Biobanks can be defined as a structured repository of biological specimens obtained from patients having a particular disease. These collections then serve as materials for future scientific and medical research, specifically, in translational medicine for the discovery of biomarkers and molecular factors involved in disease diagnosis, progression, and treatment. Biospecimens can be any many forms, such as whole blood, serum, saliva, skin cells, tissues for cryosectioning, or formalin-fixed paraffin-embedded tissues. These specimens can also originate from a wide variety of sources, such as humans, microorganisms, plants, or animals. The critical major steps in establishing specimens for a biobank consist of a synchronized collection of diseased or healthy samples, their processing, and their appropriate storage. Storage can be a major challenge requiring extensive space and refrigeration facilities. Cataloguing, categorizing, searching, and retrieving these samples can also be huge and time-consuming task. Moreover, as these specimens are used for research purposes, they have to be of excellent quality, and hence an emphasis on quality assurance (QA) protocols is paramount.
Tissue specimens excised from patients during surgery are divided, with a part used for the patient diagnostics and the remainder transferred to a biobank. On arrival at the biobank, the samples are usually sectioned and stained by standard histopathological techniques and then analyzed by a pathologist for an evaluation of the specimen composition and disease state. This process not only requires a time commitment from already over-burdened pathologists but can also expose the process to variability through differences in both inter- and intra- pathologist interpretation.
Digital Pathology to the rescue?
The digitization of tissue samples has extraordinary value in biobanking. The deployment of automated slide scanning systems in conjunction with computer-assisted morphometric image analysis capabilities, has the power to revolutionize tissue analysis processes and can even potentially serve as an alternative to the physical storage of biomaterials. The entire tissue sample can be optically scanned, providing a digital image data file that can be stored in a database, potentially obviating the need for refrigeration and cryogenic storage. The simple interface, high resolution display and instantaneous zoom capabilities found in most modern digital pathology systems, enable the pathologist to navigate and make observations in a similar way to standard microscopy. Moreover, pathologist and microscope slide no longer need to be in the same room; the image can be instantly retrieved for analysis at a time of his or her choosing and from any location. Image analysis software can then provide a non-variable quantitative assessment of the sample and automatically highlight regions of potential interest. By removing the need to mail physical slides, digital pathology also facilitates the exchange of histopathological images between biobanks and research organizations. Images can be inspected instantaneously by a partner laboratory for immediate sharing and consultation.
How are microscopic images analysed by digital pathology?
Computer based image analysis of whole slide images relies on dedicated algorithms which have been specifically developed to make inferences about the tissue sample based on color, texture, morphology etc. Typically, image analysis software can identify regions of interest in the digital image as well as providing a quantitative assessment. The algorithms are normally developed via an iterative training process through exposure to a representative set of samples and are then able to identify the unique spatial-spectral topographies that distinguish specific image pixels characteristic of a tissue type or a disease state. During development, algorithms are tested on study groups of unknown tissue samples to increase exposure to variability and therefore the quality of ‘the training’ for the targetted tissue type. There are hundreds of pattern recognition algorithms now in existence from a host of specialist providers. Algorithms may be designed for example, to recognise a tissue type, to highlight a region of solid tumor tissue or even to classify or grade a specific tumour type. Moreover, many of the image analysis vendors allow the user to further enhance their algorithms for specific applications, meaning that biobanks now have the possibility to create their own algorithms against a specific tissue type or disease or for quality assurance purposes. As an example, Wei and Simpsons (2014) designed algorithms (through a training set of annotated biobank samples) with mean accuracies of > 98% and with > 95 % sensitivity and specificity. These were then applied to cancer specimens present in the biobank to evaluate neoplastic cells against stromal elements. Amongst the companies offering these type of mutable image analysis algorithms are Indica Labs, Visiopharm, Leica Biosystems.
A critical partnership between digital pathology and Biobanks
The automated analysis of biobank samples using digital pathology and image analysis has almost certainly strengthened biomarker discovery and research and it has helped world-wide investigators in rapidly selecting biospecimens for a particular field of study. Canada has already approved high resolution digital whole slide imaging technology for clinical use and in 2017, the US Food and drug administration (FDA) also approved the first specific digital pathology algorithms (Philips Intellisite) for primary diagnosis. More approvals will certainly follow and the biobanks will have a critical role in the provision of the high quality tissue samples required for algorithm training. In 2019, ContextVision, a software company specializing in image analysis and artificial intelligence, based in Stockholm, signed an agreement with a large biobank to digitise samples and to develop databases for defined patient cohorts. Digital pathology and subsequent image analysis have not only allowed biobanks, specialist researchers, oncologists, and surgeons to interact through image sharing. Image duplication also provides huge advantages and for this reason, virtual repositories can be expected to spread quickly with the potential to redefine our present concept of biobanking. In the future, the incorporation of artificial intelligence and deep learning technologies into standard image analysis techniques will undoubtedly lead to a further improvement in software training and an even greater accuracy in computer-based decision-making models. The steady provision of high-quality samples from the world’s tissue banks and the continued digitization of biobanking are both critical to this continued evolution.
Digital pathology and image analysis are already hugely significant technologies for biobanking. As competition grows between commercial providers, the instrumentation and software components required for analysis will likely become increasingly more affordable, leading to an increased rate of sample digitisation. According to a report by the market research group ‘Research and Markets’, entitled the “Strategic Commercial Partnerships Transforming the Global Biobanking Market, 2020 – 2026”, the biobanking market is expected to grow by ~$71.2 billion during this period and more affordable digital pathology services are likely to further escalate this market growth.