Chaim Linhart, PhD, co-founder and CTO, Ibex Medical Analytics, writes about the development of uses of artificial intelligence in cancer services.
MOR_RDT
Pathology serves as the cornerstone of modern medicine, supporting all fields of healthcare. With innovative, digital diagnostic technologies, pathologists can help clinical teams managing complex conditions to provide better treatment plans for patients.
Cancer remains one of the most challenging diseases we face, resulting in millions of lives lost each year. As a disease which becomes more difficult to treat if not detected early enough, accurate and timely diagnosis of biopsies is crucial for a patient’s survival. Pathologists possess the invaluable training and unique skills needed for diagnosis. However, over the last decade in the United States and in other countries, cancer cases continue to rise while the number of pathologists has significantly declined. As a result, there are fewer practicing pathologists, as the demand for high-quality, rapid diagnosis rises.
We have seen the steady adoption of automated digital tools across different medical practices, and this can hint to the future of pathology. However, while digital transformation in pathology holds promise, its pace across the industry has been slow. A key factor at play is that digital pathology tools remain limited in the scope of decision support, quality enhancement and productivity gain they can offer, while requiring significant investment in equipment, new lab workflows and training.
Considering the range of tasks pathologists are responsible for, a digital solution in pathology can bring considerably more value by including advanced computational technologies and well-trained AI algorithms. AI-based tools span a gamut of applications including simple tasks such as tumour sizing and cell counting, or technical tasks such as ordering additional staining protocols for certain biopsies. An accurate AI algorithm trained to detect cancer can also be helpful in triaging malignant cases for a more prompt review. Moreover, pathologists who review cases using AI were found to improve their overall accuracy, reduce misdiagnosis rates and lower turnaround times. By taking advantage of such capabilities, pathologists’ work can be considerably optimised, and they can focus their expertise on diagnosing complex and rare cases.
The field of pathology is undergoing a monumental shift as pathologists take an active role in working with software developers and data scientists to create AI solutions across the pathology workflow. This collaboration is crucial to foster digitisation and enable algorithms that correspond to how pathologists are trained to diagnose disease. Physicians need to develop a high level of trust in each new medical technology before they agree to use it and therefore having pathologists directly involved in research and development can help accelerate their adoption in the laboratory.
Artificial Intelligence: The opportunities before us
Unlike other areas of healthcare where AI may be more easily adopted and applied, pathology presents a unique opportunity in which the final product requires direct involvement from the end-user to ensure its continued evolution and development. There are several areas where this applies.
Pathology images generated from scanning glass slides are huge, typically containing billions of pixels. Moreover, they have unique characteristics in terms of colors, shapes, orientation and spatial relationships between the various cells and tissue structures – making them very different from other medical imaging technologies, such as X-ray and CT scans. To create an AI tool that can analyse and diagnose these complex images, developers must use sophisticated, cutting-edge techniques, such as Deep Learning models, and train these algorithms on massive and diverse datasets that include rare cases. Pathology departments with large and accessible archives play an important role in sharing such datasets with developers.
As these datasets evolve, pathologists can enrich their algorithms to identify different subtypes of cancer and train them to go beyond just cancer detection to include cancer grading and tumour sizing, in addition to identifying other clinically relevant features. These algorithms can then be rigorously tested on new datasets to validate their accuracy by comparing it to a ground truth call made by expert pathologists.
What is most exciting is that as these tools are now rolling out in pathology labs around the world, they enable developers to collect real-time user feedback, directly contributing to the continued evolution of these programs. The systematic, repetitive nature of pathology work makes it easier for pathologists to give feedback on how the AI fits into their diagnostic workflows, prompting subsequent training to fit their needs. What’s no less exciting, is to note the growing number of patients who have already benefited from the fact that their biopsies were diagnosed by a pathologist that was assisted by an AI algorithm, contributing to the quality of diagnosis, patient safety and overall level of care.
The field of pathology is on the cusp of radical transformation as pathologists play an active role in developing, and refining AI algorithms to advance their practice. Coopetition and collaboration between care systems, pathology departments and software developers are creating a bright future for pathology, as innovative technologies help meet the challenges of 21st century healthcare. We have only just begun to scratch the surface and there is much more to be done.
