Cookies

Like most websites The Pathologist uses cookies. In order to deliver a personalized, responsive service and to improve the site, we remember and store information about how you use it. Learn more.
Inside the Lab Digital and computational pathology

Beyond Digital

Pathology is involved in two-thirds of all diagnoses made in healthcare systems such as the UK’s National Health Service – and an estimated 95 percent of clinical pathways rely on patients having access to efficient, timely, and cost-effective pathology services (1). However, the classical histopathology workflow from biopsy to a diagnostic report (Figure 1) takes up to 10 days on average in the US (2) and 14 days in the UK (3) – an excruciating wait for patients and their families.

Figure 1: The classical histopathology workflow.

The workload for diagnostic services will only continue to increase. On one hand, there is an increase in demand with the growing and aging population in the UK, and advancements in early detection and treatment pathways resulting in a predicted 28.4 million cancer cases in 2040, a nearly 50 percent rise from 2020 (4) – meaning that one in two people are expected to receive a cancer diagnosis in their lifetime (5). On the other hand, the number of practicing pathologists is declining. A 2018 workforce census from the Royal College of Pathologists showed that a quarter of all histopathologists are over 55, most of whom are expected to retire by 2023 (6). Furthermore, an all-time low number of trainee doctors are choosing to specialize in pathology with only 3 percent of NHS histopathology departments having enough staff to meet clinical demand (6).

To add to the problem, the COVID-19 pandemic has significantly disrupted healthcare systems with the shutdown of nonessential services and drastic changes in patient behavior. By March 2021, 4.7 million patients were waiting for treatment (7). An additional £2 billion per year is needed to recover this backlog over the next three years, requiring an estimated 11 percent increase in NHS activity – that is, an extra 4,000 consultants and 17,000 nurses per year (7). With pathology involved in 95 percent of clinical pathways and diagnostics, the pandemic has exacerbated the pressure on pathology departments.

The rise of digital pathology 

Telepathology and digital pathology emerged in the 2000s (8). The development and introduction of whole-slide imaging (WSI) enabled pathologists to “read” digital images on a computer screen instead of on a physical slide under a microscope (Figure 2).

Figure 2. The digital pathology workflow.

The digitization of histology slides with remote image access offers several immediate benefits. It allows pathologists to report anytime, anywhere. This in turn reduces delays associated with the transportation of glass slides, improves laboratory workflow with reduced costs and increased workforce capacity, and provides pathologists with easier access to colleagues for second and expert opinions. Datasets of annotated digital images can also serve as valuable platforms for training junior pathologists and developing automated tools that increasingly support and streamline clinical decision-making. More recently, the COVID-19 pandemic has essentially demanded the adoption of digital pathology due to restricted access to pathology labs and a need for remote work. However, despite these benefits, the digitization of pathology has been slow, business cases have failed, and implementation projects have stalled (9,10).

The cause of the stall

For the past 15 years, digital pathology has promised to change pathologist workflows. However, efforts to digitize pathology are yet to yield the promised increases in operational efficiency (10). This is primarily due to barriers in clinical adoption, infrastructure implementation, and operational excellence. These barriers are reflected in the underwhelming digital transformation of pathology laboratories, with only 31 percent of healthcare providers across the UK starting to invest in applying digital pathology technologies to steps in their clinical diagnosis workflows (11).

To deliver the promises of digital pathology, key stakeholders across the clinical workflow must work together to facilitate clinical adoption. Implementation and optimization of digital pathology is typically initiated and supported by laboratory managers and lead pathologists as a collaboration. Hospital laboratories invest in the hardware and software, as well as biomedical and IT staff to manage and deliver this digital infrastructure. Pathologists and clinicians define how the technology is integrated into clinical workflows and decision-making. Thus, digital pathology companies must play their part in providing first-class customer success programs and staff training schemes to encourage clinical buy-in and adoption.

Moreover, the implementation of digital pathology in a hospital or laboratory first requires the setup and integration of back-end infrastructure such as scanning equipment, image viewer software, and network capacity to store images. Commercial solutions in digital pathology currently typically center on software for workflow management and operations for pathology laboratories in hospitals. However, such solutions do not account for variations between scanner hardware and image annotation techniques, hindering technology acceptance by laboratory staff and contributing to interobserver variability between consultant pathologists. To overcome this barrier, companies must ensure that their solutions integrate with on-site infrastructure, as well as provide training programs and quality assurance schemes for reporting pathologists and laboratory staff.

A recent evaluation of traditional analogue and digital pathology workflows concluded that the logistical savings from digital pathology would not be enough for a financial business case (10). The researchers compared the impact of traditional analogue versus digital pathology on the efficiency of five laboratory workflows, concluding that digital pathology saved over 19 hours on an average day across all laboratory workflows. Critically, however, remote reporting on digital images instead of glass slides saved the pathologist only one hour per day.

Altogether, the uptake of digital pathology has stalled because the digitization of glass slides alone does not resolve the pressures of an increasing workload on a diminishing workforce of pathologists. Instead, digital pathology solutions add to hospital overheads in operating and capital expenditures, weakening their own business case. Notably, if the integration of digital pathology infrastructure stalls diagnostic workflows, this inevitably results in an increasingly volatile workload and pressure for reporting pathologists. This backlog may be outsourced to company laboratories who offer a centralized resource of pathologists in a hub-and-spoke manner. Moving forward, to supply and scale pathology services in response to increasing demands, we need a fully managed diagnostic service that integrates digital pathology infrastructure with the automation of both laboratory and pathologist workflows (Figure 3).

Figure 3. The key barriers in the implementation of digital pathology technologies (in circles) and their resolutions (in overlaps).

The computational promise

To relieve the pressure on pathologists, we need to streamline clinical decision-making. Repetitive pattern recognition tasks in pathology, such as cell counting or object classification, are ideally suited to number-crunching computers. This would allow us to reserve our human expert pathologists for more nuanced tasks and cases. However, this baseline computation depends on data that are captured and generated through digital pathology infrastructure. Just as the Internet provides a foundation on which to improve the efficiency of communication, digital pathology is the foundation on which we can build computer-automated tools to support and streamline clinical decisions.

This emerging field of “computational pathology” leverages artificial intelligence (AI) technology for diagnostic pathology by extracting information from digital images with a “big data” approach – using mathematical models to make diagnostic inferences and presenting clinically actionable recommendations for pathologists and clinicians (Figure 4).

Figure 4. The computational pathology workflow.

Over the past six years, two defining computational pathology models have emerged in academic research and are increasingly being deployed in the clinic:

  1. Semi-automated review: providing routine and repetitive tasks, such as cell counting and cell classification, to help categorize and triage samples for review by a human expert pathologist.
  2. Fully automated review: providing grade scoring, outcome prediction, and survival analysis to provide second opinions for pathologists and perhaps soon even diagnostic reports for clinicians.

Research across academia and industry demonstrates how AI tools can be developed for clinical decision support across different pathologies and disease types. For example, Grand Challenges in Pathology encourage multidisciplinary teams to build and optimize AI tools to tackle fundamental steps in routine pathology reporting that currently depend on human experts. The AI tools developed to tackle areas of unmet need will increasingly move from an initial validation environment through real-world evaluation to regulatory approval and national clinical development.

One such Grand Challenge seeks to streamline prostate cancer biopsy grading. Gleason scoring is a strong predictor of patient prognosis, but is challenging and time-consuming due to tumor heterogeneity. This leads to significant variability between the conclusions reached by expert pathologists. Notably, in September 2021, the FDA approved the first AI tool to increase the rate of prostate cancer detection on digital slides by identifying and highlighting areas of interest for pathologists (12). The tool is now also being evaluated by the UK’s National Health Service through a national implementation pilot.

Similarly, the Cytosponge diagnostic test is being evaluated and rolled out by the NHS to drive the earlier detection of esophageal cancer. Here, pathologists review digital images of over three million esophageal cells, a resource-intensive and repetitive process. Research demonstrated that, for diagnostic testing of cell samples for early esophageal cancer, a semi-automated review can reduce pathologist workload by 57 percent while maintaining the diagnostic performance of human expert pathologists (13). This will enable the test to scale to a national level without placing unsustainable pressure on our pathologist workforce.

Notably, fully automated models for computational pathology are still being validated in research. Thus, such tools are unlikely to materialize anytime soon – likely not until perceived barriers in regulation, ethics, human-AI interaction, and technology implementation are addressed. These primarily revolve around issues of trust, public acceptance, transparency, and explainability. To put it into perspective, we are more likely to see self-driving cars globally accepted and in use before fully automated pathological review.

Issues and considerations surrounding patient-AI interaction models are being explored and debated by think tanks around the world. For example, to what extent should semi- and fully automated models be used – to provide a second opinion for pathologists or an independent test score for clinicians? Clearly, such ethical and regulatory considerations impact healthcare pathways and must be addressed to deliver the future of patient-centered medicine.

Computational pathology offers to relieve pressures on reporting pathologists by increasing throughput without compromising accuracy. Digital pathology’s impact on the workload overwhelming a diminishing workforce of pathologists may be limited – but computational pathology can help.

The key

Although computational pathology is the answer to relieving the pressure on pathologists, the key is in digital pathology infrastructure – the backbone that enables computational pathology (Figure 5). By ensuring a fully managed digital pathology service from biopsy to laboratory processing to reporting – and by solving the infrastructural, adoption, and operational challenges in the digital pathology workflow – the burden on pathologists can be reduced.

Figure 5. Histopathology and digital pathology as building blocks of computational pathology.

However, we must not forget that the challenge in computational pathology is the full adoption of AI tools by clinical teams. To address this, stakeholders in industry must collaborate with clinicians, academics, and policy partners to address implementation gaps in digital pathology. Such interdisciplinary projects will be crucial to establishing best practices, implementation frameworks, and change management for the system-level adoption of digital and computational pathology. There is a need to standardize processes and infrastructure for data sharing, image annotation, and image analysis techniques to ensure interoperability. Pathologists and clinicians must support policymakers and regulatory bodies in critically evaluating and defining governance guidelines and regulatory protocols for data protection. All stakeholders must work together to ensure that digital and computational pathology can have a sustainable positive impact on pathologist workflows and patient-centered care.

So how can you – as a pathologist, a clinician, or a patient – tackle the problems in digital pathology and unlock the promise of computational pathology?

  1. Get involved with local initiatives at academic institutions, small-to-medium enterprises, or bigger corporations that focus on providing solutions for digital and computational pathology.
  2. Reach out to local health-focused think tanks and policy advocates.
  3. Be open to change and growth in new computational technologies that support clinical decision-making.
Receive content, products, events as well as relevant industry updates from The Pathologist and its sponsors.

When you click “Subscribe” we will email you a link, which you must click to verify the email address above and activate your subscription. If you do not receive this email, please contact us at [email protected].
If you wish to unsubscribe, you can update your preferences at any point.

  1. NHS England, “National pathology programme. Digital First: Clinical Transformation through Pathology Innovation” (2014). Available at: https://bit.ly/2XaJvsz.
  2. National Cancer Institute, “Pathology Reports” (2010). Available at: https://bit.ly/2YqcusU.
  3. NHS England, “Waiting Times for Suspected and Diagnosed Cancer Patients” (2017). Available at: https://bit.ly/3a9Yc1T.
  4. H Sung et al., “Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries,” CA Cancer J Clin, 71, 209 (2021). PMID: 33538338.
  5. American Cancer Society, “Lifetime Risk of Developing or Dying From Cancer” (2020). Available at: https://bit.ly/3DeqswL.
  6. Royal College of Pathologists, “College Report Finds UK Wide Histopathology Staff Shortages” (2018). Available at: https://bit.ly/3Fn8Hxa.
  7. A O’Dowd, “NHS waiting list hits 14 year record high of 4.7 million people,” BMJ, 373, n995 (2021). PMID: 33858845.
  8. RS Weinstein et al., “Invention and early history of telepathology (1985–2000),” J Pathol Inform, 10, 1 (2019). PMID: 30783545.
  9. A Hosny, “AI Startups in Pathology: A Meta-Review” (2020). Available at: https://bit.ly/2X7eiWF.
  10. A Baidoshvili et al., “Evaluating the benefits of digital pathology implementation: time savings in laboratory logistics,” Histopathology, 73, 784 (2018). PMID: 29924891.
  11. BJ Williams et al., “Digital pathology access and usage in the UK: results from a national survey on behalf of the National Cancer Research Institute’s CM-Path initiative,” J Clin Pathol, 71, 463 (2018). PMID: 29317516.
  12. FDA, “FDA Authorizes Software that Can Help Identify Prostate Cancer” (2021). Available at: https://bit.ly/2YuRwcz.
  13. M Gehrung et al., “Triage-driven diagnosis of Barrett’s esophagus for early detection of esophageal adenocarcinoma using deep learning,” Nat Med, 27, 833 (2021). PMID: 33859411.
About the Authors
Aishwarya Khanduja

Research Fellow at Cyted. She conducted her research with the support of Charlene Tang, Luiza Moore, Alec Hirst, and Marcel Gehrung at Cyted.


Charlene Tang

Business Development Manager, Market Access at Cyted, Cambridge, UK.

Register to The Pathologist

Register to access our FREE online portfolio, request the magazine in print and manage your preferences.

You will benefit from:
  • Unlimited access to ALL articles
  • News, interviews & opinions from leading industry experts
  • Receive print (and PDF) copies of The Pathologist magazine

Register