Chinese Medical Journal review discusses the future prospects of medical AI

In recent years, artificial intelligence (AI) has grown remarkably, fueled by innovative foundation models (FMs) such as GPT-n series by OpenAI, and derived large language models (LLMs) like Chat generative pre-trained transformer, popularly known as ChatGPT. These FMs and LLMs have found extensive applications in various fields, including medical sciences. As a result, medical AI has advanced rapidly all over the world. Specifically, MedPaLM, Google’s medical LLM, has successfully demonstrated expert-level accuracy in solving the U.S. Medical Licensing Examination.

Moreover, FMs as well as LLMs are being widely used for medical diagnostics and education, as evidenced by the research papers published by Springer Nature and the New England Journal of Medicine. Recent initiatives aimed at global medical imaging data sharing and the establishment of regulatory guidelines for medical AI by the U.S. FDA point towards the expanding horizons of AI in medical sciences. These developments herald a new era of growth and development in which AI can play a crucial role in improving diagnostic accuracy and support medical professionals worldwide.

However, like any emerging technology, medical AI faces several challenges which needs to be addressed for its future benefits. The review,  led by Prof. Io Nam Wong from the Institute for AI in Medicine, Faculty of Medicine, Macau University of Science and Technology, along with Olivia Monteiro, Zhuo Sun, Sheng Nie, and Yun Yin as the corresponding authors, provides a comprehensive analysis of the prospects of AI technologies such as FMs and LLMs for medical applications. This review was published online on 19 September 2024 and released in Volume 137, Issue 21 of the Chinese Medical Journal on 5 November 2024.

The authors note that medical AI models process and analyze various kinds of medical data. This includes images data (CT, MRI, and X-rays), textual data (patient records and physical examination findings), experimental data (cellular, animal, and clinical studies), and structured medical knowledge (anatomy, pathology, and pharmacology). The researchers further propose a novel classification framework for medical AI models, categorizing them into disease-specific, general-domain, and multi-modal models. Disease-specific models, often FMs, are tailored for diagnosing particular conditions; examples include RetFound for eye diseases, Neuro-Oph GPT for neuro-ophthalmology, UNI for computational pathology tasks, and Virchow for cancer detection. General-domain models, such as ChatGPT, MedPaLM, MedSAM, SAM-Med2D, and MedLSAM, offer broader applications, addressing the limitations of disease-specific models. Multi-modal models, such as PathChat, PLIP, OpenMEDLab, and IRENE, excel in textual information processing and data integration from multiple modalities.

Overall, medical AI models have proven highly promising for disease diagnosis and prognosis prediction, medical image segmentation, medical report generation, biomarker screening, molecular sub-typing, and medical question answering. Nevertheless, key challenges related to data collection and analysis persist, including issues with data volume, annotation, multi-modal fusion, biases, need for supervision, and data privacy concerns. According to Prof. Wong, “To facilitate progress, it is imperative to establish an environment that enables seamless data sharing supported by robust regulations. Additionally, the development of secure frameworks for handling extensive medical data is paramount to guaranteeing patient privacy while enabling efficient analyses.”

Future exploration should focus on algorithmic improvement and standardized evaluation protocols. Effective collaborations among researchers, healthcare professionals, and regulatory bodies will be essential for bringing medical AI into clinical practice.

“Medical AI models have tremendous potential for transforming healthcare. With their ability to enhance diagnostics, customize treatments, and improve patient outcomes, these models can redefine the healthcare field. However, to fully harness the transformative power of medical AI, continuous research, innovation, and careful ethical considerations are required to address the challenges that arise when implementing these tools effectively,” concludes Prof. Wong.

We hope that such advancements in this field continue to enhance the efficiency of diagnosis and treatment in the future!

***

Reference

DOI: 10.1097/CM9.0000000000003302