Designing clinically translatable artificial intelligence systems for high-dimensional medical imaging
by Roxanna Van Norman
December 8, 2021
Recent advances in computing power, deep learning machines, and sophisticated datasets have spurred the development of medical imaging artificial intelligence (AI) systems but translating them into practical clinical decision-making processes remains challenging.
That is the latest review conducted by a team of researchers from the Stanford Department of Cardiothoracic Surgery who examined the challenges unique to high-dimensional clinical imaging. The paper, "Designing clinically translatable artificial intelligence systems for high-dimensional medical imaging," was published in November issue of Nature Machine Intelligence.
Cloud-based collaborative annotation workflows. Cloud-based tools such as MD.ai can be used to generate expert-annotated datasets and evaluate them against clinical experts via a secure connection. An implementation of MD.ai in which clinical experts make a variety of 2D measurements to quantify cardiac function is shown. Credit: MD.ai Inc, NY.
The potential for deep learning to revolutionize healthcare is well understood, explained Rohan Shad, MD, a postdoctoral research fellow in Department of Cardiothoracic Surgery and the lead author on the paper. Yet, as the paper describes it, it is challenging to deploy AI systems.
"There are many practical hurdles relating to how one deals with data sharing, selecting the right architecture, and navigating challenges such as uncertainty and explainability," said Dr. Shad. "This manuscript was written to serve as a technical roadmap in many ways,
finding readership among computer scientists, biologists, imaging specialists, and those who seek to understand how AI stands to shape the future of the clinical practice."
While there is skepticism surrounding medical imaging and AI as described in the paper, the paper provides insights into why this is important.
"In my eyes, the importance of this paper lies in promoting researchers to think far ahead of the immediate technical challenges associated with building a working AI system," said Dr. Shad. "Much like ones [where scientists] plans years in advance when building a medical device, we think that researchers must think actively about addressing the challenges we describe regarding clinical deployment, from the inception of the project."
The paper's authors included John P. Cunningham, PhD, Euan A. Ashley, MD, Curtis P. Langlotz, MD, PhD, and William Hiesinger, MD.
Read more about the paper: https://www.nature.com/articles/s42256-021-00399-8
Dr. Rohan Shad