TechWatch recently caught up with the R&D manager for PathXL, David McCleary, to discuss how AI is used in its digital pathology solution.
AI is critical for PathXL’s flagship tumour detection product, TissueMark, which automates the process of analysing human tissue samples including biopsies for cancer tumour cells. The system learns how to recognise the patterns of tumour cells, to differentiate them from healthy cells. 
PathXL doesn’t put pathologists out of a job, but rather, works with them in the lab to improve outcomes in the part of their job that is prone to human error – detecting where tumour is on a slide, and counting the number of tumour nuclei. Pathologists need to estimate how many tumour cells are present in a sample, which can have a huge impact on how the patient is treated. This can be done far more accurately with an automated system like TissueMark.
TissueMark can help improve the quality and consistency of molecular results and save significant time and cost over a human-only workflow. The system is critical for the new age of medicine – all about delivering molecular testing and personalised medicine. This blog post written by PathXL discusses the need for automation in this area, with one study showing huge variation in pathologists’ evaluation of tissue samples – variation which could impact on test quality and patient care.
In the below gallery of images of a lung sample, ‘original’ is the original tissue sample, ‘boundary’ is the boundary around the tumour, and ‘heatmap’ is a boundary plus a detailed heatmap showing more precise regions of tumour. Areas in red are tumour and blue are normal.
“Once we reach the desired level of statistical significance, we ‘lock down’ the system and release an updated version of Tissuemark. This way, the technology is always ‘learning’ but it’s directed, and thus controlled, learning.”
To achieve this controlled learning, one of the technologies that PathXL makes use of is support vector machines – supervised learning models with associated algorithms that analyse data used for classification and regression analysis.
Are other companies doing this or is it unique in NI? “Others are focused on pattern recognition using IHC (special stains), but TissueMark is unique in Northern Ireland and highly differentiated from any other approach globally. TissueMark focuses on analysing H&E stained slides – the most widely-used stain, and the technology follows on from research at Queen’s University Belfast. The product is selling into a wide variety of cancer research organisations and medical centres in the US, UK and Europe,” said David.
PathXL was founded in 2004 and received Deloitte’s Irish Technology Fast 50 Award four years in a row (2012-’15). It has been regularly funded through InvestNI, and has been awarded joint R&D grants with Almac. Local VC firms have provided support, including Crescent Capital and Qubis.
Is the future of AI bright? David said, “Self-learning systems are a buzzword right now, but I think more work should be done in the fight against cancer using directed learning systems. Applied in medicine, directed learning solutions can deliver vast improvements in patient outcomes and help the NHS become more efficient. These systems can assist doctors and pathologists to speed up their delivery of medical care, reducing waiting times and crucially, giving more accurate results to patients who need a diagnosis.”
TissueMark is giving patients peace of mind that their diagnosis is spot on. It’s also a great example of humans working in concert with AI technology.
By Emily McDaid
Read more from TechWatch
