Scientists from Weill Cornell Medicine and the Dana-Farber Cancer Institute in Boston have developed and tested new artificial intelligence (AI) tools tailored to digital pathology — a field that uses high-resolution digital images from tissue samples to diagnose diseases and guide treatment.
A recent study published in The Lancet Digital Health on July 9, introduces ChatGPT, an AI language model specifically designed to understand and generate text related to digital pathology. The research shows that this AI tool can provide accurate responses to questions and compile detailed results, making it easier for pathologists without extensive coding knowledge to use complex software for analyzing tissue samples.
Dr. Mohamed Omar, the lead author of the study and an assistant professor at Weill Cornell Medicine, highlighted that general language models may not provide useful information in specialized fields like digital pathology. This led researchers to customize ChatGPT, aiming to enhance efficiency and accuracy for digital pathology decision-making.
Enhancing AI Precision for Pathology
Standard language models can often give generic responses lacking specialized information and may generate inaccurate content. To address this, Dr. Renato Umeton from Dana-Farber Cancer Institute developed a specialized version of ChatGPT called GPT4DFCI. This version was enriched with access to a curated digital pathology database of the latest research, resulting in more precise and detailed responses.
Through a technique called retrieval-augmented generation (RAG), GPT4DFCI could access the database for accurate responses regarding digital pathology. By comparing the responses from GTP4DFCI to a general model, researchers verified the higher accuracy and relevance of the specialized model.
AI Assists with Coding
In addition to text generation, the team also developed an AI tool to assist pathologists in using PathML, a software library for analyzing histopathology image datasets that requires coding knowledge. By integrating PathML with ChatGPT, users can now receive step-by-step instructions on how to analyze samples using PathML, even without prior coding experience.
Dr. Umeton emphasized the potential of these AI tools to support researchers in different medical fields. This research demonstrates that combining AI with specialized information retrieval can significantly benefit complex topics like digital pathology.
