Originally Published IVD Technology May 2005
Detection Technologies
Automated microscope slide analysis in pathology
High throughput, indefatigable observational capability, and strict objectivity in interpretation make image analysis systems useful laboratory tools.
Iqbal Habib
 |
| Figure 1. The Ariol automated imaging system by Applied Imaging Corp. (San Jose). |
The era of personalized medicine has brought with it the hope for new, better, and more-targeted cancer treatments. It has also instigated an explosion of immunological and DNA-based staining techniques that have created new challenges and opportunities in the area of pathology.
Advances in molecular testing used in cancer treatment are likely to have an impact on survival rates as great as that accompanying the advent of chemotherapy and radiotherapy. However, as high-tech as they may be, the power of these tests is limited by the ability of highly experienced pathologists to achieve deductive insights in their examination of microscope slides.
Pathologists mainly use two staining methods in molecular pathology, immunohistochemistry (IHC) and fluorescent in situ hybridization (FISH). IHC is employed to measure protein expression by way of antibodies, made visible by chemical stains. This technique has traditionally focused on whether or not a cell or group of cells is expressing a specific protein, typically indicated by the presence or absence of stain. New tests have called for the quantification of protein levels by means of both staining intensity and the pattern and location of staining in each cell.
FISH, on the other hand, is a tool to identify the presence, location, and number of specific sequences of DNA in a cell. It is fast becoming the preferred method for identification of gene amplification. In FISH, fluorescent-dyed DNA probes are hybridized to DNA sequences and located by the fluorescent light produced upon excitation.
Improvements in IHC and FISH techniques for pathology had not until recently been matched by advances in imaging technology. But automation has made that connection, addressing the challenges of establishing and maintaining analytical methods characterized by high reproducibility and accuracy.
This article discusses the challenges pathologists face in analyzing microscope slides and considers how new developments in automated image analysis can help to overcome them, as well as accommodate further advancements in staining techniques.
Automation Relieves the Human Eye
Pathologists must cope with three main areas of challenge in their field: productivity, accuracy, and objectivity.
Laboratories are under increasing pressure to improve productivity, i.e., to complete more tests and process more slides. Tackling the mountain of work manually places a constant strain on resources, staff, and finances. And in the near future, this burden is likely to worsen rapidly.
In any field of science dependent on observation, accuracy is essential. However, it is well reported that, after prolonged visual study, eye and specific cone fatigue can significantly affect a person’s ability to discern color changes and identify unusual objects.
Finally, there is the question of objectivity. The nature of the human eye is such that every person sees an object slightly differently from the way others see that same object. Subjectivity in this regard is therefore innate. Also, tints and shades can appear to change from one setting or context to another. Different observers may report seeing different features on the same object, as may a single observer at different times.
Automation can improve the practice of pathology by overcoming the limitations of manual microscopy. Some benefits are immediately evident. Digital analysis is not affected by fatigue. Computers can be programmed to recognize visual patterns accurately, and they are better than the human eye at detecting subtle variations in shading and intensity. Also, computers are consistent, impartial, and objective in their analytical observations.
A case in point is specific cone fatigue, which is a sort of temporary color blindness. When it afflicts a tired pathologist, it causes that person to be momentarily unable to register a range of colors. An automated instrument is not susceptible to such debility.
Indeed, image analysis computing can manage and process huge quantities of data at a phenomenal speed without suffering from any human limitation.
Applied Imaging Corp. (San Jose) began working in this area in the mid-1990s. Its aim was to automate the spadework of anatomical pathology by combining the capabilities of the pathologist’s eye with the power of digital image analysis. The efforts of a team of computer scientists, pathologists, and imaging engineers have resulted in the Ariol automated image analysis system (see Figure 1).
The Ariol provides a new investigative toolbox for the pathologist. Applied Imaging designed the system to supplement pathologists’ experience and training by presenting an objective, consistent, high-resolution view of specimens amenable to the generation of quantifiable results. The technology has application in a variety of pathology settings, including pharmaceutical and biotechnology laboratories, cancer research institutes, university medical centers, and clinical pathology laboratories. It promises to benefit the healthcare professionals whose work relies on pathology data.
Slide Handling and Imaging
 |
| Figure 2. The Ariol slide loader by Applied Imaging Corp. |
Throughput limitations have been a major bottleneck for microscope slide analysis. Managing today’s growing mountain of slides and tests requires a system that can process a large volume of slides at one time and provides easy slide handling and manipulation. The Ariol system’s slide loader can manage up to 50 slides in robust trays and facilitates the transport of slides to and from the microscope stage (see Figure 2). Slides can be scanned automatically unattended—even overnight—and trays can be continually loaded and unloaded for around-the-clock operation with virtually no system downtime.
The image analysis system is also designed to read the slides’ unique bar codes and relate the coded information to a specific set of procedures required to analyze the slide. This mechanism enables the system to analyze slides of more than one type in a single run. As the foundation of the system’s database, the bar codes are keys to a treasure trove of sample history and data.
Rather than acting as a dumbwaiter that simply pushes slides through the system, the slide loader is a delicate and responsive tool that can recall slides to the stage and even relocate the stage itself. Slides are automatically placed within 5 µm of the same position.
Pathological investigations of tissue or cells are based on the meticulous microscopic examination of samples on glass slides. Using manual methods, this process is inherently time consuming and requires diligent concentration from a highly trained operator.
The Ariol system is designed to remove the burden of manual slide scanning by carrying out frame-by-frame image capture of the tissue or cell suspension. This is done by means of a high-resolution charge-coupled device (CCD) camera optimized for low-light-sensitive imaging across the full spectral band, which works with a high-performance Olympus upright BX-61 microscope (Olympus America Inc.; Melville, NY). The resulting images are of very high quality, and their on-screen presentation as high-definition 1600 ¥ 1200-pixel displays becomes the basis for the system’s automated scoring.
Graphic cards used with the system ensure high image quality that can liberate staff from hours of staring down a microscope. However, the importance of direct microscopic examination in this field of science is not forgotten. With just a click of the mouse, the user of the system can recall the slide onto the stage for closer, ocular viewing.
The purpose of an automated image analysis system should not be to remove or diminish human control, but rather to provide the efficiency and high throughput of automation with the option of full manual control as necessary. The Ariol is fitted with its own fully programmable joystick, keyboard, and mouse and a series of interfaces through which the operator can work interactively with the images.
Dual-display operation allows live side-by-side comparison or lets the user run multiple interfaces simultaneously.
Image Analysis and Results Manipulation
 |
| Figure 3. The Ariol imaging system enables multiple units to be run simultaneously through a centralized server. The network supports the system’s laptop data entry stations and review stations as well (click to enlarge). |
In pathology, imaging algorithms are necessarily the core of any system’s analytical capabilities. The Ariol system’s imaging algorithms were designed specifically for the purposes of pathology image analysis. Each module’s imaging algorithm was developed for a specific type of IHC protocol or FISH technique.
To make the system flexible in its imaging capabilities, the analysis algorithms are supported by a so-called trainable classifier, a feature that enables the user to habituate the system to his or her individual laboratory techniques quickly so that it can accommodate day-to-day run variation and methodology.
System flexibility is further evidenced by the variety of applications in routine use in laboratories worldwide. The Microsight cellular rare-event detection, Hersight membrane IHC, and ERsight and PRsight nuclear IHC modules are for IVD use. Other modules, for research use only and not for diagnostic procedures, are specified for cytoplasmic IHC, DNA ploidy, immunofluorescence, microvessel density, sentinel lymph node cellular FISH, tissue FISH, and tissue microarray (TMA) applications.
Image analysis is a resource-intensive process. Consequently, this system features dual Xeon processors, server-grade hard drives and memory, and the Windows XP Professional operating system, ensuring fast analysis with no backlog.
Besides analyzing slides, an automated image analysis system should be equipped with features that enhance postanalysis efficiency in the pathology laboratory. Accordingly, Applied Imaging gave Ariol report generation, data archiving, and networking functionality.
The system generates full-color electronic reports that can include images of significant cells or tissues analyzed, numerical data and scores, diagnostic comments, and space for a signature. Eliminating the time commitment required for paper reports, this feature ensures clarity and consistency in the communication of data, and results in reports that are e-mail and fax ready.
To support tidy, cost-effective archiving, the system offers a full clinical data management server, terabytes of hard-disk space, 10-Gb digital video–random-access memory discs, and, of course, Zip and even floppy disks.
 |
| Figure 4. Analysis of HER-2/neu IHC images using membrane masking enables visualization of membrane-specific staining patterns (click to enlarge). |
The potential of this digital technology for high-throughput operation is fully realized through system networking. Pathologists need not huddle around the microscope when they can review cases, including crisp, high-quality images, from their office desktop PC. Besides operating as a single stand-alone system, the Ariol can be configured with multiple review stations and laptop data entry stations situated in various places around the laboratory or department (see Figure 3). Review stations enable the pathologist or laboratory director to review ongoing work, review and sign off on cases, and optimize the system of reporting and archiving data. The data entry stations can be positioned to register each microscope slide as it comes into the laboratory.
Both images and data generated by the Ariol system can be integrated seamlessly into existing third-party laboratory information systems. This is accomplished through the system’s advanced Web services, which employ platform-independent, non-vendor-specific XML technology.
Automated-System Applications
The Ariol system has application in the quantification of a number of biomarkers that aid pathologists in cancer cell analysis. Most compelling is its potential use in the detection and analysis of breast cancer cells in women. To demonstrate this utility, the system was put to the test in four laboratory applications used in breast cancer profiling: HER-2/neu IHC, HER-2/neu FISH in tissue, occult tumor cell scanning, and routine tissue IHC.
Automated Scoring for HER-2/neu IHC. The American Cancer Society estimates that 211,240 women in the United States will be diagnosed with invasive breast cancer in 2005. About 40,410 women will die from the disease this year.1
Fortunately, genetic approaches toward improving the treatment of breast cancer are already proving beneficial, and indicating the brightness of the future of personalized medicine. One of the best-established examples involves the overexpression of the HER-2/neu gene that occurs in about 30% of metastatic breast cancer patients. The HER-2/neu protein, a member of the epidermal growth factor receptor family, resides in the membrane of expressing cells. The challenge has been to identify the 30% of patients at risk by finding a way to measure overexpression of HER-2/neu dependably.
The market-leading IHC test for HER-2/neu is the HercepTest from DakoCytomation a/s (Glostrup, Denmark). Analysis of this test calls for measurements of staining intensity, pattern continuity, and stained-cell frequency to be made on the region of the invasive tumor. The method has proved to be a considerable challenge for the pathology community, having been demonstrated to show poor reproducibility in scoring by manual methods.2 In a recent comparison of local and centralized testing of HER-2/neu status, the concordance between results was measured at only 74%.3 Digital image analysis has been shown to improve accuracy and reproducibility in scoring this test.4
 |
| Figure 5. Z-stack imaging in HER-2/neu FISH analysis, showing benefits in image quality and spot resolution (click to enlarge). |
In the experiment, the Ariol was used to provide high-throughput analysis for HER-2/neu IHC-stained slides. The slides were transferred automatically by the system’s slide loader, then scanned at low magnification to produce a satellite view of the entire tissue section. Automatic or user-defined target regions were then scanned frame by frame at higher magnification, with three bright-field-filter images being collected per frame. The use of bright-field filters and a monochrome CCD camera—as opposed to a color camera—is unique to the Ariol system and is recognized as producing very accurate high-resolution images without the need for the artificial distortion of colors or shapes.
The captured images were brought together seamlessly by the system to produce a high-resolution image of the whole slide into and out of which the analyst could zoom while maintaining excellent image quality.
Analysis of defined tissue regions was then conducted using the system’s cell-masking template to quantify the number of cells and to analyze stain intensity and pattern continuity specifically within the cell membrane (see Figure 4). This membrane specificity makes for truly objective analysis.
Automated Tissue and Cellular FISH Scoring. Current advised HER-2/neu clinical practices dictate reflexing to tissue FISH analysis for patients with borderline IHC scoring.5 The Ariol was designed to provide applications for automated interphase FISH fluorescent-spot counting by which fluorescent loci-specific signals within a cell or a number of cells are rapidly and accurately measured. Each signal, or spot, reveals the location of a specific sequence of DNA. In the pathology environment, the most common use of this technique is to count the number of chromosomes or specific genes in a cell.
Distinguishing between fluorescent spots can be difficult in traditional manual microscopic imaging. Problems arise mainly when two fluorescent spots lie one on top of the other within the thickness of the tissue section and appear as just one spot.
But Z-stacking with an automated imaging system can automatically capture individual images of planes at slightly different focal distances through fine control of the focus motor (see Figure 5). With the Ariol system, various Z-stacks can be viewed individually or compressed into a single image—a 2-D presentation of a 3-D scene—with optimal resolution of the images being maintained. The result is greater accuracy and reproducibility in FISH analysis.
This approach using automated image analysis has been highly successful in the automated FISH scoring of HER-2/neu in tissue samples as described by the PathVysion test (Vysis Inc.; Downers Grove, IL). The Ariol system uses multiple light filters and its high-end camera to capture images of the FISH-stained tissue in a number of focal planes and perform complex cell segmentation. Image-analysis algorithms are employed to quantify the different fluorescent signals for the HER-2/neu gene and chromosome 17, correcting for three-dimensional alignment of fluors and even calculating the likely number of individual spots within a cluster of FISH spots—a difficult task to perform manually, and one prone to variability among observers.
The results are then presented on a screen in a grid with the visual evidence and numerical data side by side. The pathologist can review and recall individual cells for further examination, if required, prior to generating automatically a full-color report.
Seeking Occult Tumor Cells in Bone Marrow. A quickly emerging field in the study of cancer is the search for epithelial tumor cells in bone marrow tissue. Shed from their primary tumor site, the cells can circulate systemically to become lodged in the patient’s bone marrow. Identification of these tumor cells by manual means is taxing, as only one such cell may be present among a million others.
The detection of such occult tumor cells may be a reliable and important prognostic determinant of cancer and is a focus of energetic research for future drug development. The clinical accuracy of this method and the speed at which it can be completed are therefore vitally important to this technique reaching its full clinical potential.
The Ariol automated hardware platform and imaging algorithms were designed to enable automated scanning for rare labeled cells, and for users to review and classify recognized objects either on-screen or directly down the microscope. The system rapidly scans frame after frame, analyzing for candidate IHC-stained objects discriminated by color and no fewer than 22 morphometric parameters. Objects are presented for review by the user, who can quickly classify the findings and, if necessary, analyze the data using powerful analytical and statistical tools.
The work-flow improvement contributed by the slide loader and the system’s intuitive interface enhances manual analysis considerably. With the additional benefits of high accuracy and objectivity, imaging automation makes the search for occult tumor cells in bone marrow a good example of how the technology may advance cancer research.
The value of the technology was demonstrated quantifiably by a Norwegian specialist in occult tumor cell detection. In her study, the performance of Applied Imaging’s technology was measured against that of trained pathologists. The automated image analysis system identified 38% more positive samples than manual methods alone in a set of 120 clinical slides, and it identified 13% more tumor cells in a set of 30 spiked samples.6 The Ariol system is now used routinely in that researcher’s laboratory.
Automated Analysis for Tissue IHC. Perhaps the potential of automated image analysis is most impressive with respect to routine IHC markers. Used in immunohistochemistry since the beginning, nuclear and cytoplasmic IHC markers have been the bread and butter of many laboratory investigations. Despite this fact, however, this area of pathology still exhibits considerable diversity in terms of scoring methods and analysis patterns. Some slides may require only a quick glance down a microscope by an experienced pathologist to yield a report of positive or negative, while other slides may call for the counting of more than a thousand cells.
To accommodate these practices, the Ariol was designed for work-flow flexibility. For rapid examination, the system provides clear, easy-to-view tissue images and full cataloging and quick porting. For more expansive quantification, it offers a choice of nuclear, cytoplasmic, and membrane-masking methods for high-specificity object analysis or convenient area analysis. Both system configurations engage in pixel-by-pixel analysis for highly objective results and are integrated with a review interface for ease of use.
Automated Analysis for Tissue Microarrays. Although tissue microarrays are used for research now, there are already early adopters integrating this methodology for clinical purposes. Because of the difficulty of managing TMA data sets, automated slide scanning and image analysis will become important in this practice.
Conclusion
Automated image analysis does not replace pathologists; rather, it aids them in their work. Designed never to fade in concentration or accuracy, automated image analysis systems perform slide analysis at levels previously unachievable by manual means.
The positive implications of automation for this field of laboratory endeavor are several. First, staff are released from long hours at the microscope. The time savings resulting from improved work flow can benefit many areas of pathology where maximizing staff productivity is an issue. Also, technologists, pathologists, and clinicians can be networked to more easily compare data. The role of the pathologist is augmented by a tool that allows more-comprehensive analysis and generates greater confidence in results. Finally, advantages can be realized in the areas of operational costs and reimbursements.
For all these reasons and more, automated image analysis will have a significant effect in the field of pathology.
References
 |
| Iqbal Habib is technical product manager for the pathology market at Applied Imaging Corp. (San Jose). |
1. “Cancer Reference Information,” the American Cancer Society Web site (Atlanta: 2005); available from Internet: www.cancer.org/docroot/CRI/content/CRI_2_4_1X_What_are_the_key_statistics_for_breast_cancer_5.asp.
2. JMS Bartlett et al., “Evaluating HER2 Amplification and Overexpression in Breast Cancer,” Journal of Pathology 195 (2001): 422–428.
3. PC Roche et al., “Concordance between Local and Central Laboratory HER2 Testing in the Breast Intergroup Trial N9831,” Journal of the National Cancer Institute 94, no. 11 (2002): 855–857.
4. Y Hatanaka et al., “Quantitative Immunohistochemistry Evaluation of HER 2/neu Expression with HercepTest in Breast Carcinoma by Image Analysis,” Pathology International 51 (2001): 33–36.
5. RL Ridolfi, MR Jamehdor, and JM Arber, “HER-2/neu Testing in Breast Carcinoma: A Combined Immunohistochemical and Fluorescence In Situ Hybridization Approach,” Modern Pathology 13 (2000): 866–873.
6. E Borgen et al., “Use of Automated Microscopy for the Detection of Disseminated Tumor Cells in Bone Marrow Samples,” Cytometry 46 (2001): 215–221.
Copyright ©2005 IVD Technology
