Computational pathology is, among other things, about how technology can support specialists with exceptional expertise in diagnosis. In the future, artificial intelligence (AI) is expected to improve workflow and drive accuracy in both disease detection and diagnosis, a leading expert told Guido Gebhardt during the German Society of Pathology’s annual meeting last May in Munich.
Before AI can help to produce a diagnosis from a tissue sample in a pathology laboratory, technology is necessary – special hardware such as scanners and special workstations, software and interfaces, a proper network for high internet speed and a fair amount of storage space.
It may sound trivial, but costs related to the investment can rise up exponentially, according to Peter Schüffler, a professor of Computational Pathology at the Institute of General Pathology and Pathological Anatomy at the Technical University (TU) of Munich.
“For a large pathology department with a high throughput, we’re talking about seven-figure investment costs,” said Schüffler, a bioinformatician who has created a fully digitized pathology department at his institution. This is virtually unique in Germany and it is setting the foundations of an entirely new world, in which pathologists can work much more efficiently to improve patient care.
Digital Slides
Over the past two years, Schüffler and his team have created the technological prerequisites for analyzing tissue with AI and digitizing the pathology workflow.
The digitization of sectional specimens using so-called slide scanners is particularly important in the overall process, as the aim is to create whole slide images (WSIs).
At the Institute of General Pathology and Pathological Anatomy at TU Munich, eight scanners are used to generate digital images. In addition, suitable workstations for pathologists as well as a redundant, fail-safe image management system, interfaces to the laboratory information system and sufficient storage space were procured and seamlessly implemented.
External cloud solutions to store digital images also require a lot of storage space. “As cloud solutions are not legally possible at the moment, we have procured servers, and I hope that this is only temporary,” he said.
Despite the hurdles, the pathology department in Munich is a pioneer and has received support from relevant departments at the TU Munich and the Rechts der Isar Hospital, who are open to innovation and provide human and financial resources.
With AI to the Primary Tumor
Following the digitalization of the workflow, the Institute of Pathology is now training AI models on new computers with state-of-the-art graphics processors, in addition to its own high-performance cluster.
In the future, these will be able to analyze tissue sections, generate predictions on prognosis or treatment response, or quantitatively record objects. However, it will take some time before AI learns to make verifiable and comprehensible statements, Schüffler explained.
The aim is to have algorithms as soon as possible that can help pathologists with difficult cases, such as cancer with an unknown primary tumor. “It is not at all uncommon for metastases to be detected without being able to determine the associated original tumor,” he said.
Schüffler is convinced that AI-supported diagnostics will help solve problems in the coming years. AI recognizes and combines patterns that humans cannot see with the naked eye. In the near future, a whole range of algorithms will be available that can be used in pathology laboratories for diagnosis and therapy selection.
