From my vantage point in pathology, I’ve seen the significant patient benefit AI can bring. I’m happy to share some examples here, as well as my thoughts on the strategies we need to adopt to continue to advance.
Muhammad Aslam
Consultant Histopathologist, Betsi Cadwaladr University Health Board
How pathology has been leading AI adoption
As well as my role as a consultant, I am clinical lead for the NHS Executive’s All-Wales Digital Pathology and AI project. In this guise, I’ve developed a strong appreciation for how AI can improve our patient pathways. In particular, pathology has been leading the way in adoption in Wales.
Four solutions using AI that are in use, or we’re actively working on, are:
Ibex Prostate, which scans samples before the pathologist and uses a traffic light system to rate possible abnormalities. Our trial in north Wales saw a 13% increase in cancer detection, and we quickly moved to a national rollout with funding from Welsh Government.
The North Wales breast AI project quickly followed, tackling the complexities of breast cancer diagnosis and treatment. Using an AI tool to predict diagnosis and automatically request further tests for the best treatment options can shorten the usual process by around 1.5 days, ensuring the patient has quick confirmation of their treatment plan by the Multi-Disciplinary Cancer Team. This has also moved to a national rollout, underway right now.
The upper GI cancer AI project has similar goals to the prostate and breast cancer projects, but for stomach biopsies. This project pilot is complete and the data is under evaluation currently.
The Pan-Cancer AI deployment project is in its early stages but could save many patients who would otherwise have lengthy delays in their diagnosis. Nearly 20% of referrals originally classed ‘non-urgent, non-cancer’ turn out to be malignant after histological examination. The tool can triage samples across all cancers and, where abnormalities are found, upgrade to urgent cancer pathway.
Strategies for success
There are a number of strategies we must adopt in Wales if the opportunities for AI to improve health and social care provision are to grow.
National digital pathology procurement
The success in AI adoption in pathology comes despite a serious barrier. Pathology is behind other areas when it comes to digitisation. 80% of our workload still uses glass microscopy – a person analysing a physical glass slide – with no digitisation of the myriad of data that sample could offer. It will be impossible to adopt AI tools without progressing the business case for national procurement of a digital pathology system. But, the good news is that we are in right direction and hoping to get business case approved sooner. A central source of funding will, of course, be essential to this.
A consistent digital workflow
Across many areas – and of particular relevance in pathology – there is a mix of analogue and digital workflows. In many cases, whether a patient’s case is digitised (and what stage of the process this happens) is based on a combination of location and service provision. Switching back and forth between different workflows can lead to errors, omissions and inefficiencies.
Consistent digital workflows will allow us to employ AI, as well as more basic algorithmic tools that can spot errors and omissions in a set of samples.
Adoption of an approach to consent for data use
AI tools cannot be developed without data. And to develop tools that properly address the needs of our population, they should be trained on data from our own patients. As it stands, however, we don’t have an accepted approach to consent for using a patient’s anonymised data to train AI. This leaves us reliant on suppliers who develop training tools using other populations.
By developing clear, transparent guidelines and processes for gaining consent, we can both train new tools and audit existing tools, to provide better care for our patients.
Build awareness and engagement with clinicians and the public
My experience of seeing AI deployed in pathology shows how swiftly clinicians accept tools when presented with evidence of the benefits. Colleagues immediately appreciated AI’s ability to supplement the work of a pathologist and get the most out of the available material for patients. This high level of acceptance can be replicated across other areas, as awareness of the potential benefits AI can bring grows.
Similarly, when patients have seen the faster diagnosis we’ve achieved in prostate cancer, they’ve demonstrated strong acceptance of the tool.
Multi-disciplinary working groups and open, transparent engagement with patients will be essential to more widespread adoption.
Infrastructure improvements
Access to high-speed internet is critical to support the use of AI tools. Currently, a minimum speed of 25 megabits per second is recommended and this will only increase as technology advances. In some areas, centres struggle to keep up with current usage, so IT network infrastructure needs to be strengthened to ensure everyone has equal access.
Our priorities
I’ve primarily focused on pathology here, not only because it’s my area of expertise, but because pathology is so vital to healthcare. Without pathology, we can’t diagnose, and without diagnosis we can’t treat.
Pathology is also an area where AI can offer gains extremely quickly. Pattern recognition in images has been one of the fastest advancing areas of AI, so, the basis of most of our work in pathology is a prime candidate for support from AI.
As well as this, the potential for developing and training new tools for pathology is huge. One typical slide translates to around 5-6GB of data. As we know, AI thrives on data, so if we can use the data we collect in pathology to train tools, we can quickly reap significant rewards.
To achieve this, a 100% digital process will be key and I have a personal ambition that we’ll achieve this by 2026.
To learn more about the innovative solutions being used in cancer detection, contact Muhammad at muhammad.aslam3@wales.nhs.uk.