Dr. Anant Madabhushi, Professor of FAIMBE, FIEEE and FNAI, Emory University, said at the beginning of his keynote address: “When you think of artificial intelligence, I think AI could potentially be the interpreter of all diseases.” It has secured over $1000 in research grants and stands out as a global leader in artificial intelligence (AI) to improve outcomes for cancer patients. In his session, “Interpreter of Maladies: Applications of AI to Problems in Skin Cancer and Beyond,” he emphasized the potential of AI to improve the accuracy and efficiency of cancer diagnosis, and to promote more informed treatment. We created a platform to help you make decisions. He will be attending the 2023 American College of Morse Surgery in Seattle, Washington.1
Dr. Anant Madhavshi. Photo credit: Twitter/@anantm
Employment security by AI
One of Madabhushi’s most popular questions is whether AI will replace the job of pathology. The short answer to that is “no”. Only useful in areas with medical disparities. The role of AI and computational imaging will help diagnostic, prognostic, and predictive tasks to save pathologists and dermatologists time.
He cited a study showing an assessment of the global pathologist shortage in 134 countries and territories representing 95% of the world’s population. As no database currently exists to locate all pathologists worldwide, study results were calculated using resources such as journals, communications with professional societies, and World Health Organization health worker reports. As a result, he has 1 pathologist for every 125 doctors, and 70% of his workforce of pathologists is located in his 10 countries, mainly the United States, India, China, Brazil and Russia. shown to be focused.2
First experience with neural networks
“I think you can call it old wine in a new bottle because deep learning has been based on machine learning algorithms called neural networks since the 60s,” Madabhushi said. He went on to explain that within the last decade, we can use better algorithms to increase computational power and build capacity for complex tasks. One of Madabhushi’s first examples of this approach in digital pathology was an image of a breast cancer, annotated with cancer cell locations and fed to the algorithm. The algorithm was able to learn representations associated with cancer cells and used those representations to identify cancer cells in new images. “The great thing about this approach is that without actually knowing anything about pathology, one of my students, he, could make an algorithm based solely on the location of these cancer cells, which was provided by the pathologist. I was able to train,” he explained.
AI isn’t always smart
Six years ago, when Madabhushi was collaborating with the University of Pennsylvania, his research team looked at endomyocardial biopsies and predicted the presence of heart failure from MMI-targeted biopsies. They were tasked with feeding the algorithm about 100 images of patients undergoing heart transplants. When measuring effectiveness, the algorithm came back with his 97% accuracy. For further validation, the same images were sent to a cardiac pathologist to identify the presence of heart failure, and the AI algorithm was approximately 20% more accurate than the pathologist. Needless to say, Madabushi and his team were nonchalant and the research made the news.3
The same technology was tested a few months later and performance dropped from 97% accuracy to 75%. He admits that one of his graduate students noticed that a remote software upgrade of the technology subtly changed the appearance of the images and hampered their performance. Mr Madabhushi said: One of the things our group has been trying to do is figure out how to deliberately try to deal with some of these variations in what we call the batch effect. So it’s a variation on how these slides are cheated when you get data from one lab versus another. ”
in conclusion
Madabhushi shared various recent AI studies that have led research into automated diagnosis of non-melanoma skin cancerFour, which was a study published last year. He also gave insight into preliminary studies on detection of squamous cell carcinoma from cryo-embedded hematoxylin- and eosin-stained slides in Mohs surgery. Both studies gave participants ideas for how his AI technology could help dermatologists in their daily routines in the future.
Keynote speakers made it clear that AI is not magic and requires thoughtful and purposeful development. Routine image processing and data-based computational analysis may help address issues in precision medicine, particularly in predicting prognosis and response to therapy. AI development has great advantages such as low cost, time saving, and availability. He made head-to-head comparisons showing that molecular testing costs about $4,000, takes two weeks to get results back to him, is only accessible in the US, and destroys tissue specimens. . A pathology test using AI technology, on the other hand, costs $4, gives results in 20 minutes, could be available worldwide, and has digitized tissue specimens.
References
1. Madabhushi, A. Disease Interpreter: Application of AI to skin cancer and beyond. Presented at: 2023 American College of Morse Surgery Conference. May 4-7, 2023. Seattle, Washington.
2. Bychkov A, Fukuoka J. Assessment of the global supply of pathologists. modern pathology. 2022;35:1473-1522. doi:10.1038/s41379-022-01050-6
3. Nirschl JJ, Janowczyk A, Peyster EG, et al. A deep learning classifier uses whole-slide images of H&E tissues to identify patients with clinical heart failure. PLoS One. 2018;13(4):e0192726. Published April 3, 2018. doi:10.1371/journal.pone.0192726
4. Zhou Y, Koyuncu C, Lu C, et al. Multisite Cross-Organ Calibration Deep Learning (MuSClD): Automated Diagnosis of Nonmelanoma Skin Cancer. Creampie image anal. 2023;84:102702. doi:10.1016/j.media.2022.102702
