Klang, Eyal

Abstract Most clinical large language model (LLM) benchmarks rely on clean, concise vignettes that do not reflect the noisy, long-form documentation typical of real clinical records. How LLM performance degrades under realistic chart conditions remains poorly characterised. Here we test whether structured retrieval workflows protect National Institutes of Health Stroke Scale (NIHSS) scoring accuracy under systematic context stress. Using 100 de-identified acute stroke cases and a fully crossed 4 x 4 x 3 x 3 condition matrix (144 conditions per case), we vary context acquisition method, document length, distractor load and critical-information position across four Gemini models (57,047 retained runs). Structured retrieval reduces mean absolute error (MAE) from 4.58 to 2.96 points relative to non-agentic baselines (mean gain 1.62 MAE points; 95% CI 1.57–1.67; 35% relative reduction), with consistent gains across all 36 stress combinations. Lower-cost models show disproportionately larger gains (2.76 versus 0.45 MAE points). Tool-retrieved pipelines outperform retrieval-augmented generation in 33 of 36 combinations. These findings indicate that retrieval architecture, rather than model scale alone, is a tractable lever for robust, equitable clinical LLM deployment.

Background Large language models (LLMs) are entering clinical workflows, yet their alignment with medical-ethical principles is unclear. Objective To evaluate how LLMs reconcile deontological and utilitarian reasoning in medically adapted “Trolley dilemma” scenarios. Methods We tested 14 LLMs, including GPT-o1-preview and DeepSeek-R1-Distill-Llama, using five medically adapted Trolley dilemmas sourced from philosophical and real-world scenarios. Each model generated 10 independent answers per case (overall 700 responses). Responses were forced to “yes” (utilitarian) or “no” (deontological). The primary outcome was the proportion of utilitarian choices. χ 2 tests compared models. Results LLM responses varied. Some models, such as GPT-4 and Gemma, selected the utilitarian option in 20% of cases, while others, like Qwen-2-7B, reached 80% ( p < .001). GPT-o1-preview and DeepSeek-R1-Distill-Llama selected utilitarian responses in 44% and 38% of cases, respectively. Some models selected the consequence-maximizing option in vignettes where doing so required overriding consent or endorsing intentional harm within the scenario framing. Seven of the fourteen models tested recommended impermissible medical actions, including nonconsensual limb amputation, killing a healthy person for organ harvesting, and transferring a patient to hospice against their wishes, in 44 of 700 outputs (6.3 %). Conclusions In this constrained forced-choice stress test, models occasionally endorsed boundary-violating actions within the vignette framing. These results do not generalize to routine clinical decision-making, but motivate further benchmarked evaluation of refusal behavior and boundary constraints in ethically sensitive medical contexts. We must carefully define how these models influence clinical decisions, establish robust ethical guidelines for their use, and ensure that fundamental ethical norms are not compromised.

Background Artificial intelligence (AI)-generated text to video is emerging in medical education, but its effectiveness, accuracy and safety remain uncertain. We aimed to synthesise empirical studies evaluating these tools in learner or patient education. Methods A comprehensive search was conducted in MEDLINE/PubMed, Google Scholar, Scopus, Cochrane Review and Web of Science for studies published up to January 2025. Eligible studies evaluated AI-generated text to video for medical or patient education, reporting both quantitative and qualitative outcomes. Two reviewers screened and extracted data. The review adhered to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Results Out of 103 identified studies, five studies met the inclusion criteria: four evaluated patient education and one evaluated physician training. Clinical areas were ophthalmology (2/5), plastic surgery (1/5), dysphagia rehabilitation (1/5) and neurosurgical training (1/5). In ophthalmology, control materials outperformed AI-generated videos on image/script accuracy (p<0.005), with similar script-image alignment. In dysphagia rehabilitation, a randomised trial reported improvements in swallowing function and related outcomes with an AI-assisted video game intervention (p<0.001). A plastic surgery study reported greater user preference for a video avatar tool compared with a text chatbot (63.5% vs 28.1%). Across the reviewed studies, samples were small and CIs were rarely reported. Outcome measures were heterogeneous, and meta-analysis was not feasible. Conclusion AI-generated videos can enhance engagement or selected outcomes in certain contexts, yet concerns about accuracy and inconsistent measurement persist. Current evidence is sparse and mixed. Currently, these tools can complement, rather than replace, standard resources until non-inferiority is demonstrated for the prespecified primary outcomes. PROSPERO registration number CRD42025640042.

Summary: Peripheral Artery Disease (PAD) is a significant global health concern, leading to morbidity through progressive stenosis and eventually occlusion of the lower extremity arterial vasculature. Advanced imaging modalities play a major role in diagnosing PAD, in planning endovascular as well as surgical interventions, and in monitoring post-treatment outcomes. This review highlights the current major imaging techniques, including duplex ultrasound, computed tomography angiography (CTA) and magnetic resonance angiography (MRA). It provides insights into their applications, advantages, limitations and the importance of individualized imaging strategies for optimizing patient outcomes.

Large language models (LLMs) have the potential to transform healthcare by assisting in documentation, diagnosis, patient communication, and medical education. However, their integration into clinical practice remains a challenge. This perspective explores the barriers to implementation by synthesizing recent evidence across five challenge domains: workflow misalignment and diagnostic safety, bias and equity, regulatory and legal governance, technical vulnerabilities such as hallucinations or data poisoning, and the preservation of patient trust and human connection. While the perspective focuses on barriers, LLM capabilities and mitigation strategies are advancing rapidly, raising the likelihood of near-term clinical impact. Drawing on recent empirical studies, we propose a framework for understanding the key technical, ethical, and practical challenges associated with deploying LLMs in clinical environments and provide directions for future research, governance, and responsible deployment.

BACKGROUND: While clinical practice guidelines advocate for multidisciplinary heart team (MDHT) discussions in coronary revascularization, variability in implementation across health care settings remains a challenge. This variability could potentially be addressed by language learning models like ChatGPT, offering decision-making support in diverse health care environments. Our study aims to critically evaluate the concordance between recommendations made by MDHT and those generated by language learning models in coronary revascularization decision-making. METHODS: From March 2023 to July 2023, consecutive coronary angiography cases (n=86) that were referred for revascularization (either percutaneous or surgical) were analyzed using both ChatGPT-3.5 and ChatGPT-4. Case presentation formats included demographics, medical background, detailed description of angiographic findings, and SYNTAX score (Synergy Between Percutaneous Coronary Intervention With Taxus and Cardiac Surgery; I and II), which were presented in 3 different formats. The recommendations of the models were compared with those of an MDHT. RESULTS: ChatGPT-4 showed high concordance with decisions made by the MDHT (accuracy 0.82, sensitivity 0.8, specificity 0.83, and kappa 0.59), while ChatGPT-3.5 (0.67, 0.27, 0.84, and 0.12, respectively) showed lower concordance. Entropy and Fleiss kappa of ChatGPT-4 were 0.09 and 0.9, respectively, indicating high reliability and repeatability. The best correlation between ChatGPT-4 and MDHT was achieved when clinical cases were presented in a detailed context. Specific subgroups of patients yielded high accuracy (>0.9) of ChatGPT-4, including those with left main disease, 3 vessel disease, and diabetic patients. CONCLUSIONS: The present study demonstrates that advanced language learning models like ChatGPT-4 may be able to predict clinical recommendations for coronary artery disease revascularization with reasonable accuracy, especially in specific patient groups, underscoring their potential role as a supportive tool in clinical decision-making.

Multimodal technology is poised to revolutionize clinical practice by integrating artificial intelligence with traditional diagnostic modalities. This evolution traces its roots from Hippocrates’ humoral theory to the use of sophisticated AI-driven platforms that synthesize data across multiple sensory channels. The interplay between historical medical practices and modern technology challenges conventional patient–clinician interactions and redefines diagnostic accuracy. Highlighting applications from neurology to radiology, the potential of multimodal technology emerges, suggesting a future where AI not only supports but enhances human sensory inputs in medical diagnostics. This shift invites the medical community to navigate the ethical, practical, and technological changes reshaping the landscape of clinical medicine.

AbstractLarge language models (LLMs) are transforming the field of natural language processing (NLP). These models offer opportunities for radiologists to make a meaningful impact in their field. NLP is a part of artificial intelligence (AI) that uses computer algorithms to study and understand text data. Recent advances in NLP include the Attention mechanism and the Transformer architecture. Transformer-based LLMs, such as GPT-4 and Gemini, are trained on massive amounts of data and generate human-like text. They are ideal for analysing large text data in academic research and clinical practice in radiology. Despite their promise, LLMs have limitations, including their dependency on the diversity and quality of their training data and the potential for false outputs. Albeit these limitations, the use of LLMs in radiology holds promise and is gaining momentum. By embracing the potential of LLMs, radiologists can gain valuable insights and improve the efficiency of their work. This can ultimately lead to improved patient care.

Abstract Coronary artery disease (CAD) is a leading health challenge worldwide. Exercise stress testing is a foundational non-invasive diagnostic tool. Nonetheless, its variable accuracy prompts the exploration of more reliable methods. Recent advancements in machine learning (ML), including deep learning and natural language processing, have shown potential in refining the interpretation of stress testing data. Adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, we conducted a systematic review of ML applications in stress electrocardiogram (ECG) and stress echocardiography for CAD prognosis. Medical Literature Analysis and Retrieval System Online, Web of Science, and the Cochrane Library were used as databases. We analysed the ML models, outcomes, and performance metrics. Overall, seven relevant studies were identified. Machine-learning applications in stress ECGs resulted in sensitivity and specificity improvements. Some models achieved rates of above 96% in both metrics and reduced false positives by up to 21%. In stress echocardiography, ML models demonstrated an increase in diagnostic precision. Some models achieved specificity and sensitivity rates of up to 92.7 and 84.4%, respectively. Natural language processing applications enabled the categorization of stress echocardiography reports, with accuracy rates nearing 98%. Limitations include a small, retrospective study pool and the exclusion of nuclear stress testing, due to its well-documented status. This review indicates the potential of artificial intelligence applications in refining CAD stress testing assessment. Further development for real-world use is warranted.

Background: The normal limits of nerve conduction studies are commonly determined by testing healthy subjects. However, in comprehensive real-life nerve conduction electrodiagnostic (EDX) evaluations, multiple nerves are tested, including normal nerves, for purposes of comparison with abnormal ones. Objective: This study aims to evaluate the average values of normal nerve conduction studies in a large population and examined the influence of age and sex. Methods: EDX parameters were extracted from an electronic database of studies performed from May 2016 to February 2022. Established normal values were used to determine the classification of a nerve study as normal. Results: We identified 10,648 EDX reports with 5077 normally interpreted nerve conduction studies (47.6%) of which 57% ( n = 2890) were for females. The median age of studies with no abnormalities was 45.1 years (range < 1 to 92) overall and 42.5 years (range: 0.16 –89.5 years) for males and 47.5 years (range:<1 –91.7) for females. Correlations between age and amplitude, latency, and velocity ( p < 0.001) were observed in most nerves. Amplitude correlated negatively with age in adults in all nerves with a mean of –0.44 (range: –0.24 to –0.62). However, in the pediatric population (age < 18 years), amplitude as well as velocity increased significantly with age. In the adult cohort, sex differences were noted, where females had higher mean sensory nerve action potentials in ulnar, median, and radial evaluations ( p < 0.001). In older patients (aged > 70 years) with normally interpreted EDX studies (845 records of 528 patients), sural responses were present in 97%. Conclusions: This real-life study confirms that advanced aging is associated with decreased nerve conduction amplitudes, increased latency, and the slowing of conduction velocity. The findings also indicate higher sensory amplitudes and conduction velocities in females. Sural nerve responses were identified in most adults over age 70.