Case study
Game-changing technology opens door to faster, more accurate cancer diagnoses
November 7, 2022
With support from EAHN™, Huron Digital Pathology and pathologists at Hamilton Health Sciences completed a validation study of a new image search technology for pathology. This unique use of AI promises to improve the speed and quality of cancer diagnoses.
Behind every cancer diagnosis, lies a pathology report—an expert assessment of cells or tissues that helps determine the type and extent of a cancer and the best treatment plan.
With most hospitals generating thousands of pathology slides each month, these slides and reports represent an immense body of knowledge that could lead to faster, more accurate diagnoses for patients. An Ontario company, Huron Digital Pathology, is poised to unlock that knowledge with the world’s first commercial image search engine for pathology.
The company’s software (Lagotto™) indexes archives of digitized biopsy slides, representing each as a series of distinctive barcodes that capture the key features of the cell or tissue sample. A pathologist reviewing a new slide can perform a search, and a sophisticated algorithm—a form of artificial intelligence (AI)— quickly compares barcodes to find similar images and the associated pathology reports and diagnoses.
This ability to immediately access the expertise of colleagues when reviewing images is significant; pathologists, particularly those with sub-specialty expertise, are often in short supply and older pathologists are retiring. Cancer cases are also increasing as Canada’s population ages. Requesting a consultation with a colleague to assess a difficult case could delay a diagnosis by weeks. Reviewing multiple similar cases also helps pathologists improve the accuracy of their own diagnoses.
Key to the technology’s success is its ability to search a large database quickly. Scanned pathology images are typically very large—one image can be a gigabyte of data—making it difficult to search images quickly. By using barcodes, the software can convert these images into just 10 to 15 kilobytes of data. As a result, Lagotto™ can search up to 100,000 images a second.
EAHN™ project offers opportunity to validate technology
Not surprisingly, the value of the search technology depends entirely on its accuracy in finding similar images. By 2020, the company had validated its search algorithms using a publicly available data set of 30,000 images from the US National Health Institute/National Cancer Institute. The next step was applying the technology in a clinical environment. EAHN™ provided that opportunity.
Through EAHN™, companies with innovative health technologies are matched with health organizations to develop, test, refine, adopt, promote, and disseminate novel technologies. In this case, Huron Digital Pathology was matched with Hamilton Health Sciences. Working with the hospital’s IT support and a lead pathologist, the company installed a high-quality scanner and scanned and indexed 1000 pathology slides. The search engine was then used to conduct searches and several hematopathologists (specialists in blood and bone marrow) provided feedback on the accuracy of its matches.
Study provides important feedback to enhance search algorithms
The results were not as expected, and the team soon realized the reason. The search algorithm assumes large data sets of 10,000 images or more; the 1000 images scanned for the project were insufficient to generate accurate results. It was valuable learning, and Huron Digital Pathology is modifying its algorithms to obtain higher accuracy with fewer images. The project also helped the company consider how to implement the technology on a much larger scale.
For Huron Digital Pathology, the experience demonstrates the value of EAHN’s program. The development and refinement of innovative technology depends on working with collaborative partners who are open to and engaged in the problem-solving process. Hamilton Health Sciences was that kind of partner.
The hospital is also positive about the experience. EAHN™ projects provide the opportunity for the hospital to access high potential technology from companies that have been pre-vetted and without going through a traditional procurement process. EAHN’s contract and project management addressed privacy, security and other issues that routinely require significant staff time. And the funding provided by EAHN™ removed any financial barrier that might deter the hospital from taking on this kind of project.
Leading the shift to digital pathology
Digital pathology is relatively new, and the shift from microscope to reviewing digitized slides on a computer screen will take some time. Radiology digitized in the 1980s and it was a decade before many hospitals had a fully digital workflow in place. Pathology has its own challenges. One cancer diagnosis can involve ten to 200 slides, and some hospitals would need to digitally scan up to 500,000 slides a year.
Currently, digital pathology is more common in the US and some parts of Europe. In Canada, Quebec is moving in that direction and Huron Digital Pathology expects larger hospitals in other provinces to follow. The efficiencies of a digital workflow make the change inevitable and the Lagotto™ platform and other AI applications point to the possibilities that digitization brings. The company estimates the diagnostic imaging market to be $10+ billion.
For now, Huron Digital Pathology will continue to validate and evolve its algorithms to improve accuracy, particularly as it scales up to larger data sets. A contract with the US Joint Pathology Centre will help them in this process. The Centre is the pathology tissue repository for the US Department of Defense and Department of Veterans Affairs, and Huron Digital Pathology is scanning and indexing their 55 million slides as the Centre moves to digitize its holdings. The Centre will use the company’s image search technology to enable data sharing with researchers, clinicians and educators, providing access to 100 years of medical and scientific knowledge.
Get in touch 