In the European Society of Uropathology Symposium (ESUP) on Day 1 of EMUC24, expert speakers delivered updates on the clinical application of artificial intelligence in genitourinary cancer pathology for tumour classification, slide triage, quality control and improving biomarker evaluation. The session was chaired by Prof. Maurizio Colecchia (IT).
Dr. Yuri Tolkach (DE) presented a lecture on ‘AI in genitourinary cancer pathology’, focusing on the clinical relevance. He emphasised that there is work to be done before using AI tools for pathology, and even though digital pathology uses 50% less time, there are a lot of costs involved in the setup, the integration can be very difficult, and there are no reimbursement mechanisms for AI tools. “We need some support from clinical disciplines as this is something that urologists can help us with a lot, and other colleagues from oncology, because the application of AI tools brings a lot of benefits to the patients.”
According to Dr. Tolkach, “AI has virtually no limits in pathology but there is still only a handful of tools (which is the same as 3 to 4 years ago), there is still very limited use, mostly because we are not digitalised enough yet. There is a lot of ‘AI hype’ and lots of studies that have no clinical application.”
He cites the results of his digital pathology validation study (Tolkach Y, et. al. NPJ Precision Oncology, 2023). From over 7,000 biopsy cores, there was no 100% accuracy, but the sensitivity of the tools was very high, which is useful for alerting pathologists to areas considered suspicious, rather than the obvious.
A new access to tumour complexity?
“There are huge efforts ongoing, especially in the industry,” stated pathologist Dr. Markus Eckstein (DE) in his lecture on ‘Spatial transcriptomics/proteomics: How AI could improve our understanding of tumour complexity.’ He shared several methods with a review of their resolution, complexity and price, including array-based transcriptome, digital microdissection, mass spectrometry, CyTOF, High Plex ISH and RNA Scope. “All of these methods have pros and cons. You must know what spatial biology study you want to perform and then you can choose the best option.”
Dr. Eckstein was enthusiastic about the discovery earlier this year of the spatial niche interactions from Cellcharter (Varrone et al, Nature Genetics 2024). Cellcharter is an algorithmic framework to identify, characterise, and compare cellular niches in spatially resolved datasets.
Extending diagnostic capabilities
“There is promise in AI and pathology but there are also barriers to overcome, but they are not brick walls,” explained Dr. Gladell Paner (US) in his lecture on ‘Artificial intelligence in grading of urological cancers.’ Firstly, he clarified that “AI in the grading of prostate cancer is architecture-based and AI grading of bladder cancer is cytomorphology-based.” Dr. Paner shared several research studies with results indicating “AI has a comparable accuracy and reproductivity in the grading of PCa with uropathologists. It can be used to enhance the grading of PCa by assisting pathologists, and can identify other grade-derived elements, such as cribriform pattern. For BCa, AI can help identify objective morphometric features for grading BCa.”
According to Dr. Paner there are still challenges in the advancement of AI-based histopathology, and a lengthy list of limitations. “None of the current available AI algorithms are 100% perfect and they need human supervision. It is not superior to a specialist GU pathologist, and the digital pathology workflow for integration of AI-based diagnostics is a costly set up.”
You can watch the full ESUP presentations on the EMUC24 Resource Centre.
