Dr. Chaim Linhart, CTO and co-founder of Ibex Medical Analytics, considers the effect that artificial intelligence will have on the population through healthcare.
Taking cancer diagnosis to the next level
In the UK and worldwide, cancer rates are on the rise but there are not enough pathologists on board to deal with the increased number of cases. This has led to delays in cancer diagnosis exacerbated by the COVID-19 crisis. In fact, Cancer Research UK estimates that, since lockdowns began in March, more than 350,000 referrals have been missed. This is exerting tremendous pressure on pathology departments dealing with an influx of later-stage and more complex cases, while also raising concerns about diagnostic accuracy.
Pathologists have an especially difficult job. They are overworked and there are just not enough of them to meet demand; even before COVID-19 hit, one in ten diagnostic posts in the NHS were vacant. Cancer Research UK warned in October that, as vacancies pile up and demand for cancer services increases, diagnostic teams in all four UK nations are missing waiting times targets. A key issue is that most pathologists manually examine specimens - a process that takes time and effort and is at times subjective and prone to human error. It's clear that pathology needs a new approach if it is going to cope with the workload.
This is especially true in anatomic pathology. Pathologists examine tissue samples to determine whether or not they contain cancerous cells, usually under a microscope – as they have been doing for over a century. While well-trained and experienced pathologists will be able to discern the presence of those cells, anything that is done manually is going to be subject to error. Quality control is a challenge, and random second reads are quite labour intensive - requiring pathologists to repeat the entire diagnosis - and therefore not very effective on a broader scale. And, according to a National Institute of Health study, the rate of misdiagnosis in anatomic pathology is as high as 9% - meaning that as many as one out of ten patients are either getting the wrong treatment or being treated when they in fact should not be treated at all.
One way to reduce that error rate, and improve care for patients, is by implementing the use of advanced digital pathology technologies, and especially machine learning and artificial intelligence (AI), into the pathology toolbox. Using advanced artificial intelligence techniques, pathologists will be able to far more accurately determine if the tissue samples they are looking at indeed indicate the presence of cancer and provide more rapid diagnosis.
Advanced machine learning techniques use samples from large databases of biopsies to train an AI algorithm to identify patterns that will indicate whether or not a particular piece of tissue is cancerous. The larger the database used for training, the smarter the algorithm gets in “understanding” what it is looking for, and the more capable it is of making more accurate assessments. In recent studies, such AI algorithms demonstrated near perfect accuracy, making them a perfect companion to pathologists, helping them in the following ways:
Improved accuracy and quality control
With many pathologists overworked and basing their diagnosis solely on what they see on the microscope, AI can be essential for quality control. It ensures that despite the difficulties of the profession, errors are detected, traced, kept to a minimum and most importantly, that patients receive an accurate diagnosis. Our AI solution, Galen Prostate, is used in routine practice at pathology laboratories around the world, including the UK, Puerto Rico, France and Israel, and has already helped detect ‘false negative’ errors in prostate biopsies - cancers that pathologists missed and would have otherwise gone untreated. These labs can now offer their patients improved quality of diagnosis, with AI providing a safety net for 100% of the cases.
Better workflows
AI-based systems can make pathology workflows faster and far more efficient. They automatically classify and triage cases based on whether they are cancerous or benign, cutting down diagnosis time by highlighting suspicious areas for review. AI can also be used for singling out certain types of cancer for further analysis, routing more complex cases to expert pathologists, or triggering time-saving processes in the lab automatically, based on the algorithm's findings.
An important step forward in making such technologies more broadly available in the UK is a series of initiatives put forward by the Government recently, aimed at accelerating adoption of AI and digital pathology into the NHS. One such program is the PathLAKE consortium, built around a digital pathology centre of excellence created at the University Hospitals Coventry and Warwickshire, and tasked with leading digitisation projects at NHS laboratories and driving AI innovation in pathology. Under another initiative, the AI in Health and Care Award, led by the NIHR and NHSx, six hospitals will deploy an AI solution in their pathology departments and study its impact on clinical outcomes and health economics.
The ultimate vision for AI-based systems is to become decision support tools that will provide pathologists with the insights they need to do their jobs, saving them from having to deal with routine tasks and freeing them up to deal with the most serious cases that need special attention. In addition, advanced systems will be able to expand their roles, contributing even more to diagnosis and treatment as databases grow, streamlining both to ensure that patients get what they need when they need it.
Finally, using machine learning, future algorithms will be able to examine samples and make new connections – between images and medical conditions, between patterns and results – that will help pathologists make better and more accurate assessments. This will lead to new prognostics on disease progression, new and more effective treatments, more efficient and effective care, and a slew of other benefits.
Working with these tools, pathologists will have a more significant impact on cancer diagnosis and treatment, ensuring that patients receive the most effective care, with less time wasted and more resources (and money) saved. Pathology has never been an “easy” profession, and the current working conditions are making the job harder than ever. With advanced technologies, pathologists will find their workloads becoming more reasonable again – while enabling them to accomplish more than ever.
