What is the current state of artificial intelligence and pathology?
Feb 20th 2020
Artificial Intelligence is a hot topic that touches many worlds, including medicine. We sat down with an expert to discuss AI and its roles in pathology.
Elizabeth Anne Montgomery, MD, is a professor of pathology, oncology and orthopedic surgery at the Johns Hopkins School of Medicine in Baltimore. She specializes in gastrointestinal pathology and soft tissue pathology. She reviews biopsy specimens from patients at Johns Hopkins and from around the world.
Q. One of the most discussed topics in pathology these days seems to be artificial intelligence (AI). Do you see this much in your practice?
A. No. At this point, AI is of very limited value in diagnostic pathology community, but it will improve over time. AI on a good day has to jump through many hoops simply to determine whether tissue is from the stomach or the colon. If a human performed as badly as AI does today, it would be a disaster. For example, in one study, neural networks missed half of gastric cancers.1 Of course, gastric cancers are tricky to spot. A more recent article showed that the computer could be taught to pick up Helicobacter pylori infection quite well, but the computer would miss the associated gastric cancer.2 AI is also really slow because the intelligence is artificial whereas a human can deal with glass slides very rapidly.
Q. Do you think the technology is improving?
A. Yes, it is. A recent study showed that with the massive expense and amazing numbers of scanned images, a computer could learn to triage certain types of samples that are tedious for a human to review such as spotting cancer cells in lymph nodes. However, anything flagged by AI needs a human to come behind and check it.3
Q. So, clinical utility is a number of years away, it seems.
A. Using AI requires preparing whole slide scans, which is time consuming. At this point, colleagues using scanned slides to perform routine diagnostic work (without AI) must work more slowly4. However, using scanned slides is of value when distances are great and it is faster to use scans than to drive a long distance to another site. Regardless, a colleague whom I know in pathology spends lots of time promoting whole slide scanning and AI at meetings (and is paid to do so). However, he does all his own daily work for which he earns his daily bread using traditional methods because they are lightning quick compared to using scanned slides and are far superior to current AI methods.
 Yoshida, Hiroshi, et al. “Automated histologic classification of whole-slide images of gastric biopsy specimens.” Gastric Cancer 21.2 (2018): 249-257
 Martin DR, Hanson JA, Gullapalli RR, Schultz FA, Sethi A, Clark DP. A Deep Learning Convolutional Neural Network Can Recognize Common Patterns of Injury in Gastric Pathology. Arch Pathol Lab Med. 2019 Jun 27. doi: 10.5858/arpa.2019-0004-OA. [Epub ahead of print] PubMed PMID: 31246112
 Campanella G, Hanna MG, Geneslaw L, Miraflor A, Werneck Krauss Silva V, Busam KJ, Brogi E, Reuter VE, Klimstra DS, Fuchs TJ. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images. Nat Med. 2019 Jul 15. doi: 10.1038/s41591-019-0508-1. [Epub ahead of print] PubMed PMID: 31308507
 Hanna MG, Reuter VE, Hameed MR, Tan LK, Chiang S, Sigel C, Hollmann T, Giri D, Samboy J, Moradel C, Rosado A, Otilano JR 3rd, England C, Corsale L, Stamelos E, Yagi Y, Schüffler PJ, Fuchs T, Klimstra DS, Sirintrapun SJ. Whole slide imaging equivalency and efficiency study: experience at a large academic center. Mod Pathol. 2019 Jul;32(7):916-928. PubMed PMID: 30778169