相关实验视频

Updated: Jul 8, 2025

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

域调整的大型语言模型用于核医学报告的分类

Zachary Huemann¹, Changhee Lee¹, Junjie Hu¹

¹From the Departments of Radiology (Z.H., C.L., S.Y.C., T.J.B.), Biostatistics (J.H.), and Computer Science (J.H.), University of Wisconsin-Madison, 1111 Highland Ave, Madison, WI 53705; and University of Wisconsin Carbone Cancer Center, Madison, Wis (S.Y.C.).

Radiology. Artificial intelligence

|December 11, 2023

概括

域调整显著增强了用于从PET/CT报告中预测淋巴瘤Deauville分数的语言模型. 这种人工智能方法看起来很有前途,域调整的RoBERTa表现优于人类的准确性.

科学领域:

医疗成像中的人工智能
在临床报告中使用自然语言处理 (NLP).
机器学习在核医学中的应用

背景情况:

从PET/CT报告中准确预测Deauville得分对于淋巴瘤治疗评估至关重要.
现有的语言模型可能会与核医学报告的专业术语和背景作斗争.
域调整提供了一种潜在的方法来提高NLP模型在特定医疗任务上的性能.

研究的目的:

评估域调整对语言模型在预测五分德维尔分数中的表现的影响.
将域适应的语言模型与非域适应的版本,视觉模型,多式模式模型和人类专家性能进行比较.

主要方法:

追溯分析了4542个PET/CT淋巴瘤检查报告,其中1664个是为了培训而提取的Deauville分数.
在核医学数据上使用掩面语言建模的BERT,BioClinicalBERT,RadBERT和RoBERTa模型的域调整.
使用七倍蒙特卡洛交叉验证进行五点杜维尔分数预测的模型的比较.

主要成果:

域调整显著提高了所有测试语言模型的准确性 (P = .01).
适应域的RoBERTa获得了最高的准确性 (77.4%±3.4),超过仅视觉模型和核医学医生 (66%的准确性).
性能最好的适应域的 RoBERTa 模型的性能与其多式联络对应器相提并论.

结论:

域调整是一种有效的策略,可以提高大型语言模型在预测PET/CT报告中的Deauville分数方面的性能.
人工智能模型,特别是适应领域的RoBERTa,显示出在这个任务中与人类专家的表现相匹配或超过的潜力.
这项研究强调了转移学习和NLP在改善临床成像报告分析方面的价值.

关键词:

人工智能的人工智能卷积神经网络 (CNN) 是一种神经网络.德奥维尔是什么意思德奥维尔是什么意思语言建模语言建模淋巴瘤是一种淋巴瘤.机器学习机器学习多模式学习是多模式学习.自然语言处理自然语言处理.核医学是一种核医学.在这里,PET是PET.聚乙烯/聚乙烯/聚乙烯转移学习转移学习没有监督的学习学习.

更多相关视频

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model

Published on: August 16, 2020

相关概念视频

Nuclear Localization Signals and Import

Nuclear Localization Signals and Import

Proteins targeted to the nucleus carry short stretches of amino acid sequences called the nuclear localization signal or NLS. Classical nuclear localization signals are of two types: monopartite and bipartite NLS. Monopartite classical NLS (cNLS) consists of a single cluster of 4-8 amino acids. Bipartite cNLS consists of two clusters of 2-3 amino acids and a 9-12 residue long proline-rich linker bridging the two clusters. Signal clusters are rich in positively charged amino acids such as...

Positron Emission Tomography

Positron Emission Tomography

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...

Applications Of NMR In Biology

Applications Of NMR In Biology

Nuclear magnetic resonance (NMR) spectroscopy is a very valuable analytical technique for researchers. It has been used for more than 50 years as an analytical tool. F. Bloch and E. Purcell formulated NMR in 1946 and won the 1952 Nobel Prize in Physics for their work. Biological macromolecules such as proteins, nucleic acids, lipids, and organic molecules including pharmaceutical compounds, can be studied using this versatile tool that exploits the magnetic properties of certain nuclei.