Is Artificial Intelligence the future of pathology?
Within the last ten years, essential steps have been made to bring artificial intelligence (AI) successfully into the field of pathology. Even so, most pathologists are still far away from using AI in daily pathology practice. Dr Travis Brown is a general pathologist with a special interest in Information Technology and Pathology Informatics. He explains that AI is part of the future of pathology.
“The future for AI with pathology is really exciting but is still only in the research phase. There are companies around the world that have invested a lot of time and money into this technology, which is mainly focused on Histology and uses digital images from scanned glass slides. However, this is currently very limited because it relies on Digital Pathology scanners which were only approved in the US this year for routine diagnostic work. At the moment this is prohibitively expensive and has limited benefits.
“Dr Ajit Singh in the US states that AI is the effort to mimic the workings of a human mind. This is also the underlying principle of the Turing Test which was designed by Alan Turing in the 1950s to evaluate if a person could be tricked into believing they were interacting with another human but were actually conversing with a machine or a computer. Alan Turing believed this would happen in around 50 years. We are now starting to see computers being able to achieve this level of ‘Artificial Intelligence’ in 2020,” said Dr Brown.
The integration of AI with pathology involves the creation of a pathologist’s aide or assistant to help in analysing images of biopsies taken from patients. As computers excel at repetitive and reproducible tasks, prompts may be highlighted to guide a pathologist in targeted features that are important diagnostically, such as perineural invasion, tumour infiltrating lymphocytes. This means that computers can assist in detecting relevant features on large digital images. It could even help identify small features that have the potential to be missed.
However, it is important to note there are certain limitations to AI. To start, all will require interpretation by a pathologist as many diagnoses fall into grey areas. For example, there are limitations on being able to distinguish between inflamed (reactive) cells and cancerous cells (which can also be a challenge for the most experience pathologist), or the AI program may list the potential diagnoses with the probabilities of being correct that will need a pathologist to review. Additionally, the dissemination of large images into many smaller images for their analysis loses the overall architecture of the specimen, meaning important aspects such as lesion depth and low-power morphology cannot be assessed by the computer.
“AI is part of the future of pathology, but it is not coming to replace jobs. The technology which is being developed is aimed to assist pathologists in their routine reporting tasks and I believe it will help us do our job with more accuracy and increased efficiency. However, when this technology will be ready is anyone’s guess. Mine is that it is at least a decade away, but again, that is a guess and I’m more than happy to be proven wrong.
“There is a quote that I would like to modify slightly from Professor Sam Gambhir (Stanford University) about AI and Radiologists,” said Dr Brown. ‘It’s not that AI will replace Pathologists, it’s that Pathologists using AI will replace Pathologists that are not using AI’.”