Introduction

Colon and rectal endoscopic biopsies are used in the diagnosis of serious large bowel diseases such as colitis, Crohn’s disease and cancer. The diagnosis is currently done by a pathologist (a doctor trained to assess tissue based disease) examining tissue samples (biopsies) under a microscope. Colon and rectal endoscopic biopsies form a large volume of a pathologist’s workload, and around 30-50% of these samples are normal and contain no evidence of disease. In the context of ageing population and acute shortage of expert pathologists [1], this raises the need to look for novel ways to cope with the increasing large bowel screening pathology workload.

New technology means that tissue slides of biopsies can be scanned into a computer as a digital image. As part of the PathLAKE project (recently completed), the expertise of pathologists, clinicians, computer scientists and mathematicians were brought together to develop a computer program (algorithm) that by analysing these digital images, can distinguish normal colon tissue samples from abnormal ones, including cancer.

The algorithm (COBI) is run against the pixels from the biopsy images to find any irregularities that indicate the presence of disease. To the best of our knowledge, there is no other existing tool that can do that at present.

The COBI algorithm has been trained and tested on thousands of high-resolution images of large bowel biopsy slides and has shown high accuracy [2,3]. The next step is to take the final, fully optimised version of the COBI algorithm and conduct a large-scale multi-site validation (the NIHR-funded COBIx study) of its performance on biopsy images taken directly from other hospital labs. This is important because different labs have slightly different equipment, different ways of staining the tissue and different patient populations; it is important to test that the COBI algorithm works equally well for all these sites.

The twelve NHS Hospital Trusts from England and Scotland below are taking part in the COBIx study. These trusts already have the equipment to scan and digitise their biopsy samples.

·Cambridge University Hospital NHS Foundation Trust

·County Durham and Darlington NHS Foundation Trust

·NHS Greater Glasgow and Clyde

·North Tees and Hartlepool NHS Foundation Trust

·Oxford University Hospitals NHS Foundation Trust

·Royal Wolverhampton NHS Trust

·Sheffield Teaching Hospital NHS Foundation Trust

·Southampton NHS Foundation Trust

·University Hospitals Coventry and Warwickshire NHS Trust

·University Hospital Birmingham NHS Foundation Trust

The study sponsored by University Hospitals Coventry and Warwickshire in collaboration with Tissue Image Analytics (TIA) Centre and Warwick Clinical Trials Unit, both departments of the University of Warwick. The study is also partnered with Histofy Ltd, owners of the COBIx algorithm, who are providing their expertise to the project – follow this link to learn more about them.

Method

Over the duration of this project, we will collect images from approximately 10,000 large bowel biopsies from the twelve trusts, these will be analysed by the COBI algorithm and the results will be compared with the original pathologists’ diagnosis. The goal is to see if COBIx accurately separates normal large bowel biopsies from the abnormal biopsies. The aim is that in the future normal biopsies would be reported by the computer algorithm, while abnormal biopsies would be prioritised for examination by a pathologist. Furthermore, we wish to see if the detection of serious disease by COBIx is helpful in ensuring cases containing diseases, such as cancer or severe inflammation, can be fast-tracked for urgent pathologist review.

Due to screening programmes, pathology labs have seen a doubling of slide volumes in the last decade. If the COBIx algorithm is successful, it will speed up normal biopsy reporting and focus pathologists’ expertise on the cases that need them, leading to faster diagnoses for all patients.

Alignment with NHS Long term plan, NHSX or wider government priorities

This study will help deliver the aims of the NHS Long Term Plan [4], which set the goal for three-quarters of all cancers to be detected at an early stage, to diagnose cancer earlier and to support pathologists, clinicians and patients.

COBIx will help deliver earlier detection of cancer in the following ways:

·Through removing negative samples from the pathologists workload we allow resources to be focussed on the slides which contain disease. This means pathologists are not delayed in reviewing these slides by the need to examine negative slides ahead of them in the reporting queue;

·Slides which contain cancer are recognised by the algorithm and can be prioritised ahead of other non-urgent abnormal samples allowing them to be diagnosed sooner;

·Finally, slides containing small foci of cancer which could be missed by pathologists, will be detected by the algorithm reducing the risk of missed cancer biopsies.

COBIx will support pathologists in the following ways:

·Reducing the burden of reporting negative slides;

·Finding regions of slides which contain diagnostic features and thereby alerting them to important areas of the slide to examine;

·Reducing fatigue in screening large numbers of slides and helping to reduce error through oversight;

·Providing pathologists with a new tool to help deliver consistency in interpretation for challenging areas of diagnosis.

COBIx will support clinicians in the following ways:

·Ensuring patients under their care with serious clinical problems are diagnosed sooner because of the improved pathology workflow;

·Negative results being reported directly to patients can help reduce the number of incoming enquiries to the clinical team for results.

COBIx will support patients in the following ways:

·Patients with cancer and serious disease get their diagnoses sooner so that treatment can be started sooner;

·Patients without serious disease can be sent automated reports directly thereby helping to reduce the burden of worry, and the inconvenience of having to contact busy clinical services with enquiries for results, thereby improving the efficiency of the service;

·The safety of the service is improved leading to fewer errors.

COBIx will support healthcare providers in the following ways:

·COBIx is intended to be cost saving once it is rolled out to the NHS. The cost of implementation being off-set by savings in the pathologist workforce. By reducing the reporting workload in pathology there is the potential for these resources to be redirected to address other pressures;

·COBIx reduces a proportion of the pathologists’ workflow, reducing the need for human intervention and the associated energy demands of the workplace (travel, heating, lighting, etc);

·It will also contribute to more efficient and safer services for early detection of cancer, improving patient outcomes and reducing demand on resources (e.g., fewer appointments and inpatient stays) and the environmental impact of chemotherapy. Carbon emissions from medicines, chemicals and medical equipment account for 30% of the NHS Carbon Footprint Plus.

Benefits to patients, the NHS and the wider population

These types of algorithms, also known as artificial intelligence (AI), offers cellular pathology an exclusive co-pilot to achieve efficiency gains, quality improvements and patient safety. The COBIx algorithm will deliver the following benefits to the NHS, patients and wider population:

1. Improved efficiency through automating recognition of normal large bowel biopsies;

2. Quality improvements through triaging of biopsies with serious disease for urgent review and directing pathologist’s attention to the diagnostic foci detected;

3. Reduce errors through oversight of small diagnostic foci;

4. Faster reporting with normal reports going to patients as well as the clinician; Abnormal reports delivered more quickly by relieving pathologists of examining normal or negative slides and through having diagnostic regions of the slide highlighted in advance;

5. Health service providers benefit from a cost saving innovation focussing pathologist’s time on disease containing samples needing their attention;

6. As an exemplar of how cellular pathology reporting may be automated the project opens the prospect of deploying similar solutions in other areas of the cellular pathology;

7. The project demonstrates how well-curated patient data with appropriate access mechanisms can provide a revenue stream back to NHS sites through the licence agreements reached on the use of data held for the development of AI tools;

The wider population benefits from societal and economic gains due to accelerated adoption of AI in cellular pathology and the UK as a leading centre for this technology.

References

1.Pathology staff shortages causing delays to cancer diagnosis (CRUK, 2018).

2.Graham, S., Minhas, F., Bilal, M., Ali, M., Tsang, Y. W., Eastwood, M., ... & Snead, D., Rajpoot, N. (2023). Screening of normal endoscopic large bowel biopsies with interpretable graph learning: a retrospective study. Gut, 72(9), 1709-1721.

3.Bilal, M., Tsang, Y. W., Ali, M., Graham, S., Hero, E., Wahab, N., ... & Snead, D., Rajpoot, N. (2023). Development and validation of artificial intelligence-based prescreening of large-bowel biopsies taken in the UK and Portugal: a retrospective cohort study. The Lancet Digital Health, 5(11), e786-e797.

4.The NHS Long Term Plan (NHS, 2019).