Deadline: 28 November
This document specifies requirements and gives recommendations for the digitalization of slidemounted, stained sections, processing of digital whole slide images, and artificial intelligence (AI) based image analysis to support anatomical pathology examination. This document focuses on sections from formalin-fixed paraffin-embedded (FFPE) tissues (derived e.g. from surgical resection, biopsy or autopsy). It is also applicable to paraffin-embedded tissue fixed by fixatives other than formalin.
This document is applicable to in vitro diagnostic examinations using digital pathology and AI-based image analysis performed by medical laboratories, in particular but not limited to anatomical pathology laboratories. It is also intended to be used by health institutions, in vitro diagnostics developers and manufacturers, and regulatory authorities.
NOTE International, national or regional regulations or requirements can also apply to specific topics covered in this document.
Technological advances and the increased focus on precision medicine demanding precision diagnostics created a need for application of digital pathology and artificial intelligence (AI)-based image analysis.
The global digital pathology market was valued at USD 736 million in 2021 and is expected to grow at a Compound Annual Growth Rate (CAGR) of 13.2% for the period of 2021-2026. This is driven by rising incidence of cancer, global shortage of pathologists fostering the adoption of digital pathology, telepathology, and growing applications of digital pathology in drug development, complementary diagnostics and companion diagnostics. The (immuno)histochemistry sector (using stained tissue sections) is estimated to register the highest growth rate in the market (by application).
Digital pathology using high-resolution digitized (whole slide) images of slide-mounted, stained tissue sections from e.g. surgical resection, biopsy or autopsy specimens, gained increasing relevance in the context of medical diagnostics and personalized medicine. This was enabled by the availability of slide scanners generating digital images with high resolution and information content comparable to classical light microscopes.
Digital pathology and AI-based image analysis can be applied in disease detection, diagnosis and prediction of prognosis. Furthermore, it supports the identification of the optimal treatment, and the discovery of novel biomarkers and drug targets.
The generation and processing of the digital (whole slide) images for AI-based image analysis requires the modification of diagnostic workflows and needs special medical devices: A whole slide imaging (WSI) system comprising a slide scanner and a workstation (i.e. hard- and software) for slide viewing and presenting them to pathologists for diagnosis making.
The US Food and Drug Administration (FDA) has published a guidance document and approved the first registrations of WSI equipment as medical devices in 2017.
The digital pathology together with AI-based analysis of digital images poses new requirements on the standardization of the entire workflow to obtain high-quality (whole slide) images and reproducible, qualitative and quantitative analysis results. Factors which can impact the quality of the image and the computational analysis result include, for example:
i) pre-analytical and analytical factors associated with the collection and processing of tissue specimens and production of stained slides (e.g. tissue fixation and processing, embedding, section quality (e.g. uniformity of section thickness), intactness, size, material and thickness of the slide, staining intensity and homogeneity, placing of the coverslip, air bubbles, dust, or markings on the slide,
ii) factors such as errors occurring during scanning (e.g. bar code detection failure, focus failure, incomplete slide scanning (no detection of tissue or missing of tissue parts), improper stitching, line artefacts or variable colour calibration and different settings of brightness, intensity disparity, and boundary intensity during scanning,
iii) factors associated with visualisation of digitized images on monitors (e.g. colour calibration, and
iv) factors associated with the AI-based analysis (e.g. wrong detection of features or “Clever Hans effect”, performance of an AI algorithm not demonstrated for features present on a whole slide image).
Therefore, a standardization of the entire digital pathology and AI-based image analysis workflow and requirements and recommendations for the respective devices including the performance verification and validation are needed to secure the proper quality of the image and the AI-based analysis result to support pathological examination.
The following will not be covered within this document:
- the in situ detection techniques;
- information security and privacy requirements/recommendations;
- general software-related issues (e.g. software lifecycle management, cybersecurity, interoperability with information management systems);
- image file formats;
- diagnosis making by pathologists.
You can comment via this link.