Evidence
Alert

Digital analysis of cells and tissue was as reliable as using a microscope in a new review

Diagnoses made using digital scans of cell and tissue samples are as reliable as looking at traditional slides through a microscope, a new review suggests. It provides the strongest evidence to date that digital techniques are a viable alternative to microscopy and could in future replace it.

Small samples of body tissue (biopsies) are sometimes taken from patients to diagnose disease. They are prepared on a glass slide. Cancer cells look and behave differently from normal cells; these differences can be viewed directly down a microscope. Alternatively, new digital techniques allow slides to be scanned and then examined on a computer screen (digital pathology).

Researchers pooled the evidence from previous studies in the largest systematic review carried out so far. They found that diagnosis based on digital pathology almost always agreed with microscopy (more than 98% agreement). The study also highlights specific areas where more research is needed.

Digitisation has the potential to make pathology labs more efficient and the results of biopsies more accurate. But only a small number of UK labs currently use the technology. The new results could help convince others to make the switch. 

What’s the issue?

In digital pathology, glass slides of biopsies are scanned at high resolution by a computer. The resulting detailed images are analysed on screen. In principle, this would help pathologists (doctors who specialise in the process of diagnosing diseases) look at more samples and make results more consistent. It also makes automation possible, and perhaps the use of machine learning algorithms in future.

But many pathologists do not yet trust digital techniques to do a reliable job. They still prefer to look at slides under a microscope. That is partly because, until this new study, there was no large-scale comparison of the results obtained using the new and traditional methods.

What’s new?

This review pooled the results of 25 studies that compared digital pathology with microscopy to diagnose a range of conditions across the body. It checked whether the diagnoses offered by each method agreed.

Agreement in this review means that either the results were the same or differed only slightly; the diagnosis would be the same with either technique. Analysis of 24 of the studies showed a high level of agreement overall:

  • average agreement was 98.3%
  • three studies showed 100% agreement
  • in 23 of the 24 studies, digital and traditional analysis agreed for 95% or more of the samples.

There was some disagreement which could lead to a difference in the treatment offered to patients. The review found:

  • across the 25 studies, a total of 546 disagreements between the two methods were reported
  • disagreement was most likely for gastrointestinal tract and gynaecological (of womb, cervix or vagina) samples, followed by skin, breast, genitourinary (including bladder, prostate and testis) and kidney samples
  • more than half (57%) the disagreements related to the assessment of cell nuclei (which contain most of the cell’s genetic material), or to the grading of abnormal cells (how quickly and aggressively they may grow or spread)
  • other common disagreements involved difficult diagnoses, such as melanoma (skin cancer), and spotting small objects including infection with micro-organisms.

Why is this important?

Despite the opportunities offered by digital techniques, only about 15% of pathology labs in the UK have embraced them. The new study could build confidence in digital diagnoses and help convince more labs that digital pathology is as reliable as conventional microscopy. The review also helps make the business case for investment in the technology. This is particularly relevant as the ongoing COVID-19 pandemic has disrupted workplaces and led to many more people working from home rather than in centralised labs.

The study highlights some sample types for which the digital technique and microscopy give different diagnoses. These discrepancies are not all a result of the method used. Pathologists routinely differ on how to grade some cancer cells and how to judge a cell’s features under a microscope, because this interpretation includes subjective opinion.

What’s next?

Digital pathology is not currently used to assess the single-largest source of samples in the UK: cancer screening. Although research studies have suggested such work can be done digitally, Public Health England is yet to approve this method. The researchers now want to extend their research into cancer.

The review represents a broad sweep of clinical practice and the majority of studies had a low risk of bias. But more work is needed in specific areas. No comparisons were available for analysis of ophthalmic samples, and too few studies examined renal or paediatric samples to draw meaningful conclusions.

The study also identifies some pathology specialisms that might not be suited to current digital techniques. The build-up of a rogue protein called amyloidosis may be easier to diagnose with a traditional microscope which has polarised light filters and allows an appreciation of the protein’s texture. Identifying tiny objects such as bacteria may also be more difficult with digital techniques which have a single depth of focus and lose detail when the image is created. The review suggests a need for continued improvement in digital techniques and for new non-digital techniques to be developed.

You may be interested to read

The full paper: Azam A, and others. Diagnostic concordance and discordance in digital pathology: a systematic review and meta-analysis. Journal of Clinical Pathology. 2020. doi: 10.1136/jclinpath-2020-206764

More on digital pathology from the Royal College of Pathologists

The Digital Pathology Association, a non-profit organisation, that aims to improve awareness, education and the adoption of digital pathology in healthcare and the life sciences

Funding

This project was funded by NIHR Programme grants for Health Technology Assessment (HTA).

Commentaries

Study author

I work in a pathology lab that was the first in the country to start using digital pathology for routine clinical diagnosis in 2014 and was recognised as ‘The Digital Pathology Centre of Excellence.’ My experience has been really good. My team sees all the benefits that digital pathology enables, including efficient workflow, remote access to the slides and workload via secure network, ease of access in obtaining second opinion from the colleagues, as well as exploring the role of computer assisted analysis using machine learning. And now this research has backed that up by offering substantial evidence that digital pathology is as reliable as the microscope.

Ayesha Azam, Cellular pathologist & Digital Pathology Research Fellow, University Hospitals Coventry and Warwickshire NHS Trust

Member of the public

The paper provides evidence that digital tools have good accuracy for many pathology applications. The number of cases seen is not that large and the demographic of cases, though similar to the NHS’, is not identical. But it suggests that this technology is ready for more widespread trial leading to its gradual adoption. This should help build trust in digital tools among GPs and clinical leads in hospitals.

Knowing that a patient’s pathology sample was analysed digitally should indicate that the analysis was more systematic. Digital tools will be quicker and, for many repetitive mistake-prone tasks, will perform better than humans. They also have economic advantages.

What matters and needs to be demonstrated is that digital tools reliably detect borderline cases which require multiple opinions to diagnose. Careful study of how well digital pathology tools perform in practice will build patients’ confidence. However, inevitably there will be misdiagnoses and the media will have a field day when a machine rather than an expert person can be blamed.

Nick Filer, Computer scientist and Public Contributor, Manchester

Researcher 

This paper shows the use and potential implementation of digital pathology in diagnosing cancer, but it is easily applicable across other disease areas. The research will promote professional acceptance of digital pathology and is a step towards using machine learning for diagnosis in future. This will reduce costs across the pathology and diagnosis spectrum and in turn lead to wider use and acceptance of digital pathology.

Paul Townsend, Pro-Vice-Chancellor and Executive Dean of the Faculty of Health and Medical Sciences, University of Surrey

Conflicts of Interest

One author has received fees from an equipment supplier.

Paul Townsend is a co-founder or consultant for various biotechnology companies.