Israel Medical Association Journal

Background: The rapid evolution of large language models warrants updated benchmarking in ophthalmology to determine whether newer versions offer clinically meaningful improvements over earlier models and human comparators.

Objectives: To evaluate the diagnostic accuracy of ChatGPT-4o and ChatGPT-5 in ophthalmic cases and to compare it with previously reported results of ChatGPT-3.5, residents, and specialists.

Methods: This retrospective cohort study was conducted in one academic tertiary medical center. We reviewed data of patients admitted to the ophthalmology department from June 2022 to January 2023. We then created two clinical cases for each patient. The first was according to medical history alone (Hx). The second added the clinical examination (Hx and Ex). For each case, we asked for the three most likely diagnoses from ChatGPT-4o and ChatGPT-5. We then compared the accuracy rates (at least one correct diagnosis) with previous results of ChatGPT-3.5, residents, and specialists.

Results: A total of 63 cases were analyzed, first using history alone and then with examination findings. Based on history alone, GPT-5 and GPT-4o correctly identified 73% and 70% of cases, respectively, outperforming GPT-3.5 (54%, P < 0.05) and approaching the accuracy of residents (75%) and attending physicians (71%, P < 0.05). When physical examination was included, diagnostic accuracy rose to 94% for GPT-5 and 89% for GPT-4o, surpassing GPT-3.5 (68%, P < 0.05) and closely matching or exceeding human performance (residents 94%, attendings 87%).

Conclusions: ChatGPT-4o and ChatGPT-5 significantly outperformed GPT-3.5 and achieved diagnostic accuracy similar or even higher to clinicians in diagnosing ophthalmology cases.

Background: Gait disturbances are common in patients undergoing hemodialysis and are associated with increased fall risk, mobility decline, and adverse health outcomes. Prior research suggests that hemodialysis may impact gait parameters such as speed, stride length, and variability; however, findings are inconsistent.

Objectives: To evaluate acute changes in gait metrics before and after hemodialysis using an artificial intelligence (AI) based video gait analysis system.

Methods: We initially enrolled 38 hemodialysis patients, two were excluded due to clothing interference with video analysis (27.8% female, 72.2% male). AI-driven gait analysis was performed immediately before and after dialysis. The system extracted spatiotemporal gait and joint range of motion. Statistical analyses included the Shapiro-Wilk test for normality, Wilcoxon signed-rank tests for non-normally distributed data, and paired t-tests for normally distributed data (P < 0.05).

Results: Gait speed (0.59 m/sec pre-dialysis) remained unchanged post-dialysis (P = 0.876), as did cycle length and time. However, step length significantly decreased post-dialysis (P = 0.001), suggesting a more conservative gait pattern. Knee flexion and extension increased slightly but did not reach statistical significance.

Conclusions: Dialysis does not acutely affect overall gait speed but significantly reduces stride length. Post-dialysis fatigue or hemodynamic shifts may alter walking patterns, highlighting the need for fall prevention strategies and physical rehabilitation interventions in dialysis care. AI-based gait analysis may provide a practical tool for monitoring mobility changes in hemodialysis patients.

Artificial intelligence (AI) has become the emblem of progress. We are told it learns faster, sees patterns invisible to the human eye, and will soon outthink us in every domain, from finance to philosophy, from language to life. In medicine, where decisions carry the weight of saving lives, this narrative has gained traction. Algorithms promise precision without fatigue, accuracy without bias, and reproducibility without emotion. Yet, sometimes, the data tell a quieter story.

Artificial intelligence (AI) and machine learning have moved to the forefront of scientific discourse and clinical medicine, offering improved accuracy and efficiency while raising concerns about transparency, accountability, and unintended consequences. Recent developments, particularly large-scale and generative models, have fueled these debates. However, efforts to mimic aspects of human intelligence long predate ChatGPT. These efforts include the early rule-based systems to Weizenbaum’s ELIZA program, which humorously simulated a Rogerian psychotherapist in its Doctor script [1]. For clinicians, the real test is not whether predictions become marginally more accurate on average, but whether they improve the identification of high-risk patients and meaningfully change management.

Medical education has evolved significantly since the establishment of university-based training in the 12th century, continually adapting to scientific, technological, and societal changes. In the 21st century, rapid advances in digital technologies, artificial intelligence, and the widespread availability of medical information are reshaping the competencies required of physicians. This review explores the concept of the doctor of the future and the implications for medical education and training. Future physicians will require strong cognitive abilities to interpret clinical data, utilize decision support technologies, and make informed clinical judgments while maintaining responsibility for final clinical decisions. At the same time, lifelong learning will be essential as medical knowledge and technologies rapidly evolve. Despite technological progress, core human qualities including empathy and effective communication, are essential traits that will be needed by future physicians. These skills require educational models that integrate technological competence, humanistic care, interdisciplinary teamwork, and adaptable career pathways to meet the challenges of evolving healthcare systems.

Background: Large language models (LLMs) are rapidly advancing, with the potential to improve healthcare. While LLM performance on medical licensing exams were studied extensively, their performance in rheumatology exams requires specific evaluation.

Objective: To assess Chat Generative Pre-Trained Transformer (ChatGPT) performance on 200 validated Israeli rheumatology board exam questions.

Methods: ChatGPT performance was evaluated using 200 multiple-choice questions from the 2023 and 2024 Israeli official rheumatology board examinations. Three gpt-4-turbo based variants were assessed: base model (Model 1), few-shot chain of thought (CoT) model (Model 2), and knowledge-augmented prompting model incorporating rheumatology guidelines (Model 3). Model 1 was assessed using both the original Hebrew and a validated English translated version, while Models 2 and 3 were assessed using the English version only.

Results: Overall, Model 3 achieved the highest numerical accuracy (81%), followed by Model 1 (English, 77%), Model 2 (75%), and Model 1 (Hebrew, 74.5%); however, these differences were not statistically significant. Performance varied markedly by question type. For text-only questions (n=177), accuracies ranged from 78.5% to 83.1%, with Model 3 showing the highest point estimate (83.1%). In contrast, all models demonstrated substantially lower performance on questions that included images (n=23), with accuracies ranging from 34.8% to 65.2%. Model 3 yielded the highest numerical accuracy (65.2%).

Conclusions: The study highlights the potential role of LLMs in rheumatology board examinations but also emphasizes their critical limitations. Future research should focus on addressing limitations, especially image interpretation and management of complex cases to enable efficient application of LLMs in rheumatology.

During these challenging times, following the October 7th terrorist attacks and the ongoing Iron Swords war, there is a greater need to strengthen the Israeli Society of Plastic and Aesthetic Surgery. Prof. Josef Haik, chair of the Israeli Society of Plastic and Aesthetic Surgery, leads this effort. In addition, it is vital to highlight Israel's contributions to the global Plastic and Aesthetic Surgery community and to encourage collaboration with the nursing division for plastic surgery and burns. Our department is involved in presenting our findings and collaborating with colleagues. In this article, we outlined five main topics: the role of plastic surgery in wartime, reconstructive plastic surgery, aesthetic plastic surgery, microsurgery, and innovation in plastic surgery.

Artificial Intelligence (AI), particularly large language models (LLMs) like OpenAI's ChatGPT, has shown potential in various medical fields, including pediatrics. We evaluated the utility and integration of LLMs in pediatric medicine. We conducted a search in PubMed using specific keywords related to LLMs and pediatric care. Studies were included if they assessed LLMs in pediatric settings, were published in English, peer-reviewed, and reported measurable outcomes. Sixteen studies spanning pediatric sub-specialties such as ophthalmology, cardiology, otology, and emergency medicine were analyzed. The findings indicate that LLMs provide valuable diagnostic support and information management. However, their performance varied, with limitations in complex clinical scenarios and decision-making. Despite excelling in tasks requiring data summarization and basic information delivery, the effectiveness of the models in nuanced clinical decision-making was restricted. LLMs, including ChatGPT, show promise in enhancing pediatric medical care but exhibit inconsistent performance in complex clinical situations. This finding underscores the importance of continuous human oversight. Future integration of LLMs into clinical practice should be approached with caution to ensure they supplement, rather than supplant, expert medical judgment.

In this special issue of Israel Medical Association Journal (IMAJ) we expose readers to the topic of artificial intelligence (AI) in medicine. AI has become a powerful tool, which enables healthcare professionals to personalize treatment based on many factors, including genetic analyses of tumors, and to consider other co-morbidities affecting a specific patient. AI gives physicians the ability to analyze huge amounts of data and to combine data from different sources. AI can be implemented make a diagnosis based on computed tomography (CT) scans and magnetic resonance imaging (MRI) scans using deep machine learning and data that are stored in the memory of mega computers. AI assists in tailoring more precise surgery to train surgeons before surgery and to support surgeons during procedures. This advancement may benefit surgical procedures by making them more accurate and faster without cutting unnecessary tissues (e.g., nerves and blood vessels); thus, patients face fewer complications, lower rates of infection, and more operation theater time. In this issue, we include three original studies that describe the use of AI in academia and eight review articles that discuss applications of AI in different specialties in medicine. One of the review articles addresses ethical issues and concerns that are raised due to the more advanced use of AI in medicine.

Artificial intelligence in ophthalmology is used for automatic diagnosis, data analysis, and predicting responses to possible treatments. The potential challenges in the application and assimilation of artificial intelligence include technical challenges of the algorithms, the ability to explain the algorithm, and the ability to diagnose and manage the medical course of patients. Despite these challenges, artificial intelligence is expected to revolutionize the way ophthalmology will be practiced. In this review, we compiled recent reports on the use and application of deep learning in various fields of ophthalmology, potential challenges in clinical deployment, and future directions.

Artificial intelligence (AI) has emerged as a powerful technology in medicine, with a potential to revolutionize various aspects of disease management. In recent years, substantial progress has been made in the development and implementation of AI algorithms and models for the diagnosis, screening, and monitoring of retinal diseases. We present a brief update on recent advancements in the implementation of AI in the field of retinal medicine, with a focus on age-related macular degeneration, diabetic retinopathy, and retinopathy of prematurity. AI algorithms have demonstrated remarkable capabilities in automating image analysis tasks, thus enabling accurate segmentation and classification of retinal pathologies. AI-based screening programs hold great promise in cost-effective identification of individuals at risk, thereby facilitating early intervention and prevention. Future integration of multimodal imaging data including optical coherence tomography with additional clinical parameters, will further enhance the diagnostic accuracy and support the development of personalized medicine, thus aiding in treatment selection and optimizing therapeutic outcomes. Further research and collaboration will drive the transformation of AI into an indispensable tool for improving patient outcomes and enhancing the field of retinal medicine.

Artificial intelligence (AI) has emerged as a transformative group of technologies in the field of medicine. Specifically in cardiology, numerous applications have materialized, and these are developing exponentially. AI-based risk prediction models leverage machine learning algorithms and large datasets to probe multiple variables, aid in the identification of individuals at high risk for adverse events, facilitate early interventions, and enable personalized risk assessments. Unique algorithms analyze medical images, such as electrocardiograms, echocardiograms, and cardiac computed tomography scans to enable rapid detection of abnormalities and aid in the accurate identification of cardiac pathologies. AI has also shown promise in guiding treatment decisions during coronary catheterization. In addition, AI has revolutionized remote patient monitoring and disease management by means of wearable and implantable sensing technologies. In this review, we discussed the field of cardiovascular genetics and personalized medicine, where AI holds great promise. While the applications of AI in cardiology are promising, challenges such as data privacy, interpretability of the findings, and multiple matters regarding ethics need to be addressed. We presented a succinct overview of the applications of AI in cardiology, highlighting its potential to revolutionize risk prediction, diagnosis, treatment, and personalized patient care.

A rise in the incidence of chronic health conditions, notably heart failure, is expected due to demographic shifts. Such an increase places an onerous burden on healthcare infrastructures, with recurring hospital admissions and heightened mortality rates being prominent factors. Efficient chronic disease management hinges on regular ambulatory care and preemptive action. The application of intelligent computational models is showing promise as a key resource in the ongoing management of chronic diseases, particularly in forecasting disease trajectory and informing timely interventions. In this review, we explored a pioneering intelligent computational model by Diagnostic Robotics, an Israeli start-up company. This model uses data sourced from insurance claims to forecast the progression of heart failure. The goal of the model is to identify individuals at increased risk for heart failure, thus enabling interventions to be initiated early, mitigating the risk of disease worsening, and relieving the pressure on healthcare facilities, which will result in economic efficiencies.

Over the last decade the use of artificial intelligence (AI) has reformed academic research. While clinical diagnosis of psoriasis and psoriatic arthritis is largely straightforward, the determining factors of a clinical response to therapy, and specifically to biologic agents, have not yet been found. AI may meaningfully impact attempts to unravel the prognostic factors that affect response to therapy, assist experimental techniques being used to investigate immune cell populations, examine whether these populations are associated with treatment responses, and incorporate immunophenotype data in prediction models. The aim of this mini review was to present the current state of the AI-mediated attempts in the field. We executed a Medline search in October 2023. Selection and presentation of studies were conducted following the principles of a narrative–review design. We present data regarding the impact AI can have on the management of psoriatic disease by predicting responses utilizing clinical or biological parameters. We also reviewed the ways AI has been implemented to assist development of models that revolutionize the investigation of peripheral immune cell subsets that can be used as biomarkers of response to biologic treatment. Last, we discussed future perspectives and ethical considerations regarding the use of machine learning models in the management of immune-mediated diseases.

Medical imaging data has been at the frontier of artificial intelligence innovation in medicine with many clinical applications. There have been many challenges, including patient data protection, algorithm performance, radiology workflow, user interface, and IT integration, which have been addressed and mitigated over the last decade. The AI products in imaging now fall into three main categories: triage artificial intelligence (AI), productivity AI, and augmented AI, each providing a different utility for radiologists, clinicians, and patients. Adoption of AI products into the healthcare system has been slow, but it is growing. It is typically dictated by return on investment, which can be demonstrated in each use case. It is expected to lead to wider adoption of AI products in imaging into the clinical workflow in the future.