AI Language Models like ChatGPT Show Promise in Assisting Breast Cancer Screening Decisions
Artificial intelligence (AI) language models have demonstrated the potential to assist in breast cancer screening decisions, according to a recent study published in the Journal of the American College of Radiology. The study, conducted by investigators from various institutions, including Harvard Medical School and Massachusetts General Hospital, found that AI language models like OpenAI’s ChatGPT can accurately identify appropriate imaging tests for breast cancer screenings and breast pain.
Primary care providers often struggle to select the most suitable imaging tests for patients complaining of breast pain. Radiologists typically rely on evidence-backed guidelines to make these decisions. However, these guidelines may be less familiar to non-specialists, leading to confusion and potentially unnecessary or incorrect testing.
ChatGPT’s abilities were impressive, said Dr. Marc D. Succi, one of the senior study authors. This AI language model can act as a bridge between referring health-care professionals and expert radiologists, offering real-time recommendations for the most appropriate imaging test. Dr. Succi believes that integrating AI language models like ChatGPT into the decision-making process can reduce administrative burden, optimize workflow, and improve patient care.
To assess ChatGPT’s clinical decision-making capabilities, the investigators presented the model with 21 patient scenarios related to breast cancer screenings or breast pain. They found that the more advanced version, ChatGPT-4, outperformed its predecessor, ChatGPT-3.5, especially when given multiple imaging options. ChatGPT-3.5 correctly answered an average of 88.9% of prompts regarding breast cancer screenings, while ChatGPT-4 achieved an impressive accuracy of about 98.4%.
It is important to note that this study does not compare ChatGPT to human radiologists but instead compares the model’s performance to evidence-based guidelines from the American College of Radiology. The aim is to highlight ChatGPT’s potential as an adjunct tool rather than a replacement for physicians in selecting imaging tests.
The investigators proposed integrating ChatGPT into electronic health records, where it can prompt primary care providers to consider the best imaging options and educate patients about what to expect during the tests. Additionally, they suggested developing specialized medical AI language models using data from hospitals and research institutions to cater to specific patient populations and complex diagnoses.
Before AI language models can be fully incorporated into medical decision-making, extensive testing for bias, privacy concerns, and approval for medical settings is crucial. Regulatory measures may also be required to establish parameters for patient care interactions involving AI.
This study, supported in part by a grant from the National Institute of General Medical Sciences, showcases the potential of AI language models like ChatGPT in assisting primary care providers in making informed decisions about breast cancer screenings and breast pain. However, further research and development are necessary to ensure the effective and ethical integration of these models into clinical practice.
