Concetti Chiave
Large language models (LLMs) have demonstrated remarkable potential in automating disease diagnosis, offering enhanced accuracy, efficiency, and accessibility in clinical practice.
Sintesi
This comprehensive review examines the current state of LLM-based methods for disease diagnosis. The key insights are:
Disease Types and Clinical Specialties: The review covers a wide range of disease types across 19 clinical specialties, including infectious diseases, cardiology, oncology, psychiatry, and more.
Clinical Data and Modalities: The input data spans diverse modalities, such as text (clinical notes, medical reports), images (X-rays, MRIs), time series (ECG, wearable sensors), and multimodal combinations.
LLM Techniques: The review categorizes the applied LLM techniques into four main groups: prompt-based methods, retrieval-augmented generation (RAG), fine-tuning, and pre-training. It analyzes the strengths, limitations, and appropriate use cases for each technique.
Evaluation Strategies: The review summarizes the pros and cons of automatic, human, and LLM-based evaluation methods, providing guidance on selecting suitable evaluation approaches.
Limitations and Future Directions: The review identifies key challenges, including the need for comprehensive multimodal data, adherence to clinical guidelines, human-centric perspectives (interpretability, privacy, fairness), and the development of robust, stable, and generalizable diagnostic systems.
Overall, this review offers a comprehensive blueprint for leveraging LLMs to advance disease diagnosis, inspiring future research and streamlining the development of efficient and accurate diagnostic tools.
Statistiche
"Automatic disease diagnosis has become increasingly valuable in clinical practice."
"Recent advancements in artificial intelligence (AI) have driven the development of automated diagnostic systems."
"LLMs, such as generative pre-trained transformers (GPT) and LLaMA, are generative models pre-trained on vast amounts of unlabeled data through self-supervised learning."
"Med-MLLM, a multimodal LLM pre-trained and fine-tuned on extensive medical data, including chest X-rays, CT scans, and clinical notes, demonstrated notable accuracy in COVID-19 diagnosis."
"Kim et al. employed GPT-4 with prompt engineering and found it surpassed mental health professionals in identifying obsessive-compulsive disorder."
Citazioni
"Automatic disease diagnosis has become increasingly valuable in clinical practice."
"Recent advancements in artificial intelligence (AI) have driven the development of automated diagnostic systems."
"LLMs, such as generative pre-trained transformers (GPT) and LLaMA, are generative models pre-trained on vast amounts of unlabeled data through self-supervised learning."
"Med-MLLM, a multimodal LLM pre-trained and fine-tuned on extensive medical data, including chest X-rays, CT scans, and clinical notes, demonstrated notable accuracy in COVID-19 diagnosis."
"Kim et al. employed GPT-4 with prompt engineering and found it surpassed mental health professionals in identifying obsessive-compulsive disorder."