In the era of the Internet of Medical Things, telehealth, digital therapeutics, and other cool technologies, up to 90 percent of the pathologist’s daily routine doesn’t differ much from how things were done in the 19th century. Experts spend enormous amounts of time examining tissues via a microscope, searching for old slides in hospital archives, writing reports, and performing other manual operations.
Pathology workloads only tend to increase due to the aging population, growing cancer rates, and the development of precision medicine. All these factors require more and more tests done each year — and create urgency in speeding things up. Rephrasing the old saying, “necessity is the mother of innovation.” Digital and computational pathology is here to augment — if not replace — analog lab processes.
What is digital pathology?Digital pathology (DP) encompasses the aсquisition of information from microscope glass slides, along with further management, interpretation, and analysis of this data using software tools. It has already found application in clinical diagnostics, research, medical education, and training. Also, digital pathology sees increasing adoption by large pharmaceutical companies striving to enhance the drug development process.
For labs and hospitals, the digital approach comes with such widely recognized benefits as cost efficiency, reduced time to treatment, safety, and easy access to a second opinion. And though the majority of labs still stick to analog procedures, more and more of them are considering a transition to the new workflow that involves the following steps.
Digital pathology workflow and integrated systems.
Prepare a tissue sampleThis operation remains almost the same as in the traditional workflow. A pathologist — an expert in cell and tissue diagnostics — looks at a biopsy to assess its color, size, and consistency. At this stage, the expert can spot signs suggesting cancer and also decide which regions of a large specimen are to be examined under a microscope.
Then, the selected area undergoes multi-step preparation that takes up to several days. It involves treating with chemicals to preserve the tissue structure, placing the specimen on a glass slide, staining to enhance contrasts, and applying coverslips to prevent damage. If you’re going to digitize the sample, here are some tips to capture an accurate scan.
Be careful with staining. Excessive or pale background staining makes it harder for computer software to automatically identify the tissue and isolate its components, which will lead to more manual work.
Avoid folds, wrinkles, and air bubbles. Scanners work best for flat surfaces. Any bumps on the tissue or air inclusions captured under the coverslip may affect the scanner focus and even make the slide unsuitable for digitizing.
Check slides before scanning. Make sure that they are completely dry and have no cracks. Water or the smallest particles of glass can damage the scanner.
Capture an image with a whole slide imaging scannerThe technology that makes digital pathology possible is known as whole slide imaging (WSI) or virtual microscopy. It’s core element is a WSI scanner that captures an image of a glass slide and creates its exact electronic replica called a virtual slide.
Unlike glass slides, virtual ones are easy to duplicate, store, catalog, and share. Besides that, they can be attached to electronic health records creating the full picture of a patient’s health.
Save a virtual slideThe scanner automatically preprocesses the virtual slide and saves it to on-premises or cloud storage. It can be a centralized vendor neutral archive (VNA) that adopts all types of medical imaging. Before saving, a compression algorithm is usually applied to reduce the file size.
View and edit the slideIn the digital workflow, a pathologist examines magnified tissue samples on a computer monitor rather than under a traditional microscope. Slide viewing and management software allows you to zoom out a tissue section and see its finest details. You can also change the angle of view, add annotations and comments, or open several images side by side for comparison.
Often, WSI scanners come with all necessary software and proprietory monitors from the same provider creating an end-to-end, whole slide imaging system. However, in most cases, it’s possible to access slides remotely, from your regular computer, if relevant programs are installed.
Whole slide imaging system hardware and software components.
Upload additional informationOne of the selling points of digital pathology is the ability to examine tissue alongside other medical data associated with the same patient, such as radiology scans or clinical history. To make it possible, you need first to integrate slide viewing software with your existing EHR and a radiology information system (RIS).
Share the slideIn a typical scenario, if a pathologist wants to consult with colleagues from a different hospital, city, or state, the samples in question have to be packaged and sent to the intended specialist. Such shipping is quite costly, not to mention that getting a second opinion will take days to weeks. Also, there is always a risk of slide damage or loss in transit.
But all those hurdles disappear once tissue specimens become available in electronic form allowing them to be exchanged via the Internet. You can easily share slides with other specialists for getting a second opinion as well as with patients, research centers, and other stakeholders.
Report the resultsSome image viewing solutions have reporting functionality. Yet, typically, this task is solved via integration with the laboratory information or laboratory information management system (LIS/LIMS). Again, the connection to EHR is important for this stage, as it automates sending reports to a physician who will prescribe treatment.
The most advanced digital workflows also incorporate artificial intelligence (AI) and machine learning (ML) methods to recognize patterns in tissue specimens. This practice comes under the umbrella of a relatively new discipline — computational pathology.
Computational pathology: what it is and how it can helpComputational pathology refers to the use of ML and specifically deep learning for analyzing virtual slides and associated metadata, which can include information about how a sample was acquired, patient demographics, and annotations.
The ability of deep learning algorithms and particularly convolutional neural networks (CNNs) to extract features from visual data makes them a natural fit for image recognition and computer vision in healthcare. AI models serve as another pair of eyes, which can extract features invisible to humans or overlooked due to weariness and other factors.
If you search for real-life applications of deep learning in medical diagnosis, the majority of them belong to radiology which generates tons of visual data in the form of X-rays, CT scans, and MRIs. (Read how AltexSoft created an AI-based diagnostics tool for lung disorders.)
Following the steps of radiology, pathology also increasingly employs deep learning to improve the efficiency of tissue-based diagnosis. AI models, which are often more accurate and sensitive than read-outs from human pathologists, play a crucial role in precision medicine, a medical approach advocating personalized treatment. And the highest hope for AI is that it will revolutionize oncology, facilitating early cancer detection and the development of unique therapies for each individual case.
Digital pathology software with AI capabilitiesA recent survey shows that 75 percent of pathologists are interested in adopting AI to improve patient care. Moreover, 80 percent expect that deep learning will inevitably come to labs by 2028. In response to the growing demand, almost all large providers of end-to-end digital pathology platforms augmented their products with AI — in one way or another.
Digital pathology software: compatibility and available AI tools
Roche uPath enterprise software: available for testing and distribution of AI toolsuPath is a digital pathology platform by Roche, a global leader in pharmaceuticals and diagnostics headquartered in Switzerland. The software enables experts to examine multiple slides in a single view, alongside other available patient information, and make annotations.
The tool is natively compatible with other Roche products, such as the VENTANA DP 200 slide scanner and VENTANA Connect middleware that simplifies the integration of the pathology suite with your existing LIS. The AI offering from Roche is VENTANA Companion algorithm software designed specifically for breast cancer diagnosis.
The company also provides its Digital Pathology Environment for secure design, deployment, and exchange of image analysis tools. Developers can connect to the environment via Open API and distribute their products on the uPath platform. Pathologists and researchers get access to advanced Roche and third-party AI algorithms.
PathAI: driving oncology and NASH treatmentA Boston-based startup, PathAI develops AI-powered tools to identify disease biomarkers and predict a patient’s response to various therapies. Their main area of focus includes diagnosis and treatment of cancer and NASH (nonalcoholic steatohepatitis). PathAI partners with biopharma companies, bringing machine and deep learning models to research, clinical trials, and diagnostic development.
PathAI also collaborates with the above-mentioned Roche to distribute their research-use-only algorithms via the cloud version of uPath.
Sectra Digital Pathology Solution: vendor-neutral platform and AI marketplaceSwedish company Sectra specializes in medical imaging technologies. Its Digital Pathology Solution allows for viewing files in different formats, generated by different scanners, cameras, and other systems. The software comes with an image sorting feature along with built-in image analysis and reporting modules. The module is vendor-neutral and connects to third-party products using industry standards and open APIs.
Due to FDA clearance, American pathologists can utilize Sectra software paired with Leica scanners for primary diagnostics. Previously, its area of application in the US was limited to research and tumor boards (meetings of expert physicians to discuss complex cancer cases.)
As for AI capabilities, you can add them to the current workflow via Sectra Amplifier Marketplace. All applications are vetted by Sectra: They have FDA approval or CE marking (signifying permission to be sold in the EU) and seamlessly integrate with its digital pathology platform.
Visiopharm image analysis and diagnostic APPS: driving research and drug developmentImage analysis software from Visiopharm, a digital pathology pioneer from Denmark, supports research and drug development projects in more than 40 countries worldwide. Their suite of solutions includes an image viewer and a set of tools handling different aspects of image processing and analysis. Visiopharm products are compatible with scanners from their partner, Japanese manufacturer Hamamatsu Photonics.
You can also take advantage of the vast library of pretrained AI models to identify different types of pathologies in virtual slides. But most of them can be used for research only. Visiopharm also encourages experts to use their platform to create and train deep learning algorithms of their own.
Proscia Concentriq digital pathology platformThe vendor-neutral platform Concentriq is designed by Philadelphia-based software company Proscia focusing on automation of the pathology processes. It easily integrates with all popular WSI scanners as well as EHR and laboratory information systems, joining separate hardware and software components into a single lab network. Pathologists and researchers can view, manage, annotate, and share virtual slides via a unified interface.
As for computational pathology, you can integrate directly into the pathology workflow the following solutions:
- DermAI by Proscia to classify dermatopathology slides and provide support in the diagnosis of skin diseases;
- Visiopharm AI which is pre-integrated with Concentriq platform; and
- Ibex Medical Analytics (Israeli-based startup) AI for prostate cancer diagnosis.
Digital pathology limitationsThough digital pathology holds a lot of promise, only 20 percent of US labs use it in secondary diagnosis. Even fewer rely on the technology for the primary diagnosis or employ AI algorithms to support diagnostic decision making. Here are the main reasons why and some tips on how to address existing problems — if it’s possible at all.
Large image sizeFor diagnostic purposes, digital replicas must be at the very least the same quality as original glass slides when viewed under a microscope. That’s why a scanning process takes from one to twenty minutes and produces files of several hundreds of MBs to several GBs each. The resulting images may be as large as 100,000*100,000 pixels. They require high-speed and high-bandwidth Internet for sharing, a lot of memory for storing, and significant human resources for detailed annotation.
Large images also make applying AI more challenging as the model size increases proportionally to data input. Thus, the training process becomes slow and can’t be run on one GPU (graphics processing unit).
Of course, most WSI systems feature compression mechanisms for reducing file sizes. Still, pathology departments have to invest heavily in robust networks and on-premise or cloud computing and storage infrastructure to work with virtual slides and scale their use.
Ways to address the problem. The common practice is to divide a virtual slide into smaller tiled images or tiles so that computer memory can accommodate them for viewing or processing. Many of those tiles represent water spots, smudges, and background, and may be discarded as useless. In some deep learning projects, only tiles with 90 percent of the area occupied by tissue are considered to be of value for analysis. Such selectivity leads to faster and more accurate model training.
Informative tiles can be detected and selected automatically, with special algorithms implemented into the image processing flow. However, this takes additional effort and a workforce with expertise in data analytics and data science.
High acquisition and operational costsSmall and even mid-sized pathology laboratories can hardly afford WSI due to the huge price of high-throughput scanners — on average, about $237k per piece worldwide, as of 2020.
But besides initial investment, there are hidden costs relating to staff training, technical support, licensing, and system maintenance. And though a recent comparative cost analysis demonstrated that WSI could save labs $1.3 million in five years, this justifies the adoption of the technology for large centers only.
Ways to address the problem. Facilities with limited budgets can send glass slides for scanning in commercial centers. For example, such services are provided by 3DHISTECH, the first European manufacturer of devices for digital pathology, located in Budapest.
Also, less expensive hardware solutions are appearing on the market. For example, Israeli startup Augmentiqs offers an electro-optic gadget fixed under the microscope eyepiece. The AI-driven device transmits a live view of the specimen directly to the computer screen. A pathologist could point out important features, make annotations, or apply image analysis software.
All the interactions are immediately displayed as augmented reality within the eyepiece. The technology is twenty times cheaper than a digital scanner. Also, it saves time and computer resources, as you don’t need to acquire and process huge whole slide images.
Lack of standardsIt may come as a surprise, but there are no common data standards in digital pathology. Though we have DICOM, an international communication protocol for medical imaging, it currently doesn’t apply to virtual slides. Each WSI vendor uses its proprietory viewing software and slide processing algorithms. Images generated by different providers come with variations in coloring, contrast, and resolution.
File formats also vary from provider to provider — say, Philips chose .tiff while Leica relies on .scn. Other common WSI formats are .tif, .svs, .vms, .vmu, to name just a few.
All these lead to compatibility and interoperability issues. In other words, you often can’t share files across platforms or view virtual slides on third-party devices. As a result, data gets locked inside one system which hampers the widespread use of technology.
Ways to address the problem. It’s expected that in the future WSI providers either adopt DICOM or develop another singular standard to make pathology images vendor-neutral.
For now, if your hospital or lab intends to save virtual slides to an existing picture archiving communication system (PACS) or send them to the cloud, say, via Google’s Healthcare API, you first have to convert WSI files into the DICOM format. Some tools to help with this task are
- Orthanc Dicomizer command-line tool,
- Google Cloud Platform WSI-to-DICOM converter, or
- PyPi dicom-wsi package.
- PMA.start, a WSI viewer supporting 38 formats. It also offers API access to image processing functionality and plugins to integrate with QuPath, a platform designed specifically for bioimage analysis;
- Cytomine, a tool for collaborative analysis of giga-pixel images, employing ML algorithms;
- ASAP (Automated Slide Analysis Platform), a WSI viewer for making annotations and visualizing results of machine learning analysis (such as segmentation masks); and
- caMicrosope, a digital pathology image viewer supporting annotations and markups generated by humans or machines.
Lack of approvalWhole slide imaging was introduced back in 1999. For now, Europe takes advantage of numerous WSI systems but only two of them — Philips IntelliSite Pathology Solution (PIPS) and the Aperio ATD 2X scanning and viewing platform by Leica Biosystems (Germany) — got clearance from the US Food and Drug Administration (FDA). Also, the FDA approved a digital pathology module by Sectra (Sweden) if it’s paired with Leica scanners. All other WSI technologies can’t be marketed in the United States and are available only for research.
The FDA evaluation of whole slide imaging hardware and software involves technical performance assessment along with expensive clinical studies. PIPS received authorization after the revision of nearly 2,000 cases of using tissue samples to identify a disease.
AI algorithms face the same cautious attitude on the part of regulatory bodies. Most ML-driven diagnostics tools can be used in the US for research purposes only. In the EU, AI healthcare technologies must carry the CE marking (CE is the initialism for Conformité Européen) to enter hospitals and labs. Currently, this approval is simpler to obtain than FDA clearance. Yet, it all can change in 2024, when the new AI Act is expected to be implemented in Europe.
Ways to address the problem. Whatever the current situation with rules and regulations, the future definitely belongs to new technologies and particularly AI. But first, people have to adopt a responsible approach to artificial intelligence and do their best to prevent negative outcomes.