text
stringlengths
0
89.3k
Purpose to Medical Applications Datasets
Methodologies and Evaluations
Jinqiang Wang Huansheng Ning Senior Member IEEE Yi Peng Qikai Wei Daniel Tesfai Wenwei Mao Tao
Zhu Runhe Huang Senior Member IEEE
Abstract Large Language Models LLMs have demonstrated
surprising performance across various natural language pro
cessing tasks Recently medical LLMs enhanced with domain
specific knowledge have exhibited excellent capabilities in medical
consultation and diagnosis These models can smoothly simulate
doctorpatient dialogues and provide professional medical advice
Most medical LLMs are developed through continued training
of opensource general LLMs which require significantly fewer
computational resources than training LLMs from scratch Addi
tionally this approach offers better protection of patient privacy
compared to APIbased solutions This survey systematically
explores how to train medical LLMs based on general LLMs
It covers a how to acquire training corpus and construct
customized medical training sets b how to choose a appropriate
training paradigm c how to choose a suitable evaluation bench
mark and d existing challenges and promising future research
directions are discussed This survey can provide guidance for the
development of LLMs focused on various medical applications
such as medical education diagnostic planning and clinical
assistants
Index Terms Large Language Model Domain Medicine
Healthcare
I I NTRODUCTION
RECENTLY large language models LLMs such as
ChatGPT 1 have demonstrated exceptional capabil
ities in handling natural language processing NLP tasks
Their performance improves significantly as the scaling of
the model increases while the size of the dataset keeps
expanding Currently LLMs are widely applied in intelligent
education legal consultation code generation healthcare and
finance showcasing their powerful NLP capabilities and broad
applicability In intelligent education 2 3 LLMs find
application in intelligent tutoring systems and personalized
learning planning providing personalized advice by analyzing
students learning behaviors and feedback In legal consulting
4 5 LLMs facilitate legal document analysis contract
review and legal question answering improving the efficiency
and accuracy of legal services In code generation 6 7
LLMs can automatically generate code snippets perform code
completions and fixes augmenting development efficiency
Jinqiang Wang Huansheng Ning Yi Peng Qikai Wei Daniel Tesfai
and Wenwei Mao are with the School of Computer Communication
Engineering University of Science and Technology Beijing 100083 China
Corresponding author Huansheng Ning
Tao Zhu is with the School of Computer Science University of South
China 421001 China
Runhe Huang is with the Hosei University Japanand code quality In heathcare 8 9 LLMs are used for
medical literature analysis and doctorpatient dialogues assist
ing medical decisions and improving patient care In finance
10 11 LLMs are employed for risk assessment financial
analysis and customer service enhancing the intelligence and
personalization of financial services
The success of these above domain LLMs relies on inher
iting the strong knowledge of general LLMs and performing
further training injected with domainspecific knowledge The
general LLM has acquired extensive knowledge linguistic
styles and generalization abilities across a diverse corpus
enabling it to quickly learn new tasks Therefore the strategy
of training domain LLMs based on general LLMs is more
economical and effective than training domain LLMs from
scratch 12 There are three optional training phases for
training the medical LLM from the general LLM 1315
Continued Pretraining CP Instruction Finetuning IFT and
Human Alignment HA CP involves training on unstructured
domain data to help LLMs learn domainspecific knowledge
terminology and language style IFT focuses on training with
domain instruction data enabling LLMs to master domain
specific dialogue and instruction following HA trains LLMs
on human preference data equipping it with domainspecific
qualities such as harmlessness and helpfulness
Health constitutes the cornerstone of human survival and
development and advancements in medical technology and
services significantly impact quality of life and longevity 16
17 Given the aging population and rising prevalence of
chronic diseases the medical domain confronts disparities in
resource allocation necessitating more efficient solutions 18
Medical LLMs can address this challenge by integrating
rich medical data and clinical cases to help physicians
quickly and accurately make diagnoses and formulate
treatment plans 19 Furthermore it can help medical insti
tutions optimize resource allocation and improve the efficiency
and quality of medical services Existing medical LLMs focus
ing on different medical disciplines such as internal medicine
respiratory medicine and gastroenterology can provide more
specialized answers compared to general LLMs Additionally
in medical consultation tasks medical LLMs excel in inter
active diagnosis proactively prompting and guiding patients
to provide missing information for ambiguous descriptions
distinguishing them from general LLMs
Given the significance of the healthcare domain the ad
vantages of medical LLMs over general LLMs and the costarXiv240610303v1 csCL 14 Jun 2024JOURNAL OF L ATEX CLASS FILES VOL 14 NO 8 AUGUST 2021 2
Data Acquisition and Processing Sec II
Training Corpus Sources Data Processing
Training Paradigm s Sec III
General Large
Language Model