Digital pathology

Key points

Whole slide imaging (WSI) allows the creation of digital images of entire histology and cytology slides with sufficient detail to allow viewing at varying magnifications comparable to that achievable with a conventional light microscope.

Acquiring digital WSIs requires dedicated hardware systems which create composite images of individually acquired tiles or lines. Such slide scanners typically include slide loading mechanisms, a motorized stage, a light path, image capture device and software to create the composite image. Although the data files which represent these images are large, software techniques enable these images to be viewed on screen in a manner analogous to how a slide may be viewed on a physical microscope.

Digital representation of slides enables histology images to be distributed, viewed and shared over computer networks rather than relying on physical handling of the glass slide. This is likely to represent the future of histology and cytology departments.

Introduction

Whole-slide imaging refers to the creation of a digital representation of the image presented by a glass histology slide, at a level of detail comparable to that seen with a light microscope. ‘Digital pathology’ is a broader term encompassing related processes which maximize the practical utility of such images, including the storage, viewing, annotation and use in applications including educational, research and clinical practice.

Benefits of digital images over physical glass slides

Digital images have several advantages over glass slides. Unlike a physical object, a digital image file can be moved from one physical location to another almost instantaneously. Indeed, an image file can be viewed at the same time by two, or more pathologists, in different locations and potentially separated by thousands of miles. With appropriate data-security arrangements, digital images should be more durable than glass slides which are physical, fragile and prone to fading. The overhead of sorting, filing, storing and retrieving glass slides is particularly burdensome on larger laboratories and a fully digital workflow has the potential to significantly reduce this. Finally, digital images are a prerequisite for automated image analysis.

Digital images

Images can be represented in numerical form in a variety of manners. The text on this page, for example, is ultimately represented by the printing software, not as letters and words but as collections of lines and curves. The commonest method for representing complex real world scenes, including histology images, is to consider the image as a grid of individual points, each with brightness and, for color images, hue. These individual points are referred to as pixels, the term pixel being a contraction of ‘picture element’ and the smallest resolvable detail. In all commonly encountered digital image formats pixels are square, although other shapes, in particular hexagons, are in principle possible. This approach represents images mathematically as a matrix of brightness and hue values. The perceived quality of a digital image of this type relates to the total number of pixels (resolution) and a parameter called pixel depth.

Pixel depth

Image quality is not defined solely by resolution. The term ‘pixel depth’ refers to the extent to which subtly different colors can be distinguished. The crude but recognizable image in Fig. 22.1 has a pixel depth of 1, i.e. individual pixels represent either black or white. Only a single binary bit, 0 or 1, is required to capture this for each pixel. Most formats for the realistic representation of real life images use 24 bits, 8 to represent the intensity of each of the 3 colors red, green and blue. This enables 256 different intensity levels to be represented for each of these colors, a total of 16,777,216 different tones.

Histology as digital images

Whole slide images are distinguished by their sheer size. A typical 15 x 15 mm tissue section imaged at 0.25 μm/pixel at 24-bit pixel depth results in an uncompressed file size of just over 10 GB. The required file size may be compressed up to 30-fold using compression techniques without impact on diagnostic utility ( ), resulting in file sizes of the order of tens to hundreds of megabytes. It is worth noting that discernable artifacts may be introduced into the images at lower levels of compression without necessarily impacting on the diagnostic utility of the image ( ). Moreover, greater degrees of compression may be possible for non-H&E tinctorial stains than for H&E without undermining the practical utility of the image ( ).

The way in which pathologists interact with images also influences the way image data are stored. Many vendors of WSI systems use proprietary file formats, typically based around standard image compression techniques e.g. JPEG, JPEG 2000, but with additional features. Files may include the same image at multiple resolutions to support rapid zooming in and/or out as images are streamed over networks, i.e. pyramidal storage ( Fig. 22.2 ). Files may also store metadata such as a timestamp, image file type, image size, pixel depth, make and model of the scanner and objective, and resolution data such as microns/pixel. Accurate metadata is particularly important if the WSI is going to be quantitatively analyzed, either manually or by an image analysis algorithm as all measurements are calculated from resolution data.

Image acquisition

A slide scanner essentially comprises an optical microscope with a mechanized stage control and focusing, coupled to a digital image capture device, usually a charge-coupled device (CCD) similar to that found in a digital camera. Additional hardware may typically include mechanical apparatus to sequentially load slides, operator controls, a visual display area and computer control hardware. The smallest single slide systems may have a footprint of only 500 x 500 mm, whilst high capacity scanners holding 200-300 slides may be 750 x 1200 mm or more. A typical 15 x 15 mm tissue area may take 30 to 60 seconds to scan using a 40x objective at 25 µm/pixel.

Systems typically use a standard 20x or 40x objective with a light path to a CCD camera. Prior to image acquisition, the slide scanner may register the sample number by reading a barcode printed on the glass slide. A scanner may also perform a low resolution overview scan to determine where the tissue is on a slide and only scan that area, minimizing overall scan time and therefore file size.

The slide is moved by the motorized stage and images are captured by the camera. Two commonly used methods of image acquisition are line scanning and tile scanning. In line scanning, the slide is moved in a linear fashion so that the camera captures strips of the image. In tile scanning, small squares of the image are captured by the camera. A strobe light source and high frame rate camera are typically employed to reduce movement, the blur artifact. With both methods, an ‘image stitching’ algorithm is then applied to assemble the strips or squares of image into a whole slide image (WSI). Reconstruction of a tile scanned image is computationally more complex but modern multi-core processors negate the effect this has on overall scan time.

The topography of tissue on a glass slide can vary by up to 20% of the tissue thickness over a distance of 1 mm. An image scanned using one focal plane would appear blurry in places, making it diagnostically useless. To counter this, slide scanners use an image based autofocus technique. This requires the generation of a focus map either by the operator or automatically from the overview scan image. The scanner could apply autofocus on a tile-by-tile basis in the case of tile scanning or at several points along a strip of image in line scanning. This near continuous autofocus would result in prohibitively slow image acquisition. Instead, in a trade-off between speed and image fidelity, the scanner generates a representative autofocus map, or focuses on every third or fifth tile of an image.

Measures to ensure good quality digital images

The quality of the virtual image depends upon the production of a high quality physical slide and quality control processes relating to the scanner itself. When using a conventional light microscope, a pathologist can work around artifacts including tissue folds, wax on the coverslip, air bubbles and tissue not covered by the coverslip. A slide scanner will faithfully reproduce all of these artifacts, potentially diminishing the quality of the scanned slide.

Standard laboratory quality control procedures should ensure the production of the highest quality slides and the presence of artifacts need to be audited. Additionally, the scanner operator should re-check and if necessary clean or re-coverslip slides prior to scanning. Attention should also be paid to minimizing vibration during the scanning process. If a laboratory is situated close to a major road or rail line or other source of vibration it may be prudent to consider installing a scanner on a vibration-proof table ( Fig. 22.3 ).

The slide scanner should be regularly serviced and cleaned to ensure consistent lighting and focusing. A daily test slide should be scanned to assess the basic function of the scanner and detect major errors such as poor sample detection and abnormal color profiles. This procedure will also generate diagnostic information such as scanner temperature, time-to-focus and time-to-scan, all of which can create variance in digital image consistency and laboratory throughput. Color calibration of the scanner has been shown to increase diagnostic confidence and produce digital slides which are subjectively similar to slides viewed under a light microscope. This is achieved by scanning a standardized color patch affixed to a slide. The color values of this patch are known and can be compared with the on-board reference of the scanner. An adjustment to the color reproduction is then made by the scanner. This procedure is important to ensure day-to-day consistency of color reproduction by the scanner and consistency between scanners in the same department ( Fig. 22.4 ).

Accessing and viewing whole slide images