent diffusion models in combination with domain-specific large language models, in order to generate normal counterparts of abnormal images. The discrepancy between the two hence gives rise to a mapping indicating the diagnostically-relevant image components. To achieve this, we deploy image priors in conjunction with appropriate conditioning mechanisms in order to control the image generative process, including natural language text prompts acquired from medical science and applied radiology. We perform experiments and quantitatively evaluate our results on the COVID-19 Radiography Database containing labelled chest X-rays with differing pathologies via the Frechet Inception Distance (FID), Structural Similarity (SSIM) and Multi Scale Structural Similarity Metric (MS-SSIM) metrics obtained between real and generated images. The resulting system also exhibits a range of latent capabilities including zero-shot localized disease induction, which are evaluated with real examples from the cheXpert dataset.

Adaptive Treatment of Metastatic Prostate Cancer Using Generative Artificial Intelligence

Despite the expanding therapeutic options available to cancer patients, therapeutic resistance, disease recurrence, and metastasis persist as hallmark challenges in the treatment of cancer. The rise to prominence of generative artificial intelligence (GenAI) in many realms of human activities is compelling the consideration of its capabilities as a potential lever to advance the development of effective cancer treatments. This article presents a hypothetical case study on the application of generative pre-trained transformers (GPTs) to the treatment of metastatic prostate cancer (mPC). The case explores the design of GPT-supported adaptive intermittent therapy for mPC. Testosterone and prostate-specific antigen (PSA) are assumed to be repeatedly monitored while treatment may involve a combination of androgen deprivation therapy (ADT), androgen receptor-signalling inhibitors (ARSI), chemotherapy, and radiotherapy. The analysis covers various questions relevant to the configuration, training, and inferencing of GPTs for the case of mPC treatment with a particular attention to risk mitigation regarding the hallucination problem and its implications to clinical integration of GenAI technologies. The case study provides elements of an actionable pathway to the realization of GenAI-assisted adaptive treatment of metastatic prostate cancer. As such, the study is expected to help facilitate the design of clinical trials of GenAI-supported cancer treatments.

Visual Analysis of Multi-Outcome Causal Graphs

We introduce a visual analysis method for multiple causal graphs with different outcome variables, namely, multi-outcome causal graphs. Multi-outcome causal graphs are important in healthcare for understanding multimorbidity and comorbidity. To support the visual analysis, we collaborated with medical experts to devise two comparative visualization techniques at different stages of the analysis process. First, a progressive visualization method is proposed for comparing multiple state-of-the-art causal discovery algorithms. The method can handle mixed-type datasets comprising both continuous and categorical variables and assist in the creation of a fine-tuned causal graph of a single o utcome. Second, a comparative graph layout technique and specialized visual encodings are devised for the quick comparison of multiple causal graphs. In our visual analysis approach, analysts start by building individual causal graphs for each outcome variable, and then, multi-outcome causal graphs are generated and visualized with our comparative technique for analyzing differences and commonalities of these causal graphs. Evaluation includes quantitative measurements on benchmark datasets, a case study with a medical expert, and expert user studies with real-world health research data.

Promising tools for future drug discovery and development in antiarrhythmic therapy

Arrhythmia refers to irregularities in the rate and rhythm of the heart, with symptoms spanning from mild palpitations to life-threatening arrhythmias and sudden cardiac death. The complex molecular nature of arrhythmias complicates the selection of appropriate treatment. Current therapies involve the use of antiarrhythmic drugs (class I-IV) with limited efficacy and dangerous side effects and implantable pacemakers and cardioverter-defibrillators with hardware-related complications and inappropriate shocks. The number of novel antiarrhythmic drugs in the development pipeline has decreased substantially during the last decade and underscores uncertainties regarding future developments in this field. Consequently, arrhythmia treatment poses significant challenges, prompting the need for alternative approaches. Remarkably, innovative drug discovery and development technologies show promise in helping advance antiarrhythmic therapies. In this article, we review unique characteristics and the transformative potential of emerging technologies that offer unprecedented opportunities for transitioning from traditional antiarrhythmics to next-generation therapies. We assess stem cell technology, emphasizing the utility of innovative cell profiling using multiomics, high-throughput screening, and advanced computational modeling in developing treatments tailored precisely to individual genetic and physiological profiles. We offer insights into gene therapy, peptide, and peptibody approaches for drug delivery. We finally discuss potential strengths and weaknesses of such techniques in reducing adverse effects and enhancing overall treatment outcomes, leading to more effective, specific, and safer therapies. Altogether, this comprehensive overview introduces innovative avenues for personalized rhythm therapy, with particular emphasis on drug discovery, aiming to advance the arrhythmia treatment landscape and the prevention of sudden cardiac death. SIGNIFICANCE STATEMENT: Arrhythmias and sudden cardiac death account for 15%-20% of deaths worldwide. However, current antiarrhythmic therapies are ineffective and have dangerous side effects. Here, we review the field of arrhythmia treatment underscoring the slow progress in advancing the cardiac rhythm therapy pipeline and the uncertainties regarding evolution of this field. We provide information on how emerging technological and experimental tools can help accelerate progress and address the limitations of antiarrhythmic drug discovery.

Leveraging artificial intelligence for predicting spontaneous closure of perimembranous ventricular septal defect in children: a multicentre, retrospective study in China

BACKGROUND

Perimembranous ventricular septal defect (PMVSD) is a prevalent congenital heart disease, presenting challenges in predicting spontaneous closure, which is crucial for therapeutic decisions. Existing models mainly rely on structured echocardiographic parameters or restricted data. This study introduces an artificial intelligence (AI)-based model, which uses natural language processing (NLP) and machine learning with the aim of improving spontaneous closure predictability in PMVSD.

METHODS

We did a multicentre, retrospective analysis using data from 29 142 PMVSD patients across six tertiary centres in China from May, 2004, to September, 2022, for training (70%) and validation (30%; dataset 1, 27 269 patients), and from September, 2001, to December, 2009 for testing (dataset 2, 1873 patients). NLP extracted structured data from echocardiography reports and medical records, which were used to develop machine learning models. Models were evaluated for spontaneous closure occurrence and timing by use of area under the receiver operating characteristic curve (AUC), decision curve analysis, and calibration index.

FINDINGS

Spontaneous closure occurred in 3520 patients (12·1%) at a median of 31 months (IQR 16-56). Eleven NLP-derived predictors, identified via least absolute shrinkage and selection operator, highlighted the importance of defect morphology and patient age. The random survival forest algorithm, selected for its superior concordance indexes, showed excellent predictive performance with validation set AUCs (95% CI) of 0·95 (0·94-0·96) for 1-year and 3-year predictions, and 0·95 (0·95-0·96) for 5-year predictions; testing set AUCs were 0·95 (0·94-0·97) for 1-year predictions, 0·97 (0·96-0·98) for 3-year predictions, and 0·98 (0·97-0·99) for 5-year predictions. The model showed high clinical utility through decision curve analysis, calibration, and risk stratification, maintaining consistent accuracy across centres and subgroups.

INTERPRETATION

This AI-based model for predicting spontaneous closure in PMVSD patients represents a substantial advancement, potentially improving patient management, reducing risks of delayed or inappropriate treatment, and enhancing clinical outcomes.

FUNDING

National Natural Science Foundation of China, Shanghai Municipal Hospital Clinical Technology Project, Shanghai Municipal Health Commission, and Clinical Research Unit of XinHua Hospital.

Artificial Intelligence in Pediatric Electrocardiography: A Comprehensive Review

Artificial intelligence (AI) is revolutionizing healthcare by offering innovative solutions for diagnosis, treatment, and patient management. Only recently has the field of pediatric cardiology begun to explore the use of deep learning methods to analyze electrocardiogram (ECG) data, aiming to enhance diagnostic accuracy, expedite workflows, and improve patient outcomes. This review examines the current state of AI-enhanced ECG interpretation in pediatric cardiology applications, drawing insights from adult AI-ECG research given the progress in this field. It describes a broad range of AI methodologies, investigates the unique challenges inherent in pediatric ECG analysis, reviews the current state of the literature in pediatric AI-ECG, and discusses potential future directions for research and clinical practice. While AI-ECG applications have demonstrated considerable promise, widespread clinical adoption necessitates further research, rigorous validation, and careful consideration of equity, ethical, legal, and practical challenges.

An AI-Cyborg System for Adaptive Intelligent Modulation of Organoid Maturation

Recent advancements in flexible bioelectronics have enabled continuous, long-term stable interrogation and intervention of biological systems. However, effectively utilizing the interrogated data to modulate biological systems to achieve specific biomedical and biological goals remains a challenge. In this study, we introduce an AI-driven bioelectronics system that integrates tissue-like, flexible bioelectronics with cyber learning algorithms to create a long-term, real-time bidirectional b ioelectronic interface with o ptimized a daptive intelligent m odulation (BIO-AIM). When integrated with biological systems as an AI-cyborg system, BIO-AIM continuously adapts and optimizes stimulation parameters based on stable cell state mapping, allowing for real-time, closed-loop feedback through tissue-embedded flexible electrode arrays. Applied to human pluripotent stem cell-derived cardiac organoids, BIO-AIM identifies optimized stimulation conditions that accelerate functional maturation. The effectiveness of this approach is validated through enhanced extracellular spike waveforms, increased conduction velocity, and improved sarcomere organization, outperforming both fixed and no stimulation conditions.

From pixels to patients: the evolution and future of deep learning in cancer diagnostics

Deep learning has revolutionized cancer diagnostics, shifting from pixel-based image analysis to more comprehensive, patient-centric care. This opinion article explores recent advancements in neural network architectures, highlighting their evolution in biomedical research and their impact on medical imaging interpretation and multimodal data integration. We emphasize the need for domain-specific artificial intelligence (AI) systems capable of handling complex clinical tasks, advocating for the development of multimodal large language models that can integrate diverse data sources. These models have the potential to significantly enhance the precision and efficiency of cancer diagnostics, transforming AI from a supplementary tool into a core component of clinical decision-making, ultimately improving patient outcomes and advancing cancer care.

Student Pharmacists' Perceptions of Artificial Intelligence and Machine Learning in Pharmacy Practice and Pharmacy Education

OBJECTIVE

This study explored student pharmacists' perceptions and attitudes regarding artificial intelligence (AI) and machine learning (ML) in pharmacy practice. Due to AI/ML's promising prospects, understanding students' current awareness, comprehension, and hopes for their use in this field is essential.

METHODS

In April 2024, a Zoom focus group discussion was conducted with 6 student pharmacists using a self-developed interview guide. The guide included questions about the benefits, challenges, and ethical considerations of implementing AI/ML in pharmacy practice and education. The participants' demographic information was collected through a questionnaire. The research team conducted a thematic analysis of the discussion transcript. The results generated by a team member using NVivo were compared with those generated by ChatGPT, and all discrepancies were addressed.

RESULTS

Student pharmacists displayed a generally positive attitude toward the implementation of AI/ML in pharmacy practice but lacked knowledge about AI/ML applications. Participants recognized several advantages of AI/ML implementation in pharmacy practice, including improved accuracy and time-saving for pharmacists. Some identified challenges were alert fatigue, AI/ML-generated errors, and the potential obstacle to person-centered care. The study participants expressed their interest in learning about AI/ML and their desire to integrate these technologies into pharmacy education.

CONCLUSION

The demand for integrating AI/ML into pharmacy practice is increasing. Student and professional pharmacists need additional AI/ML training to equip them with knowledge and practical skills. Collaboration between pharmacists, institutions, and AI/ML companies is essential to address barriers and advance AI/ML implementation in the pharmacy field.

Causal Explanation from Mild Cognitive Impairment Progression using Graph Neural Networks

Mild Cognitive Impairment (MCI) is a transitional stage between normal cognitive aging and dementia. Some individuals with MCI revert to normal, while others progress to dementia. There are limited studies using explainable artificial intelligence on longitudinal data, particularly including genotypes, biomarkers and chronic diseases, to explore these differences. This study introduces a novel approach to understanding MCI progression using explainable graph neural networks. Utilizing longitudinal temporal data, we constructed a comprehensive graph representation of each individual in the study cohort. Our temporal graph convolutional network achieved 72.4% accuracy in predicting MCI transitions, while our causal explanation method outperformed existing explanation techniques in stability, accuracy, and faithfulness. We identified a causal subgraph with informative variables including hypertension, arrhythmia, congestive heart failure, coronary artery disease, stroke, lipid-related issues, and sex.

Integrating Interpretability in Machine Learning and Deep Neural Networks: A Novel Approach to Feature Importance and Outlier Detection in COVID-19 Symptomatology and Vaccine Efficacy

In this study, we introduce a novel approach that integrates interpretability techniques from both traditional machine learning (ML) and deep neural networks (DNN) to quantify feature importance using global and local interpretation methods. Our method bridges the gap between interpretable ML models and powerful deep learning (DL) architectures, providing comprehensive insights into the key drivers behind model predictions, especially in detecting outliers within medical data. We applied this method to analyze COVID-19 pandemic data from 2020, yielding intriguing insights. We used a dataset consisting of individuals who were tested for COVID-19 during the early stages of the pandemic in 2020. The dataset included self-reported symptoms and test results from a wide demographic, and our goal was to identify the most important symptoms that could help predict COVID-19 infection accurately. By applying interpretability techniques to both machine learning and deep learning models, we aimed to improve understanding of symptomatology and enhance early detection of COVID-19 cases. Notably, even though less than 1% of our cohort reported having a sore throat, this symptom emerged as a significant indicator of active COVID-19 infection, appearing 7 out of 9 times in the top four most important features across all methodologies. This suggests its potential as an early symptom marker. Studies have shown that individuals reporting sore throat may have a compromised immune system, where antibody generation is not functioning correctly. This aligns with our data, which indicates that 5% of patients with sore throats required hospitalization. Our analysis also revealed a concerning trend of diminished immune response post-COVID infection, increasing the likelihood of severe cases requiring hospitalization. This finding underscores the importance of monitoring patients post-recovery for potential complications and tailoring medical interventions accordingly. Our study also raises critical questions about the efficacy of COVID-19 vaccines in individuals presenting with sore throat as a symptom. The results suggest that booster shots might be necessary for this population to ensure adequate immunity, given the observed immune response patterns. The proposed method not only enhances our understanding of COVID-19 symptomatology but also demonstrates its broader utility in medical outlier detection. This research contributes valuable insights to ongoing efforts in creating interpretable models for COVID-19 management and vaccine optimization strategies. By leveraging feature importance and interpretability, these models empower physicians, healthcare workers, and researchers to understand complex relationships within medical data, facilitating more informed decision-making for patient care and public health initiatives.

Artificial Intelligence Applications in Lymphoma Diagnosis and Management: Opportunities, Challenges, and Future Directions

Lymphoma, a heterogeneous group of blood cancers, presents significant diagnostic and therapeutic challenges due to its complex subtypes and variable clinical outcomes. Artificial intelligence (AI) has emerged as a promising tool to enhance the accuracy and efficiency of lymphoma pathology. This review explores the potential of AI in lymphoma diagnosis, classification, prognosis prediction, and treatment planning, as well as addressing the challenges and future directions in this rapidly evolving field.

Constructing and exploring neuroimaging projects: a survey from clinical practice to scientific research

Over the past decades, numerous large-scale neuroimaging projects that involved the collection and release of multimodal data have been conducted globally. Distinguished initiatives such as the Human Connectome Project, UK Biobank, and Alzheimer's Disease Neuroimaging Initiative, among others, stand as remarkable international collaborations that have significantly advanced our understanding of the brain. With the advancement of big data technology, changes in healthcare models, and continuous development in biomedical research, various types of large-scale projects are being established and promoted worldwide. For project leaders, there is a need to refer to common principles in project construction and management. Users must also adhere strictly to rules and guidelines, ensuring data safety and privacy protection. Organizations must maintain data integrity, protect individual privacy, and foster stakeholders' trust. Regular updates to legislation and policies are necessary to keep pace with evolving technologies and emerging data-related challenges. CRITICAL RELEVANCE STATEMENT: By reviewing global large-scale neuroimaging projects, we have summarized the standards and norms for establishing and utilizing their data, and provided suggestions and opinions on some ethical issues, aiming to promote higher-quality neuroimaging data development. KEY POINTS: Global neuroimaging projects are increasingly advancing but still face challenges. Constructing and utilizing neuroimaging projects should follow set rules and guidelines. Effective data management and governance should be developed to support neuroimaging projects.

The Potential of Artificial Intelligence Tools for Reducing Uncertainty in Medicine and Directions for Medical Education

Unlabelled

In the field of medicine, uncertainty is inherent. Physicians are asked to make decisions on a daily basis without complete certainty, whether it is in understanding the patient's problem, performing the physical examination, interpreting the findings of diagnostic tests, or proposing a management plan. The reasons for this uncertainty are widespread, including the lack of knowledge about the patient, individual physician limitations, and the limited predictive power of objective diagnostic tools. This uncertainty poses significant problems in providing competent patient care. Research efforts and teaching are attempts to reduce uncertainty that have now become inherent to medicine. Despite this, uncertainty is rampant. Artificial intelligence (AI) tools, which are being rapidly developed and integrated into practice, may change the way we navigate uncertainty. In their strongest forms, AI tools may have the ability to improve data collection on diseases, patient beliefs, values, and preferences, thereby allowing more time for physician-patient communication. By using methods not previously considered, these tools hold the potential to reduce the uncertainty in medicine, such as those arising due to the lack of clinical information and provider skill and bias. Despite this possibility, there has been considerable resistance to the implementation of AI tools in medical practice. In this viewpoint article, we discuss the impact of AI on medical uncertainty and discuss practical approaches to teaching the use of AI tools in medical schools and residency training programs, including AI ethics, real-world skills, and technological aptitude.

Prediction of defensive success in elite soccer using machine learning - Tactical analysis of defensive play using tracking data and explainable AI

The interest in sports performance analysis is rising and tracking data holds high potential for game analysis in team sports due to its accuracy and informative content. Together with machine learning approaches one can obtain deeper and more objective insights into the performance structure. In soccer, the analysis of the defense was neglected in comparison to the offense. Therefore, the aim of this study is to predict ball gains in defense using tracking data to identify tactical variables that drive defensive success. We evaluated tracking data of 153 games of German Bundesliga season 2020/21. With it, we derived player (defensive pressure, distance to the ball, & velocity) and team metrics (inter-line distances, numerical superiority, surface area, & spread) each containing a tactical idea. Afterwards, we trained supervised machine learning classifiers (logistic regression, XGBoost, & Random Forest Classifier) to predict successful (ball gain) vs. unsuccessful defensive plays (no ball gain). The expert-reduction-model (Random Forest Classifier with 16 features) showed the best and satisfying prediction performance (F1-Score (test) = 0.57). Analyzing the most important input features of this model, we are able to identify tactical principles of defensive play that appear to be related to gaining the ball: press the ball leading player, create numerical superiority in areas close to the ball (press short pass options), compact organization of defending team. Those principles are highly interesting for practitioners to gain valuable insights in the tactical behavior of soccer players that may be related to the success of defensive play.

Disentangled Explanations of Neural Network Predictions by Finding Relevant Subspaces

Explainable AI aims to overcome the black-box nature of complex ML models like neural networks by generating explanations for their predictions. Explanations often take the form of a heatmap identifying input features (e.g. pixels) that are relevant to the model's decision. These explanations, however, entangle the potentially multiple factors that enter into the overall complex decision strategy. We propose to disentangle explanations by extracting at some intermediate layer of a neural network, subspaces that capture the multiple and distinct activation patterns (e.g. visual concepts) that are relevant to the prediction. To automatically extract these subspaces, we propose two new analyses, extending principles found in PCA or ICA to explanations. These novel analyses, which we call principal relevant component analysis (PRCA) and disentangled relevant subspace analysis (DRSA), maximize relevance instead of e.g. variance or kurtosis. This allows for a much stronger focus of the analysis on what the ML model actually uses for predicting, ignoring activations or concepts to which the model is invariant. Our approach is general enough to work alongside common attribution techniques such as Shapley Value, Integrated Gradients, or LRP. Our proposed methods show to be practically useful and compare favorably to the state of the art as demonstrated on benchmarks and three use cases.

New Paradigms in the Clinical Management of Li-Fraumeni Syndrome

Approximately 8.5%-16.2% of childhood cancers are associated with a pathogenic/likely pathogenic germline variant-a prevalence that is likely to rise with improvements in phenotype recognition, sequencing, and variant validation. One highly informative, classical hereditary cancer predisposition syndrome is Li-Fraumeni syndrome (LFS), associated with germline variants in the TP53 tumor suppressor gene, and a >90% cumulative lifetime cancer risk. In seeking to improve outcomes for young LFS patients, we must improve the specificity and sensitivity of existing cancer surveillance programs and explore how to complement early detection strategies with pharmacology-based risk-reduction interventions. Here, we describe novel precision screening technologies and clinical strategies for cancer risk reduction. In particular, we summarize the biomarkers for early diagnosis and risk stratification of LFS patients from birth, noninvasive and machine learning-based cancer screening, and drugs that have shown the potential to be repurposed for cancer prevention.

Exploring home healthcare clinicians' needs for using clinical decision support systems for early risk warning

OBJECTIVES

To explore home healthcare (HHC) clinicians' needs for Clinical Decision Support Systems (CDSS) information delivery for early risk warning within HHC workflows.

METHODS

Guided by the CDS "Five-Rights" framework, we conducted semi-structured interviews with multidisciplinary HHC clinicians from April 2023 to August 2023. We used deductive and inductive content analysis to investigate informants' responses regarding CDSS information delivery.

RESULTS

Interviews with thirteen HHC clinicians yielded 16 codes mapping to the CDS "Five-Rights" framework (right information, right person, right format, right channel, right time) and 11 codes for unintended consequences and training needs. Clinicians favored risk levels displayed in color-coded horizontal bars, concrete risk indicators in bullet points, and actionable instructions in the existing EHR system. They preferred non-intrusive risk alerts requiring mandatory confirmation. Clinicians anticipated risk information updates aligned with patient's condition severity and their visit pace. Additionally, they requested training to understand the CDSS's underlying logic, and raised concerns about information accuracy and data privacy.

DISCUSSION

While recognizing CDSS's value in enhancing early risk warning, clinicians highlighted concerns about increased workload, alert fatigue, and CDSS misuse. The top risk factors identified by machine learning algorithms, especially text features, can be ambiguous due to a lack of context. Future research should ensure that CDSS outputs align with clinical evidence and are explainable.

CONCLUSION

This study identified HHC clinicians' expectations, preferences, adaptations, and unintended uses of CDSS for early risk warning. Our findings endorse operationalizing the CDS "Five-Rights" framework to optimize CDSS information delivery and integration into HHC workflows.

Progress Achieved, Landmarks, and Future Concerns in Biomedical and Health Informatics

BACKGROUND

The biomedical and health informatics (BMHI) fields have been advancing rapidly, a trend particularly emphasised during the recent COVID-19 pandemic, introducing innovations in BMHI. Over nearly 50 years since its establishment as a scientific discipline, BMHI has encountered several challenges, such as mishaps, delays, failures, and moments of enthusiastic expectations and notable successes. This paper focuses on reviewing the progress made in the BMHI discipline, evaluating key milestones, and discussing future challenges.

METHODS

To, Structured, step-by-step qualitative methodology was developed and applied, centred on gathering expert opinions and analysing trends from the literature to provide a comprehensive assessment. Experts and pioneers in the BMHI field were assigned thematic tasks based on the research question, providing critical inputs for the thematic analysis. This led to the identification of five key dimensions used to present the findings in the paper: informatics in biomedicine and healthcare, health data in Informatics, nurses in informatics, education and accreditation in health informatics, and ethical, legal, social, and security issues.

RESULTS

Each dimension is examined through recently emerging innovations, linking them directly to the future of healthcare, like the role of artificial intelligence, innovative digital health tools, the expansion of telemedicine, and the use of mobile health apps and wearable devices. The new approach of BMHI covers newly introduced clinical needs and approaches like patient-centric, remote monitoring, and precision medicine clinical approaches.

CONCLUSIONS

These insights offer clear recommendations for improving education and developing experts to advance future innovations. Notably, this narrative review presents a body of knowledge essential for a deep understanding of the BMHI field from a human-centric perspective and, as such, could serve as a reference point for prospective analysis and innovation development.