Performance Evaluation of Large Language Models in Cervical Cancer Management Based on A Standardized Questionnaire: Comparative Study (Preprint) DOI Creative Commons

Warisijiang Kuerbanjiang,

Shengzhe Peng,

Yiershatijiang Jiamaliding

et al.

Journal of Medical Internet Research, Journal Year: 2024, Volume and Issue: 27, P. e63626 - e63626

Published: Dec. 11, 2024

Cervical cancer remains the fourth leading cause of death among women globally, with a particularly severe burden in low-resource settings. A comprehensive approach-from screening to diagnosis and treatment-is essential for effective prevention management. Large language models (LLMs) have emerged as potential tools support health care, though their specific role cervical management underexplored. This study aims systematically evaluate performance interpretability LLMs Models were selected from AlpacaEval leaderboard version 2.0 based on capabilities our computer. The questions inputted into cover aspects general knowledge, screening, diagnosis, treatment, according guidelines. prompt was developed using Context, Objective, Style, Tone, Audience, Response (CO-STAR) framework. Responses evaluated accuracy, guideline compliance, clarity, practicality, graded A, B, C, D corresponding scores 3, 2, 1, 0. rate calculated ratio B responses total number designed questions. Local Interpretable Model-Agnostic Explanations (LIME) used explain enhance physicians' trust model outputs within medical context. Nine included this study, set 100 standardized covering information, treatment international national Seven (ChatGPT-4.0 Turbo, Claude Gemini Pro, Mistral-7B-v0.2, Starling-LM-7B alpha, HuatuoGPT, BioMedLM 2.7B) provided stable responses. Among all included, ChatGPT-4.0 Turbo ranked first mean score 2.67 (95% CI 2.54-2.80; 94.00%) 2.52 2.37-2.67; 87.00%) without prompt, outperforming other 8 (P<.001). Regardless prompts, QiZhenGPT consistently lowest-performing models, P<.01 comparisons against except BioMedLM. Interpretability analysis showed that prompts improved alignment human annotations proprietary (median intersection over union 0.43), while medical-specialized exhibited limited improvement. Proprietary LLMs, show promise clinical decision-making involving logical analysis. use can accuracy some varying degrees. Medical-specialized such HuatuoGPT BioMedLM, did not perform well expected study. By contrast, those augmented demonstrated notable tasks, However, underscores need further research explore practical application practice.

Language: Английский

Digitalomics: Towards Artificial Intelligence / Machine Learning-Based Precision Cardiovascular Medicine DOI Open Access
Akihiro Nomura, Yasuaki Takeji, Masaya Shimojima

et al.

Circulation Journal, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 30, 2025

Recent advances in traditional "-omics" technologies have provided deeper insights into cardiovascular diseases through comprehensive molecular profiling. Accordingly, digitalomics has emerged as a novel transdisciplinary concept that integrates multimodal information with digitized physiological data, medical imaging, environmental electronic health records, and biometric data from wearables. This digitalomics-driven augmented multiomics approach can provide more precise personalized risk assessments optimization when combined conventional approaches. Artificial intelligence machine learning (AI/ML) technologies, alongside statistical methods, serve key analytical tools realizing this framework. review focuses on two promising AI/ML applications medicine: digital phonocardiography (PCG) AI text generators. Digital PCG uses models to objectively analyze heart sounds predict clinical parameters, potentially surpassing auscultation capabilities. In addition, large language models, such generative pretrained transformer, demonstrated remarkable performance assessing knowledge, achieving accuracy rates exceeding 80% licensing examinations, although there are issues regarding knowledge safety. Current challenges the implementation of these include maintaining up-to-date ensuring consistent outputs, but ongoing developments fine-tuning retrieval-augmented generation show promise addressing challenges. Integration practice, guided by appropriate validation strategies, may notably advance precision medicine

Language: Английский

Citations

0
Establishing a Prosperous Neurosurgical Practice: Essential Components for Emerging Surgeons DOI Creative Commons

Asem A. Muhsen,

Baha’eddin A. Muhsen

Dr Sulaiman Al Habib Medical Journal, Journal Year: 2025, Volume and Issue: 7(1), P. 1 - 7

Published: Jan. 1, 2025

Abstract This systematic review examines the multifaceted strategies essential for establishing a successful neurosurgical practice. It highlights key factors, such as collaboration with colleagues, engagement referring physicians, participation in conferences, digital presence, community engagement, and provision of exemplary patient care. Each strategy offers distinct benefits challenges, which collectively contribute to growth sustainability Collaboration senior colleagues enhances outcomes fosters innovation, whereas effective communication physicians strengthens trust ensures steady flow. Active conferences facilitates professional development; however, resource constraints may limit its feasibility new practitioners. A presence via social media well-managed website expands neurosurgeon’s reach but requires careful attention professionalism. Community increases public awareness satisfaction, although it strain clinical time. Providing care based on evidence-based practices is paramount fostering long-term relationships maintaining positive reputation. The suggests that future research should explore effects these practice outcomes. balanced integration elements crucial continued success practices.

Language: Английский

Citations

0
Convergence of Nanotechnology and Machine Learning: The State of the Art, Challenges, and Perspectives DOI Open Access
Amiya Kumar Tripathy, Akshata Y. Patne, Subhra Mohapatra

et al.

International Journal of Molecular Sciences, Journal Year: 2024, Volume and Issue: 25(22), P. 12368 - 12368

Published: Nov. 18, 2024

Nanotechnology and machine learning (ML) are rapidly emerging fields with numerous real-world applications in medicine, materials science, computer engineering, data processing. ML enhances nanotechnology by facilitating the processing of dataset nanomaterial synthesis, characterization, optimization nanoscale properties. Conversely, improves speed efficiency computing power, which is crucial for algorithms. Although capabilities still their infancy, a review research literature provides insights into exciting frontiers these suggests that integration can be transformative. Future directions include developing tools manipulating nanomaterials ensuring ethical unbiased collection models. This emphasizes importance coevolution technologies mutual reinforcement to advance scientific societal goals.

Language: Английский

Citations

2
Performance Evaluation of Large Language Models in Cervical Cancer Management Based on A Standardized Questionnaire: Comparative Study (Preprint) DOI Creative Commons

Warisijiang Kuerbanjiang,

Shengzhe Peng,

Yiershatijiang Jiamaliding

et al.

Journal of Medical Internet Research, Journal Year: 2024, Volume and Issue: 27, P. e63626 - e63626

Published: Dec. 11, 2024

Cervical cancer remains the fourth leading cause of death among women globally, with a particularly severe burden in low-resource settings. A comprehensive approach-from screening to diagnosis and treatment-is essential for effective prevention management. Large language models (LLMs) have emerged as potential tools support health care, though their specific role cervical management underexplored. This study aims systematically evaluate performance interpretability LLMs Models were selected from AlpacaEval leaderboard version 2.0 based on capabilities our computer. The questions inputted into cover aspects general knowledge, screening, diagnosis, treatment, according guidelines. prompt was developed using Context, Objective, Style, Tone, Audience, Response (CO-STAR) framework. Responses evaluated accuracy, guideline compliance, clarity, practicality, graded A, B, C, D corresponding scores 3, 2, 1, 0. rate calculated ratio B responses total number designed questions. Local Interpretable Model-Agnostic Explanations (LIME) used explain enhance physicians' trust model outputs within medical context. Nine included this study, set 100 standardized covering information, treatment international national Seven (ChatGPT-4.0 Turbo, Claude Gemini Pro, Mistral-7B-v0.2, Starling-LM-7B alpha, HuatuoGPT, BioMedLM 2.7B) provided stable responses. Among all included, ChatGPT-4.0 Turbo ranked first mean score 2.67 (95% CI 2.54-2.80; 94.00%) 2.52 2.37-2.67; 87.00%) without prompt, outperforming other 8 (P<.001). Regardless prompts, QiZhenGPT consistently lowest-performing models, P<.01 comparisons against except BioMedLM. Interpretability analysis showed that prompts improved alignment human annotations proprietary (median intersection over union 0.43), while medical-specialized exhibited limited improvement. Proprietary LLMs, show promise clinical decision-making involving logical analysis. use can accuracy some varying degrees. Medical-specialized such HuatuoGPT BioMedLM, did not perform well expected study. By contrast, those augmented demonstrated notable tasks, However, underscores need further research explore practical application practice.

Language: Английский

Citations

0