Assessing the Accuracy of Artificial Intelligence Models in Scoliosis Classification and Suggested Therapeutic Approaches

Journal of Clinical Medicine, Journal Year: 2024, Volume and Issue: 13(14), P. 4013 - 4013

Published: July 9, 2024

Background: Open-source artificial intelligence models (OSAIMs) are increasingly being applied in various fields, including IT and medicine, offering promising solutions for diagnostic therapeutic interventions. In response to the growing interest AI clinical diagnostics, we evaluated several OSAIMs—such as ChatGPT 4, Microsoft Copilot, Gemini, PopAi, You Chat, Claude, specialized PMC-LLaMA 13B—assessing their abilities classify scoliosis severity recommend treatments based on radiological descriptions from AP radiographs. Methods: Our study employed a two-stage methodology, where of single-curve were analyzed by following evaluation two independent neurosurgeons. Statistical analysis involved Shapiro–Wilk test normality, with non-normal distributions described using medians interquartile ranges. Inter-rater reliability was assessed Fleiss’ kappa, performance metrics, like accuracy, sensitivity, specificity, F1 scores, used evaluate systems’ classification accuracy. Results: The indicated that although some systems, accurately reflected recommended Cobb angle ranges disease treatment, others, such Gemini required further calibration. Particularly, 13B expanded range moderate scoliosis, potentially influencing decisions delaying Conclusions: These findings highlight need continuous refinement enhance applicability.

Language: Английский

---

Towards an AI Tutor for Undergraduate Geotechnical Engineering: A Comparative Study of Evaluating the Efficiency of Large Language Model Application Programming Interfaces

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: July 25, 2024

This study investigates the efficiency of Large Language Model (LLM) Application Programming Interfaces (APIs)—specifically GPT-4 and Llama-3—as AI tutors for undergraduate Geotechnical Engineering education. As educational needs in specialised fields like become increasingly complex, innovative teaching tools that provide personalised learning experiences are essential. research evaluates capabilities GPT-4’s Llama-3’s APIs integrating applying formulas, offering accurate problem-solving explanatory responses, adapting to varied requirements. Using comparative analysis, employs a formula integration approach known as Retrieval-Augmented Generation (RAG) with two widely used LLM models, Llama-3. A set 20 challenging questions, previously identified problematic zero-shot solutions GPT-4, served evaluation basis. The models were assessed on accuracy, integration, clarity explanation, adaptability. Results indicate Llama-3 have significant potential Engineering. utilising RAG, demonstrated superior performance, correctly answering 95% questions at temperature setting 0.1, 82.5% 0.5, 60% 1. In contrast, answered 25% tasks 45% API by 0.1. underscores need advanced techniques domain-specific training enhance utility APIs. Future should focus refining methods, expanding knowledge bases, assessing long-term outcomes. work contributes ongoing dialogue education, providing insights into deploying LLMs personalised, effective aids engineering disciplines.

Language: Английский

---

Leveraging Large Language Models for Enhancing Literature-Based Discovery

Big Data and Cognitive Computing, Journal Year: 2024, Volume and Issue: 8(11), P. 146 - 146

Published: Oct. 25, 2024

The exponential growth of biomedical literature necessitates advanced methods for Literature-Based Discovery (LBD) to uncover hidden, meaningful relationships and generate novel hypotheses. This research integrates Large Language Models (LLMs), particularly transformer-based models, enhance LBD processes. Leveraging LLMs’ capabilities in natural language understanding, information extraction, hypothesis generation, we propose a framework that improves the scalability precision traditional methods. Our approach LLMs with semantic enhancement tools, continuous learning, domain-specific fine-tuning, robust data cleansing processes, enabling automated analysis vast text identification subtle patterns. Empirical validations, including scenarios on effects garlic blood pressure nutritional supplements health outcomes, demonstrate effectiveness our LLM-based generating testable advances methodologies, fosters interdisciplinary research, accelerates discovery domain. Additionally, discuss potential drug discovery, highlighting their ability extract present key from literature. Detailed comparisons methods, Swanson’s ABC model, highlight approach’s advantages. comprehensive opens new avenues knowledge has revolutionize practices. Future work will refine LLM techniques, explore Retrieval-Augmented Generation (RAG), expand other domains, focus dehallucination.

Language: Английский

---

TransSMPL: Efficient Human Pose Estimation with Pruned and Quantized Transformer Networks

Electronics, Journal Year: 2024, Volume and Issue: 13(24), P. 4980 - 4980

Published: Dec. 18, 2024

Existing Transformers for 3D human pose and shape estimation models often struggle with computational complexity, particularly when handling high-resolution feature maps. These challenges limit their ability to efficiently utilize fine-grained features, leading suboptimal performance in accurate body reconstruction. In this work, we propose TransSMPL, a novel Transformer framework built upon the SMPL model, specifically designed address of complexity inefficient utilization maps estimation. By replacing HRNet MobileNetV3 lightweight extraction, applying pruning quantization techniques, incorporating an early exit mechanism, TransSMPL significantly reduces both cost memory usage. introduces two key innovations: (1) multi-scale attention reduced from four scales two, allowing more efficient global local integration, (2) confidence-based strategy, which enables model halt further computations high-confidence predictions are achieved, enhancing efficiency. Extensive dynamic also applied reduce size while maintaining competitive performance. Quantitative qualitative experiments on Human3.6M dataset demonstrate efficacy TransSMPL. Our achieves MPJPE (Mean Per Joint Position Error) 48.5 mm, reducing by over 16% compared existing methods similar level accuracy.

Language: Английский

---