Genovese, Ariana

Background: Large language models (LLMs) are increasingly integrated into clinical documentation, decision support, and patient-facing applications across healthcare, including plastic and reconstructive surgery. Yet, their evaluation remains bottlenecked by costly, time-consuming human review. This has given rise to LLM-as-a-judge, in which LLMs are used to evaluate the outputs of other AI systems. Methods: This review examines LLM-as-a-judge in healthcare with particular attention to judging architectures, validation strategies, and emerging applications. A narrative review of the literature was conducted, synthesizing LLM judge methodologies as well as judging paradigms, including those applied to clinical documentation, medical question-answering systems, and clinical conversation assessment. Results: Across tasks, LLM judges align most closely with clinicians on objective criteria (e.g., factuality, grammaticality, internal consistency), benefit from structured evaluation and chain-of-thought prompting, and can approach or exceed inter-clinician agreement, but remain limited for subjective or affective judgments and by dataset quality and task specificity. Conclusions: The literature indicates that LLM judges can enable efficient, standardized evaluation in controlled settings; however, their appropriate role remains supportive rather than substitutive, and their performance may not generalize to complex plastic surgery environments. Their safe use depends on rigorous human oversight and explicit governance structures.

Retrieval-augmented generation (RAG) quality depends on how source documents are segmented before indexing; fixed-length chunks can split concepts or add noise, reducing precision. We evaluated whether proposition, semantic, and adaptive chunking improve accuracy and relevance for safer clinical decision support. Using a curated domain knowledge base with Gemini 1.0 Pro, we built four otherwise identical RAG pipelines that differed only in the chunking strategy: adaptive length, proposition, semantic, and a fixed token-dependent baseline. Thirty common postoperative rhinoplasty questions were submitted to each pipeline. Outcomes included medical accuracy and clinical relevance (3-point Likert scale) and retrieval precision, recall, and F1; group differences were tested with ANOVA and Tukey post hoc analyses. Adaptive chunking achieved the highest accuracy—87% (Likert 2.37 ± 0.72) versus baseline 50% (1.63 ± 0.72; p = 0.001)—and the highest relevance (93%, 2.90 ± 0.40). Retrieval metrics were strongest with adaptive (precision 0.50, recall 0.88, F1 0.64) versus baseline (0.17, 0.40, 0.24). Proposition and semantic strategies improved all metrics relative to baseline, though less than adaptive. Aligning chunks to logical topic boundaries yielded more accurate, relevant answers without modifying the language model, offering a model-agnostic, data-source-neutral lever to enhance the safety and utility of LLM-based clinical decision support.

Background: Artificial Intelligence (AI)-driven Decision Support Systems (DSSs) promise improvements in diagnostic accuracy and clinical workflow efficiency, but their adoption is hindered by inadequate confidence calibration, limited transparency, and poor alignment with real-world decision processes, which limit clinician trust and lead to high override rates. Methods: We developed and validated a dynamic scoring framework to enhance trust in AI-generated diagnoses by integrating AI confidence scores, semantic similarity measures, and transparency weighting into the override decision process using 6689 cardiovascular cases from the MIMIC-III dataset. Override thresholds were calibrated and validated across varying transparency and confidence levels, with override rate as the primary acceptance measure. Results: The implementation of this framework reduced the override rate to 33.29%, with high-confidence predictions (90–99%) overridden at a rate of only 1.7%, and low-confidence predictions (70–79%) at a rate of 99.3%. Minimal transparency diagnoses had a 73.9% override rate compared to 49.3% for moderate transparency. Statistical analyses confirmed significant associations between confidence, transparency, and override rates (p < 0.001). Conclusions: These findings suggest that enhanced transparency and confidence calibration can substantially reduce override rates and promote clinician acceptance of AI diagnostics. Future work should focus on clinical validation to optimize patient safety, diagnostic accuracy, and efficiency.

Retrieval-Augmented Generation (RAG) offers a promising strategy to harness large language models (LLMs) for delivering up-to-date, accurate clinical guidance while reducing physicians’ cognitive burden, yet its effectiveness hinges on query clarity and structure. We propose an adaptive Self-Query Retrieval (SQR) framework that integrates three refinement modules—PICOT (Population, Intervention, Comparison, Outcome, Time), SPICE (Setting, Population, Intervention, Comparison, Evaluation), and Iterative Query Refinement (IQR)—to automatically restructure and iteratively enhance clinical questions until they meet predefined retrieval-quality thresholds. Implemented on Gemini-1.0 Pro, we benchmarked SQR using thirty postoperative rhinoplasty queries, evaluating responses for accuracy and relevance on a three-point Likert scale and for retrieval quality via precision, recall, and F1 score; statistical significance was assessed by one-way ANOVA with Tukey post-hoc testing. The full SQR pipeline achieved 87% accuracy (Likert 2.4 ± 0.7) and 100% relevance (Likert 3.0 ± 0.0), significantly outperforming a non-refined RAG baseline (50% accuracy, 80% relevance; p < 0.01 and p = 0.03). Precision, recall, and F1 rose from 0.17, 0.39 and 0.24 to 0.53, 1.00, and 0.70, respectively, while PICOT-only and SPICE-only variants yielded intermediate improvements. These findings demonstrate that automated structuring and iterative enhancement of queries via SQR substantially elevate LLM-based clinical decision support, and its model-agnostic architecture enables rapid adaptation across specialties, data sources, and LLM platforms.

The increasing integration of large language models (LLMs) into healthcare presents significant opportunities, but also critical risks related to patient safety, accuracy, and ethical alignment. Despite these concerns, no standardized framework exists for systematically evaluating and stress testing LLM behavior in clinical decision-making. The PIEE cycle—Planning and Preparation, Information Gathering and Prompt Generation, Execution, and Evaluation—is a structured red-teaming framework developed specifically to address artificial intelligence (AI) safety risks in healthcare decision-making. PIEE enables clinicians and informatics teams to simulate adversarial prompts, including jailbreaking, social engineering, and distractor attacks, to stress-test language models in real-world clinical scenarios. Model performance is evaluated using specific metrics such as true positive and false positive rates for detecting harmful content, hallucination rates measured through adapted TruthfulQA scoring, safety and reliability assessments, bias detection via adapted BBQ benchmarks, and ethical evaluation using structured Likert-based scoring rubrics. The framework is illustrated using examples from plastic surgery, but is adaptable across specialties, and is intended for use by all medical providers, regardless of their backgrounds or familiarity with artificial intelligence. While the framework is currently conceptual and validation is ongoing, PIEE provides a practical foundation for assessing the clinical reliability and ethical robustness of LLMs in medicine.

(1) Burn injuries demand multidisciplinary, evidence-based care, yet the extensive literature complicates timely decision making. Retrieval-augmented generation (RAG) synthesizes research while addressing inaccuracies in pretrained models. However, citation bias in sourcing for RAG often prioritizes highly cited studies, overlooking less-cited but valuable research. This study examines RAG’s performance in burn management, comparing citation levels to enhance evidence synthesis, reduce selection bias, and guide decisions. (2) Two burn management datasets were assembled: 30 highly cited (mean: 303) and 30 less-cited (mean: 21). The Gemini-1.0-Pro-002 RAG model addressed 30 questions, ranging from foundational principles to advanced surgical approaches. Responses were evaluated for accuracy (5-point scale), readability (Flesch–Kincaid metrics), and response time with Wilcoxon rank sum tests (p < 0.05). (3) RAG achieved comparable accuracy (4.6 vs. 4.2, p = 0.49), readability (Flesch Reading Ease: 42.8 vs. 46.5, p = 0.26; Grade Level: 9.9 vs. 9.5, p = 0.29), and response time (2.8 vs. 2.5 s, p = 0.39) for the highly and less-cited datasets. (4) Less-cited research performed similarly to highly cited sources. This equivalence broadens clinicians’ access to novel, diverse insights without sacrificing quality. As plastic surgery evolves, RAG’s inclusive approach fosters innovation, improves patient care, and reduces cognitive burden by integrating underutilized studies. Embracing RAG could propel the field toward dynamic, forward-thinking care.

Background/Objectives: Facial analysis is critical for preoperative planning in facial plastic surgery, but traditional methods can be time consuming and subjective. This study investigated the potential of Artificial Intelligence (AI) for objective and efficient facial analysis in plastic surgery, with a specific focus on Multimodal Large Language Models (MLLMs). We evaluated their ability to analyze facial skin quality, volume, symmetry, and adherence to aesthetic standards such as neoclassical facial canons and the golden ratio. Methods: We evaluated four MLLMs—ChatGPT-4o, ChatGPT-4, Gemini 1.5 Pro, and Claude 3.5 Sonnet—using two evaluation forms and 15 diverse facial images generated by a Generative Adversarial Network (GAN). The general analysis form evaluated qualitative skin features (texture, type, thickness, wrinkling, photoaging, and overall symmetry). The facial ratios form assessed quantitative structural proportions, including division into equal fifths, adherence to the rule of thirds, and compatibility with the golden ratio. MLLM assessments were compared with evaluations from a plastic surgeon and manual measurements of facial ratios. Results: The MLLMs showed promise in analyzing qualitative features, but they struggled with precise quantitative measurements of facial ratios. Mean accuracy for general analysis were ChatGPT-4o (0.61 ± 0.49), Gemini 1.5 Pro (0.60 ± 0.49), ChatGPT-4 (0.57 ± 0.50), and Claude 3.5 Sonnet (0.52 ± 0.50). In facial ratio assessments, scores were lower, with Gemini 1.5 Pro achieving the highest mean accuracy (0.39 ± 0.49). Inter-rater reliability, based on Cohen’s Kappa values, ranged from poor to high for qualitative assessments (κ > 0.7 for some questions) but was generally poor (near or below zero) for quantitative assessments. Conclusions: Current general purpose MLLMs are not yet ready to replace manual clinical assessments but may assist in general facial feature analysis. These findings are based on testing models not specifically trained for facial analysis and serve to raise awareness among clinicians regarding the current capabilities and inherent limitations of readily available MLLMs in this specialized domain. This limitation may stem from challenges with spatial reasoning and fine-grained detail extraction, which are inherent limitations of current MLLMs. Future research should focus on enhancing the numerical accuracy and reliability of MLLMs for broader application in plastic surgery, potentially through improved training methods and integration with other AI technologies such as specialized computer vision algorithms for precise landmark detection and measurement.

Background: Anatomically accurate illustrations are imperative in medical education, serving as crucial tools to facilitate comprehension of complex anatomical structures. While traditional illustration methods involving human artists remain the gold standard, the rapid advancement of Generative Artificial Intelligence (GAI) models presents a new opportunity to automate and accelerate this process. This study evaluated the potential of GAI models to produce craniofacial anatomy illustrations for educational purposes. Methods: Four GAI models, including Midjourney v6.0, DALL-E 3, Gemini Ultra 1.0, and Stable Diffusion 2.0 were used to generate 736 images across multiple views of surface anatomy, bones, muscles, blood vessels, and nerves of the cranium in both oil painting and realistic photograph styles. Four reviewers evaluated the images for anatomical detail, aesthetic quality, usability, and cost-effectiveness. Inter-rater reliability analysis assessed evaluation consistency. Results: Midjourney v6.0 scored highest for aesthetic quality and cost-effectiveness, and DALL-E 3 performed best for anatomical detail and usability. The inter-rater reliability analysis demonstrated a high level of agreement among reviewers (ICC = 0.858, 95% CI). However, all models showed significant flaws in depicting crucial anatomical details such as foramina, suture lines, muscular origins/insertions, and neurovascular structures. These limitations were further characterized by abstract depictions, mixing of layers, shadowing, abnormal muscle arrangements, and labeling errors. Conclusions: These findings highlight GAI’s potential for rapidly creating craniofacial anatomy illustrations but also its current limitations due to inadequate training data and incomplete understanding of complex anatomy. Refining these models through precise training data and expert feedback is vital. Ethical considerations, such as potential biases, copyright challenges, and the risks of propagating inaccurate information, must also be carefully navigated. Further refinement of GAI models and ethical safeguards are essential for safe use.

Artificial intelligence (AI) is not merely a tool for the future of clinical medicine; it is already reshaping the landscape, challenging traditional paradigms, and expanding the horizons of what is achievable in healthcare [...]