最近对LLMs在辐射瘤物理学方面的表现进行评估,使用随机混合选项问题
Peilong Wang1, Jason Holmes1, Zhengliang Liu2
1Department of Radiation Oncology, Mayo Clinic, Phoenix, AZ, United States.
Frontiers in oncology
|June 9, 2025
概括
最近发布的大型语言模型 (LLM) 显示了辐射瘤物理学的专家级性能. 虽然它们通常是准确的,但当正确的答案被掩盖时,它们的推理会下降,尽管特定的提示可以提高性能.
科学领域:
- 医学物理 医学物理
- 人工智能的人工智能
背景情况:
- 大型语言模型 (LLM) 越来越能够完成复杂的任务.
- 评估LLM在专业科学领域的表现,如辐射瘤物理学至关重要.
研究的目的:
- 评估最近发布的LLMs在辐射瘤学物理问题上的表现.
- 调查LLM的演推理能力和特定提示策略的影响.
主要方法:
- 使用了一套100个多选题辐射瘤物理问题.
- 测试了五个LLM (o1-预览,GPT-4o,LLaMA 3.1,Gemini 1.5 Pro,Claude 3.5 Sonnet) 进行了测试.
- 通过将正确答案替换为"上述任何一项"并使用"首先解释"和"逐步"提示,评估了演推理.
主要成果:
- 所有经过测试的LLM都表现出专家水平的表现,o1预览超过了医学物理学家的共识.
- 当正确答案选项被掩盖时,表现显著下降.
- 特定提示 ("首先解释"",逐步解释") 改善了LLaMA 3.1,Gemini 1.5 Pro和Claude 3.5 Sonnet的推理.
结论:
- 最近的LLM在辐射瘤物理学方面取得了专家水平的熟练程度.
- 法学士的推理能力需要进一步发展,特别是在复杂的问题解决场景中.
- 快速工程可以提高专业科学领域的LLM绩效.
更多相关视频
06:20Irradiator Commissioning and Dosimetry for Assessment of LQ α and β Parameters, Radiation Dosing Schema, and in vivo Dose Deposition
Published on: March 11, 2021
7.2K
08:25Radiation Planning Assistant - A Streamlined, Fully Automated Radiotherapy Treatment Planning System
Published on: April 11, 2018
15.2K
相关概念视频
Radiation Pressure: Problem Solving
467
The radiation pressure applied by an electromagnetic wave on a perfectly absorbing surface equals the energy density of the wave. The wave's momentum also gets transferred to the surface when an electromagnetic wave is entirely absorbed by it. The rate at which momentum is transmitted to an absorbing surface perpendicular to the propagation direction equals the force on the surface.
The average value of the rate of momentum transfer divided by the absorbing area represents the average force...
The average value of the rate of momentum transfer divided by the absorbing area represents the average force...
467
Positron Emission Tomography
6.0K
Positron emission tomography (PET) is a medical imaging technique involving radiopharmaceuticals — substances that emit short-lived radiation. Although the first PET scanner was introduced in 1961, it took 15 more years before radiopharmaceuticals were combined with the technique and revolutionized its potential.
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body...
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body...
6.0K
Cancer Survival Analysis
470
Cancer survival analysis focuses on quantifying and interpreting the time from a key starting point, such as diagnosis or the initiation of treatment, to a specific endpoint, such as remission or death. This analysis provides critical insights into treatment effectiveness and factors that influence patient outcomes, helping to shape clinical decisions and guide prognostic evaluations. A cornerstone of oncology research, survival analysis tackles the challenges of skewed, non-normally...
470
Radiation: Applications
1.3K
The average temperature of Earth is the subject of much current discussion. Earth is in radiative contact with both the Sun and dark space; it receives almost all its energy from the radiation of the Sun and reflects some of it into outer space. Dark space is very cold, about 3 K, so Earth radiates energy into it. For instance, heat transfer occurs from soil and grasses, the rate of which can be so rapid that frost can occur on clear summer evenings, even in warm latitudes.
The average...
The average...
1.3K
Maxwell-Boltzmann Distribution: Problem Solving
1.8K
Individual molecules in a gas move in random directions, but a gas containing numerous molecules has a predictable distribution of molecular speeds, which is known as the Maxwell-Boltzmann distribution, f(v).
This distribution function f(v) is defined by saying that the expected number N (v1,v2) of particles with speeds between v1 and v2 is given by
This distribution function f(v) is defined by saying that the expected number N (v1,v2) of particles with speeds between v1 and v2 is given by
1.8K
