Secondary Logo

Journal Logo

Review Articles

Clinical Application of Image Analysis in Pathology

Cornish, Toby C. MD, PhD

Author Information
Advances In Anatomic Pathology: July 2020 - Volume 27 - Issue 4 - p 227-235
doi: 10.1097/PAP.0000000000000263

Abstract

Interest in digital image analysis (DIA) in the field of pathology has been steadily increasing over the past several decades. This has been driven by several factors including the development of new detection techniques, increased accessibility of computing power, advancements in digital imaging, an evolving role for biomarkers, and the emergence of whole-slide imaging (WSI).1,2

This waxing trend in DIA is reflected in the pathology literature. Figure 1 illustrates 2 eras of significant growth in the number of pathology DIA publications per year indexed in Medline. The first growth period stretched from the late 1970s through the early 1990s. Notably, this period saw a massive increase in the accessibility of computing resources by noncomputer scientists, in particular the emergence of personal computers. The field of digital imaging was also emerging at this time, with charge-coupled device image sensors, first invented in 1969, finding their way into microscope-compatible cameras in the late 1980s.1 Detection methods in pathology also advanced significantly during this time period. In particular, the application of immunohistochemistry (IHC) to formalin-fixed paraffin-embedded tissue, considered nearly indispensable today, did not achieve widespread acceptance as a standard diagnostic technique until the mid-to-late 1980s.3 Labeling of nucleic acids with fluorescent probes was developed in the early 1980s and fluorescence in situ hybridization (FISH) was being used routinely in cytogenetics laboratories by 1990.1,4 Curiously, the number of publications per year remained steady from around 1995 to 2010 (Fig. 1). This stagnation occurred despite the emergence of commercial WSI and may have been the result of few truly new techniques for detection.2 The second period of growth stretched from 2011 to 2016. The start of this era roughly corresponds to the emergence of a new generation of clinically-oriented WSI systems and numerous developments in the digital pathology vendor space.5 It also corresponds to the rise of deep learning-based convolutional neural networks running on graphics processing units, which began dominating image analysis competitions around 2011.6 Publications leveled off again after 2016 (Fig. 1) despite the recent Food and Drug Administration (FDA) clearances for use of WSI for primary diagnosis.7 Although it would be tempting to interpret the data for 2017 to 2019 as a slight reduction in interest, this most likely represents noise in the data.

FIGURE 1
FIGURE 1:
Pathology image analysis articles indexed in Medline. The Medline database was searched via PubMed using the search string (“image analysis” AND pathology) AND (clinical OR predictive OR prognostic OR prognosis OR prediction). This search was designed to capture most clinical applications of image analysis in pathology and exclude nonpathology, nonclinical publications. Publications per year rose steadily from 1976 (no articles were identified before 1976) until the mid-1990s. From around 1996 to 2010, publications per year remained steady, followed by a spike from 2011 to 2016. Since 2016, publications have remained steady or declined slightly. The asterisk denotes that the article count for 2019 is incomplete (results as of January 2020).

While the literature is very much alive with new possibilities for DIA in pathology, very few pathologist are engaged in the activity clinically. In fact, much of the clinical DIA performed today is concentrated in a few subspecialty areas and, even then, is usually performed either at large reference laboratories or academic medical centers. The concentration of clinical DIA is such that many practicing pathologists might be surprised to know that around 25% of all quantitative IHC is performed using DIA. Given how common DIA actually is, why hasn’t it diffused into general pathology practice? Technical and economic barriers to adoption have predominantly driven the current state, but a broadening of applications for DIA (enabled by AI and driven by a demand for biomarkers and companion testing) may see an explosion of DIA on the horizon. The current and future state of DIA in pathology are discussed below.

DIGITAL IMAGING TECHNOLOGY

Digital pathology is sometimes defined narrowly as a synonym for WSI, while others promote a broader definition of digital pathology that encompasses many digital imaging modalities and associated technologies. The former definition, for example, is adopted by the Digital Pathology Association (whose founder, Dirk Soenksen, is often credited with originating the term), while the latter is frequently used by academic thought leaders.8,9 This review takes the latter approach, favoring a broader discussion of DIA that includes diverse digital imaging modalities in addition to WSI. This is not necessarily an endorsement of the broader definition of digital pathology as much as a recognition that DIA is very much the same process regardless of how the digital images themselves are acquired. Where there are significant differences between modalities, these will be noted.

Digital Photomicrography

In comparison to WSI scanners, microscope-mounted digital cameras are inexpensive, plentiful, and more democratically distributed. In academic centers, for example, it is not uncommon for all histopathologists to have digital cameras in their offices, while the few WSI scanners will be concentrated in the histology laboratory or scanning core facility. For those without access to dedicated microscope-mounted cameras, smartphone cameras can be used with inexpensive eyepiece adapters to obtain high-quality digital images. The literature demonstrates numerous methods for analyzing digital photomicrographs including commercial packages, open source software, and web-based image analysis sites.

While digital photomicrography is highly accessible, it has a number of significant disadvantages when compared with automated digital microscopy and WSI. The biggest issue with using manual acquisition of images is the lack of standardization in acquisition parameters such as color and lighting. This can result in significant issues with test reproducibility. While it is not impossible to overcome these issues, the use of automated microscopes and WSI scanners specifically designed with DIA in mind is far preferred in a clinical setting.

Automated Digital Microscopy

Before the advent of WSI, automated microscopes were the standard for clinical DIA. This class of devices includes the first system approved by the FDA for DIA of IHC, the Automated Cellular Imaging System from Chromavision (Table 1). The Automated Cellular Imaging System went on to serve as the predicate for most of the early FDA applications for estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2/neu (HER2/neu) DIA, including several other DIA-oriented automated digital microscopy systems. Like WSI systems, these systems overcome many of the issues with reproducibility that arise with manual imaging. While these systems have mostly been replaced by WSI, a few remain in service in clinical laboratories, despite waning support by vendors.

TABLE 1
TABLE 1:
FDA 510(k) Cleared Products for Clinical Image Analysis in Pathology

Whole-slide Imaging

For brightfield imaging of IHC, WSI has become the mainstay for DIA. WSI has several key advantages in terms of both technical and workflow perspectives. From a technical perspective, WSI scanners enable a more reproducible and well-controlled imaging product as compared with the use of digital photomicrography. This includes consistent color reproduction and lighting, both of which are key to accurate and precise DIA. From a workflow perspective, WSIs allow the burden of imaging slides to be transferred from the pathologist to a technician. This permits the testing to be more readily standardized and helps ensure better documentation and compliance with regulatory requirements.

CLINICAL APPLICATIONS OF DIGITAL IMAGE ANALYSIS

While the literature has numerous examples of DIA applied to a variety of biomarkers and disease processes, the common clinical applications of DIA are actually quite restricted. Breast cancer biomarkers are by far the largest clinical application of DIA in surgical pathology. The markers include HER2/neu, ER, PR, and Ki-67.

Hormone Receptors in Breast Cancer

Hormones, especially estrogen, have long been recognized as a driver of proliferation in some breast cancers. Hormone receptors were originally measured in breast cancer tissue using the dextran-coated charcoal method, then with enzyme immunoassay and enzyme-linked immunosorbent assay, and eventually using quantitative IHC on tissue sections.10 Patients with ER and PR positive breast cancers have a better prognosis, and are likely to benefit from endocrine therapy. For this reason, the National Comprehensive Cancer Network (NCCN) recommends that ER and PR status should be evaluated by IHC for all invasive breast cancer and DCIS.11

Per the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guideline breast cancers are considered positive for ER or PR if ≥1% of tumor nuclei are positive by IHC.12 Reporting of positivity using the Allred or H score is also acceptable, but does not change the 1% threshold.13 Some evidence suggests that carcinomas with low expression of ER and PR (1% to 10%) may not respond as well to endocrine therapy, and a threshold of 10% is used in Europe.

Image analysis has been used to quantify ER in breast tissues since the late 1980s with early examples employing specialized imaging devices and analysis techniques.14,15 As ER and PR are expressed in the nucleus, application of modern DIA techniques is rather straightforward (Fig. 2). DIA has been routinely applied to WSI for over a decade, and there are numerous examples of excellent performance in the literature for both ER and PR.16,17

FIGURE 2
FIGURE 2:
Image analysis algorithm for nuclear immunohistochemistry. This example shows the general steps of nuclear immunohistochemical image analysis using whole-slide imaging. A, A section of pancreatic neuroendocrine tumor is stained with hematoxylin and Mib-1 antibody using DAB. A “hotspot” region of interest is selected (green box) and shown at higher magnification (B). C, The image is segmented into the tumor (green) and nontumor (red). This step is frequently manual, but can also be performed using machine learning classifiers. D, Color deconvolution is applied to (B) to separate the hematoxylin (left) and DAB (right) stains into separate grayscale images, pseudocolored blue and brown, respectively. E, The nuclei in the hematoxylin and DAB images are processed slightly and segmented using grayscale thresholding. F, The segmentation in image (C) is applied to the images in (E) to exclude nontumor nuclei. G, The final overlay, showing the negative tumor nuclei in blue and the positive tumor nuclei in brown. Nine positive nuclei were identified out of 570 total tumor nuclei; the Ki-67 proliferation index is 1.6%.

HER2/neu

HER2/neu (or erbB2) is a receptor tyrosine-protein kinase and proto-oncogene involved in cell division, proliferation, and differentiation. Amplification of the ERBB2 gene results in overexpression of the HER2/neu protein and has been identified in a number of cancer types, including breast, ovarian, bladder, lung, and stomach cancers. HER2/neu amplification is associated with a worse prognosis, but also predicts response to trastuzumab, a HER2/neu-blocking antibody that is FDA-approved for use in certain types of breast and gastroesophageal cancers with HER2/neu overexpression. In the case of both breast and gastric carcinoma, HER2/neu expression is first evaluated by IHC, which is scored as 0 (negative), +1 (negative), +2 (equivocal), or +3 (positive). The scoring criteria are different for breast and gastric cancer, but in both cases equivocal scores are reflexed to further evaluation of gene copy number by FISH.18,19 A unique aspect to DIA for HER2/neu is that the protein is membranous. Unlike DIA for nuclear proteins, for which hematoxylin or DAPI can be used to define the nuclear compartment (Fig. 2), membrane counterstains are not typically employed. This makes HER2/neu DIA considerably more challenging and requires the use of sophisticated image analysis algorithms.

HER2/neu in Breast Cancer

HER2/neu testing should be performed on every primary invasive breast cancer and metastasis, if tissue is available.19 HER2/neu status predicts response to HER2-targeted therapy (eg, trastuzumab) and must be established before initiating therapy.20 DIA is commonly performed in clinical laboratories to measure HER2/neu expression. A few years ago, UT Southwestern Medical Center reported 10 years of their experience with clinical DIA for HER2/neu.20 They reported and overall concordance between IHC and FISH of 87.3% (1768/2026). Their concordance improved steadily over the decade and was consistently around 95% for the last 3 years they report.20

Recently, the College of American Pathologists (CAP) convened a panel of experts and published a guideline for quantitative DIA of HER2/neu IHC in breast cancer.21 A systematic review of the literature was performed, which resulted in 11 recommendations with the intent of accuracy and reproducibility of results on clinical specimens. These are discussed in more detail below.

HER2/neu in Gastroesophageal Adenocarcinoma

In 2010, the ToGA (Trastuzumab for Gastric Cancer) clinical trial demonstrated that the addition of trastuzumab significantly improved survival of patients with advanced (unresectable local-regional, recurrent, or metastatic) HER2/neu-positive gastric and gastroesophageal adenocarcinoma.22 A recent guideline from the CAP, ASCP, and ASCO recommends that all patients with advanced gastroesophageal adenocarcinoma who are candidates for trastuzumab therapy should have their carcinoma tissue tested for HER2 overexpression and/or amplification.19

While there is significant methodological overlap in the quantitation of HER2/neu expression in gastroesophageal and breast cancer, important differences exist. Notably, gastroesophageal adenocarcinomas exhibit a greater degree of within-patient heterogeneity in HER2/neu expression. Furthermore, gastroesophageal adenocarcinomas with amplification rarely show the complete membranous staining required to score breast cancer cases as positive by IHC. To account for these differences, the modified criteria of Hofmann and colleagues were adopted in the gastroesophageal guideline.19,23

The majority of IHC testing of HER2/neu for gastroesophageal adenocarcinoma is performed manually, however, Koopman et al24 recently showed strong agreement between manual and DIA methods. This group confirmed the need to adjust DIA parameters for gastroesophageal adenocarcinoma, achieving an 85.6% agreement between manual and DIA scores when gastroesophageal adenocarcinoma-specific cutoffs were used as compared with 75.8% for breast cancer cutoffs.24 Disagreement consisted entirely of equivocal (2+) cases, and DIA showed superior agreement with subsequent in situ hybridization (ISH) results, which would have reduced the number of cases reflexed to ISH. Nielsen et al25 found a similar reduction in the number of equivocal cases requiring ISH when DIA is employed.

Ki-67 Proliferation Index (PI)

Ki-67 is expressed by cells during the active portions of the cell cycle, but absent during the G0 phase. For this reason, Ki-67 has become a common biomarker for cell proliferation in a variety of malignancies. Like ER and PR, the Ki-67 protein is localized to the nucleus, so that a hematoxylin counterstain can be readily used as a counterstain. A simplified example of a DIA algorithm for Ki-67 is shown in Figure 2.

Ki-67 Proliferation Index in Breast Cancer

A number of studies support a role for Ki-67 as a prognostic and predictive biomarker.26–28 Ki-67 is capable of identifying patients with ER-positive breast cancer that would benefit from additional chemotherapeutic regimens, however it has not been found to be useful in all breast cancer cases.29 And other studies have not supported a predictive role for Ki-67.30 The Breast Cancer Panel of the NCCN has reviewed the existing evidence for Ki-67 as a predictive and prognostic biomarker and found it inconclusive. For this reason, the 2019 NCCN Guidelines do not recommend Ki-67 PI assessment for breast cancer cases.31 Despite this fact, many pathologists regularly perform quantitative Ki-67 IHC by request of treating oncologists or as part of clinical trials.

Standardization is a recognized issue with measuring Ki-67 PI, regardless of whether it is done manually or using DIA. To address issues of standardization, the International Ki-67 in Breast Cancer Working Group has conducted a series of studies on the reproducibility of intralaboratory and interlaboratory Ki-67 PI in breast cancer.32–34 This group established that there is poor concordance of manual Ki-67 PI between laboratories unless analysis methods are systematically standardized between the laboratories. The group subsequently used the same set of breast cancer slides to study DIA of Ki-67 PI in 14 different laboratories employing 10 different software packages.35 This revealed an intraclass correlation coefficient (ICC) of 0.83 (95% credible interval: 0.73-0.91) for the unstandardized DIA methods, similar to ICC of 0.87 (95% credible interval: 0.81-0.93) from their prior study using standardized manual methods.34,35

Ki-67 Proliferation Index in Gastrointestinal and Pancreatic Neuroendocrine Tumors (NETs)

The grading of pancreatic and gastrointestinal NETs is based entirely on the proliferation rate of the neoplastic cells. Two methods can be used for determining proliferation rate in NETs: (1) counting mitoses in hematoxylin and eosin (H&E) slides (ie, mitotic rate) or (2) measuring the Ki-67 PI. Mitotic rate has generally fallen out of favor due to the subjectivity of identifying mitotic figures and the need to evaluate large numbers of high-power fields, which is awkward, time-consuming, and sometimes impossible (in small samples). Evidence also suggests that Ki-67 PI is a better predictor of outcomes when mitotic rate and Ki-67 PI are discordant.36 While pathologists generally prefer using Ki-67 PI, the process is not without difficulties. A feature of almost all NETs is the abundance of admixed stroma which includes various cells types that can be briskly proliferative. These stromal cells, especially those that are positive for Ki-67, act as significant distractors when attempting to quantify Ki-67 using manual and traditional DIA methods.37 When manually-drawn regions of interest are used to exclude stromal cells, DIA and systematic manual counting are highly concordant (ICC=0.98) and superior to estimation by the “eyeball” method.38

FDA 510(k) CLEARANCES

In the United States, devices intended for clinical DIA of IHC are considered class II in vitro devices and are submitted to the FDA using the 510(k) approval process. The 510(k) application is a premarket submission demonstrating that the in vitro device is safe, effective and substantially equivalent to an already approved predicate device. Table 1 summarizes all the FDA cleared devices for automated image analysis of IHC (see the Supplemental Digital Content 1, http://links.lww.com/PAP/A29 for a more detailed listing of FDA clearances). The earliest of the cleared devices are based on automated microscopes and most of these early systems are no longer marketed and may not even be supported by their manufacturers. More recent devices are based on WSI scanners. At the time of this writing, no new devices of this type have achieved 510(k) clearance since 2015.

VERIFICATION AND VALIDATION

All clinical laboratories in the United States are regulated by the Clinical Laboratory Improvement Amendments (CLIA), which was passed in 1988 and revised in 2003. Section 493.1253 of CLIA requires that laboratories determine the performance characteristics of tests before they are used clinically. Broadly speaking, there are 2 approaches to establishing the performance of tests: verification and validation.

Verification is the process of establishing that FDA-approved tests perform according to the specifications established by the manufacturer for the population of patients served by the laboratory. The characteristics that should be verified include accuracy, precision, reportable range, and reference interval, as applicable.

Validation is considered a more rigorous study of test performance, intended for laboratory-developed tests or for modified/off-label use of FDA-approved tests. The characteristics to be measured include accuracy, precision, reportable range, reference interval, analytical sensitivity, analytical specificity, and other characteristics, as applicable. In practice (especially in anatomic pathology), the term “validation” is often used to refer to both verification and validation, but validation correctly applies to the de novo determination of the clinical performance of a test rather than confirmation of a manufacturer’s specifications.

For details on validating IHC assays, please see Fitzgibbons et al.39 Also, as noted above, the CAP recently published a guideline for quantitative DIA of HER2/neu IHC in breast cancer.21 Although this guideline reviews evidence for DIA of HER2/neu, the guideline contains a lot of good general advice about implementing clinical DIA and performing test validation. Hopefully the CAP will introduce similar guidelines for other common applications of DIA in the near future.

DIGITAL IMAGE ANALYSIS WORKFLOW

Although DIA has been used clinically for several decades, standard models for integration into clinical workflows do not exist. As a result, each laboratory is left to integrate DIA with their information systems and to develop their own workflows. For smaller laboratories with lower quantitative IHC volumes, this can be quite challenging and present a significant barrier to the adoption of DIA. For this reason, it is not surprising that the vast majority of DIA testing in the United States is performed at either large reference laboratories or in the academic setting.

Integration of DIA with information systems is discussed below, but in general, almost all laboratories performing DIA still rely heavily on personnel to plug various holes in the overall workflow. This includes performing tasks like monitoring work queues for DIA orders, pulling slides for scanning, setting up DIA algorithms on a per-slide basis, selecting fields of view, annotating WSIs, and transferring DIA results back into the patient report. While some of this work can be done by trained imaging specialists, pathologists are frequently required to select fields, perform annotations on slides, and validate results. The required commitment in full time equivalents is frequently surprising to those unfamiliar with DIA, and the additional effort can often outweigh the inherent benefits of using automated methods for biomarker quantitation.

Digital Image Analysis and Information Systems

DIA systems exist in diverse information system environments. While almost all laboratories with DIA will have a laboratory information systems (LIS), other digital pathology components will vary considerably. For example, while some laboratories have implemented a digital pathology system specifically for clinical use, others may piggyback their DIA on a general purpose or research-oriented instance. Even those laboratories that have clinical digital pathology system may or may not have an integration with their clinical LIS. Furthermore, not all LIS integrations are created equally. Ideally, DIA should be orderable by pathologists in the LIS. In turn, that order should be transmitted to the DIA system and the result returned to the LIS as discrete data. In practice, few laboratories performing DIA have achieved this level of integration with the LIS.

With regard to digital pathology systems, DIA software can either be (1) an integrated solution or (2) an interfaced solution. These terms will be familiar to many as the situation mirrors similar arrangements between the electronic health record and LIS. As with the electronic health record and LIS, integrated and interfaced approaches each have respective strengths and weaknesses.

Integrated solutions (sometimes also called “monolithic”) are single vendor offerings that seamlessly integrate DIA functionality within the digital pathology system (ie, WSI repository and viewer). A few of the digital pathology systems offer integrated DIA functionality, typically as an add-on to the base product. These DIA tools, while they may not represent best-of-breed solutions, typically support the most common DIA applications, and many of them even have 510(k) clearance for use in breast tissue (Table 1). The biggest advantage that these tools have is that they greatly simplify clinical deployment.

The second option is to interface third party DIA tools to a separate digital pathology system. Unlike other medical system interfaces, which usually rely on HL7, these “interfaces” are typically implemented using application programming interfaces specific to the digital pathology system and DIA software. While no industry standards for these interfaces exist, a few DIA companies have achieved widespread compatibility with industry-leading digital pathology systems. Other DIA companies have very limited experience with interfacing, and those inquiring about interfaces may be directed to their software development group rather than sales. This is never an encouraging sign for a new interface project.

Of course, many clinical laboratories performing DIA have neither an integrated nor an interfaced DIA solution. In these cases, there may be no clinical digital pathology system at all, with slides digitized on demand and manually transferred to a standalone DIA system. While this may sound inefficient, numerous laboratories have used this approach successfully in the clinical laboratory.

Digital Imaging and Communications in Medicine (DICOM) Standards and Digital Image Analysis

DICOM is the international standard for the storage, retrieval, transmission, and manipulation of medical images and related information. DICOM has its origins in radiology in the 1980s as a response to the rise of digital diagnostic imaging modalities such as computed tomography and magnetic resonance imaging.40 It has subsequently expanded to cover other specialties, including gastroenterology, cardiology, dentistry, ophthalmology, and others. DICOM Working Group 26 (WG-26) is charged with supporting the pathology domain and is actively working with Integrating the Healthcare Enterprise (IHE) Pathology and Laboratory Medicine (PaLM) to create profiles relevant to the practice of pathology. WG-26 added support for anatomic pathology to the DICOM standard by defining a specimen model in 2008 and a WSI standard in 2010.41,42 While there has been little-to-no adoption of DICOM for pathology in the interim, the recent rise of clinical digital pathology has renewed interest. This has resulted in several informal DICOM “Connectathons” at national and international meetings in the last few years.43 Participants have included major WSI scanner vendors, digital pathology system vendors, and several PACS vendors. DIA software companies, however, have been notably absent from these Connectathons.

While DICOM is commonly thought of as standard for encoding and exchanging the pixel data of medical images, it also defines standards for image annotation and DIA results. This makes DICOM theoretically capable of mediating the exchange of pixel data, input regions of interest, and output overlays between a PACS, a WSI viewer, and DIA software. In addition, DICOM could facilitate the transfer of orders and results between DIA software and the LIS. While this is a promising model of interoperability for the future, some have questioned whether the current DICOM standard is robust enough to address gigapixel-scale coordinate systems and the potential for hundreds of thousands of annotation objects. In late 2019, DICOM WG-26 formed the WSI Annotations Ad-hoc Group to address this issue directly.

REIMBURSEMENT FOR DIGITAL IMAGE ANALYSIS

While clinical DIA has advantages over manual analysis, it tends to be expensive for low-to-medium volume testing. Additional capital costs include imaging hardware (eg, digital imaging equipment/automated microscopes or WSI slide scanners), DIA software, and system interfaces. Recurring costs include the salaries of additional personnel to support DIA workflows, hardware/software support contracts, and ongoing software licensing costs. Notably, most of these expenses are variable costs. Per test cost will decrease as the testing volume increases, which explains why most clinical DIA is performed at high volume laboratories.

The Medicare Physician Fee Schedule (PFS)

A discussion of reimbursement for clinical DIA that reflects all payers and all locales is beyond the scope of this review. Furthermore, robust reimbursement information from private payers is difficult if not impossible to obtain, making a “complete discussion” impossible. Reimbursement rates are, however, readily available for the Centers for Medicare and Medicaid (CMS) which publishes PFS information annually in both the Federal Register and on the CMS website.44Table 2 summarizes the 2019 PFS for CMS’s Healthcare Common Procedure Coding System (HCPCS) codes for the various types of interpretation of IHC/immunocytochemistry. HCPCS is composed of 3 levels of codes with level I codes corresponding to Current Procedural Terminology (CPT) codes (copyright 2019 American Medical Association). The procedure codes that include quantitative and semiquantitative “morphometric analysis” of IHC are part of the HCPCS level I/CPT code sets. Because the HCPCS code system is (essentially) a superset of the CPT code set, we will refer to these billing codes as HCPCS codes.

TABLE 2
TABLE 2:
Centers for Medicare and Medicaid (CMS) Fee Schedule for Immunohistochemistry Procedures

The PFS lists maximum fees that Medicare will pay physicians or other providers and suppliers for services rendered on a fee-for-service basis. For both the professional component and technical component, Table 2 lists 3 prices: the national payment, the minimum payment, and the maximum payment. These prices reflect variations in payments among the numerous Medicare Administrative Contractor (MAC) localities in the United States. For each MAC, the relative value units assigned to a HCPCS code are further weighted to reflect local costs of providing the service. The national payment amount has a weight of 1.000 applied. The minimum and maximum prices in Table 2 reflect the MAC with the lowest and highest payment amounts, respectively.

Reimbursement for Manual Versus “Computer-assisted” Analysis

To compare reimbursement for manual analysis and DIA, we will focus on just the national reimbursement payment. Two HCPCS codes cover “quantitative or semiquantitative” analysis of tumor IHC: 88360 and 88361. Both codes are billable for “each single antibody stain procedure,” and there aren’t any additional codes for analysis of multiplex antibodies. Code 88360 should be billed for manual quantitative or semiquantitative analysis, and 88361 should be billed for quantitative or semiquantitative analysis “using computer-assisted technology.” While the existence of a differential payment between 88360 and 88361 is well-known and frequently cited, the magnitude of this differential is surprisingly small. For the national payment rate, the difference in professional component is $3.24 and the difference in technical component is $1.08, an increase in reimbursement of only 7.3% and 1.3%, respectively, for using DIA.

Digital Image Analysis That May Not Be Reimbursable

All existing HCPCS codes for quantitative or semiquantitative analysis “using computer-assisted technology” apply only to IHC and ISH staining techniques (ISH reimbursement was not covered in detail). Notably absent is a code for DIA applied to H&E-stained sections or non-IHC special stains. While this may seem odd at first, this situation parallels the absence of manual “quantitative or semiquantitative” codes for non-IHC stains. So while there may be valid and compelling reasons to use DIA to quantify H&E or special stains, no applicable procedure codes exist for increased technical or professional reimbursement.

Codes 88360 and 88361 are intended specifically for “tumor IHC.” “Tumor” was an unintentionally poor choice of word in the description for the codes, as it is generally understood that they apply to a malignancy or neoplasm rather than a nonspecific swelling.45 In most instances this definition is not limiting since almost all quantitative IHC is performed on neoplasms such as breast carcinomas or NETs. Quantitative IHC is far less commonly performed on “nontumor” specimens, despite the existence of valid clinical applications, including the measurement of immune and inflammatory cell densities (eg, lymphocytes, plasma cells, and mast cells) in various medical conditions.46 Whether these applications of DIA are reimbursable will depend heavily on the payer being billed.

THE FUTURE OF DIGITAL IMAGE ANALYSIS IN CLINICAL LABORATORIES

As detailed above, adoption of DIA in the clinical laboratory still faces a number of barriers. Issues include awkward workflows, increased turnaround time and increased costs (including hardware, software, and personnel costs) that are not effectively offset by reimbursement unless the DIA is being performed at very high volumes. The benefits of DIA, such as reproducibility and objectivity, have been demonstrated in the literature, but are not always sufficient to drive adoption. A significant rebalancing of this equation is coming, though, as more laboratories integrate digital pathology in their routine clinical workflow. With digital workflows in place, the added cost and complexity of DIA becomes negligible when compared with the benefits.

Adoption of DIA will also be driven by novel applications such as the expansion of existing biomarkers to new tissue types, the implementation of new biomarkers and companion tests, such as the measurement programmed death-ligand 1 and tumor-infiltrating lymphocytes using CD8 to predict response to immune checkpoint inhibitors.47–49

Finally, no discussion of the future of DIA can ignore the ongoing artificial intelligence revolution. Both classic machine learning and deep learning techniques (such as convolutional neural networks) are already being applied to many of the actionable biomarkers described in this review, either as a tissue classifier or to perform the DIA directly.50–52 Some approaches even bypass IHC entirely, predicting outcomes or biomarker status directly from H&E sections.53–55

REFERENCES

1. McNamara G, Difilippantonio MJ, Ried T. Microscopy and image analysis. Curr Protoc Hum Genet. 2005;46:4.4.1–4.4.34.
2. Pantanowitz L, Sharma A, Carter A, et al. Twenty years of digital pathology: an overview of the road travelled, what is on the horizon, and the emergence of vendor-neutral archives. J Pathol Inform. 2018;9:40.
3. Mason DY, Gatter KC. The role of immunocytochemistry in diagnostic pathology. J Clin Pathol. 1987;40:1042–1054.
4. Lichter P, Ward DC. Is non-isotopic in situ hybridization finally coming of age? Nature. 1990;345:93–94.
5. Kaplan KJ. Year in review. Tissuepathology.com. Digit Pathol Blog; 2010. Available at: https://tissuepathology.com/2010/12/30/year-in-review/. Accessed September 5, 2019.
6. Schmidhuber J. History of computer vision contests won by deep CNNs on GPUs. 2017. Available at: http://people.idsia.ch/~juergen/computer-vision-contests-won-by-gpu-cnns.html. Accessed September 5, 2019.
7. US Food and Drug Administartion (FDA), Office of the Commissioner. FDA allows marketing of first whole slide imaging system for digital pathology; 2017. Available at: www.fda.gov/news-events/press-announcements/fda-allows-marketing-first-whole-slide-imaging-system-digital-pathology. Accessed June 4, 2019.
8. Pantanowitz L. Digital images and the future of digital pathology. J Pathol Inform. 2010;1:15.
9. Jara-Lazaro AR, Thamboo TP, Teh M, et al. Digital pathology: exploring its applications in diagnostic surgical pathology practice. Pathology (Phila). 2010;42:512–518.
10. Nofech-Mozes S, Vella ET, Dhesy-Thind S, et al. Systematic review on hormone receptor testing in breast cancer. Appl Immunohistochem Mol Morphol. 2012;20:214–263.
11. Allred DC, Carlson RW, Berry DA, et al. NCCN Task Force Report: estrogen receptor and progesterone receptor testing in breast cancer by immunohistochemistry. J Natl Compr Canc Netw. 2009;7(suppl 6):S1–S21.
12. Hammond MEH, Hayes DF, Dowsett M, et al. American Society of Clinical Oncology/College of American Pathologists Guideline Recommendations for Immunohistochemical Testing of Estrogen and Progesterone Receptors in Breast Cancer. J Clin Oncol. 2010;28:2784–2795.
13. Shousha S. Oestrogen receptor status of breast carcinoma: Allred/H score conversion table. Histopathology. 2008;53:346–347.
14. Colley M, Kommoss F, Bibbo M, et al. Assessment of hormone receptors in breast carcinoma by immunocytochemistry and image analysis. II. Estrogen receptors. Anal Quant Cytol Histol. 1989;11:307–314.
15. Bacus S, Flowers JL, Press MF, et al. The evaluation of estrogen receptor in primary breast carcinoma by computer-assisted image analysis. Am J Clin Pathol. 1988;90:233–239.
16. Faratian D, Kay C, Robson T, et al. Automated image analysis for high-throughput quantitative detection of ER and PR expression levels in large-scale clinical studies: the TEAM Trial Experience. Histopathology. 2009;55:587–593.
17. Turbin DA, Leung S, Cheang MCU, et al. Automated quantitative analysis of estrogen receptor expression in breast carcinoma does not differ from expert pathologist scoring: a tissue microarray study of 3,484 cases. Breast Cancer Res Treat. 2008;110:417–426.
18. Wolff AC, Hammond MEH, Allison KH, et al. Human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists Clinical Practice Guideline Focused Update. J Clin Oncol. 2018;36:2105–2122.
19. Bartley AN, Washington MK, Colasacco C, et al. HER2 Testing and Clinical Decision Making in Gastroesophageal Adenocarcinoma: Guideline From the College of American Pathologists, American Society for Clinical Pathology, and the American Society of Clinical Oncology. J Clin Oncol. 2017;35:446–464.
20. Sarode VR, Xiang QD, Christie A, et al. Evaluation of HER2/neu status by immunohistochemistry using computer-based image analysis and correlation with gene amplification by fluorescence in situ hybridization assay: a 10-year experience and impact of test standardization on concordance rate. Arch Pathol Lab Med. 2015;139:922–928.
21. Bui MM, Riben MW, Allison KH, et al. Quantitative image analysis of human epidermal growth factor receptor 2 immunohistochemistry for breast cancer: guideline From the College of American Pathologists. Arch Pathol Lab Med. 2019;143:1180–1195.
22. Bang Y-J, Van Cutsem E, Feyereislova A, et al. Trastuzumab in combination with chemotherapy versus chemotherapy alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet. 2010;376:687–697.
23. Hofmann M, Stoss O, Shi D, et al. Assessment of a HER2 scoring system for gastric cancer: results from a validation study. Histopathology. 2008;52:797–805.
24. Koopman T, de Bock GH, Buikema HJ, et al. Digital image analysis of HER2 immunohistochemistry in gastric- and oesophageal adenocarcinoma: a validation study on biopsies and surgical specimens. Histopathology. 2018;72:191–200.
25. Nielsen SL, Nielsen S, Vyberg M. Digital image analysis of HER2 immunostained gastric and gastroesophageal junction adenocarcinomas. Appl Immunohistochem Mol Morphol. 2017;25:320–328.
26. Dumontet C, Krajewska M, Treilleux I, et al. BCIRG 001 molecular analysis: prognostic factors in node-positive breast cancer patients receiving adjuvant chemotherapy. Clin Cancer Res. 2010;16:3988–3997.
27. Luporsi E, André F, Spyratos F, et al. Ki-67: level of evidence and methodological considerations for its role in the clinical management of breast cancer: analytical and critical review. Breast Cancer Res Treat. 2012;132:895–915.
28. Jones RL, Salter J, A’Hern R, et al. The prognostic significance of Ki67 before and after neoadjuvant chemotherapy in breast cancer. Breast Cancer Res Treat. 2009;116:53–68.
29. Andre F, Arnedos M, Goubar A, et al. Ki67—no evidence for its use in node-positive breast cancer. Nat Rev Clin Oncol. 2015;12:296–301.
30. Viale G, Regan MM, Mastropasqua MM, et al. On the behalf of the International Breast Cancer Study Group. Predictive value of tumor Ki-67 expression in two randomized trials of adjuvant chemoendocrine therapy for node-negative breast cancer. J Natl Cancer Inst. 2008;100:207–212.
31. Gradishar WJ, Anderson BO, Abraham J. NCCN guidelines index table of contents discussion. Breast Cancer. 2019:216.
32. Polley M-YC, Leung SCY, McShane LM, et al. An International Ki67 Reproducibility Study. J Natl Cancer Inst. 2013;105:1897–1906.
33. Polley M-YC, Leung SCY, Gao D, et al. An international study to increase concordance in Ki67 scoring. Mod Pathol. 2015;28:778–786.
34. Leung SCY, Nielsen TO, Zabaglo L, et al. Analytical validation of a standardized scoring protocol for Ki67: phase 3 of an international multicenter collaboration. NPJ Breast Cancer. 2016;2:16014.
35. Rimm DL, Leung SCY, McShane LM, et al. An international multicenter study to evaluate reproducibility of automated scoring for assessment of Ki67 in breast cancer. Mod Pathol. 2019;32:59–69.
36. McCall CM, Shi C, Cornish TC, et al. Grading of well-differentiated pancreatic neuroendocrine tumors is improved by the inclusion of both Ki67 proliferative index and mitotic rate. Am J Surg Pathol. 2013;37:1671–1677.
37. Reid MD, Bagci P, Ohike N, et al. Calculation of the Ki67 index in pancreatic neuroendocrine tumors: a comparative analysis of four counting methodologies. Mod Pathol. 2015;28:686–694.
38. Tang LH, Gonen M, Modlin IM. Objective quantification of the Ki67 proliferative index in neuroendocrine tumors of the gastroenteropancreatic system. Am J Surg Pathol. 2012;36:10.
39. Fitzgibbons PL, Goldsmith JD, Souers RJ, et al. Analytic validation of immunohistochemical assays: a comparison of laboratory practices before and after introduction of an evidence-based guideline. Arch Pathol Lab Med. 2017;141:1247–1254.
40. DICOM Standard. History. 2019. Available at: www.dicomstandard.org/history/. Accessed August 1, 2019.
41. DICOM Standards Committee, Working Groups 26, Pathology. Digital Imaging and Communications in Medicine (DICOM) Supplement 122: Specimen Module and Revised Pathology SOP Classes; 2008. Available at: https://medical.nema.org/medical/dicom/final/sup122_ft2.pdf. Accessed August 1, 2019.
42. DICOM Standards Committee, Working Groups 26, Pathology. Digital Imaging and Communications in Medicine (DICOM) Supplement 145: Whole Slide Microscopic Image IOD and SOP Classes; 2010. Available at: https://medical.nema.org/medical/dicom/final/sup145_ft.pdf. Accessed August 1, 2019.
43. Clunie D, Hosseinzadeh D, Wintell M, et al. Digital imaging and communications in medicine whole slide imaging connectathon at Digital Pathology Association Pathology Visions 2017. J Pathol Inform. 2018;9:6.
44. Centers for Medicare and Medicaid Services (CMS). Physician Fee Schedule search. 2019. Available at: www.cms.gov/apps/physician-fee-schedule/search/search-criteria.aspx. Accessed August 1, 2019.
45. Zaydfudim V, Sarr MG, Harken AH, et al. Tumor vs neoplasm: it is not just semantics. Surgery. 2013;153:603–604.
46. Akiyama Y, Maeda D, Morikawa T, et al. Digital quantitative analysis of mast cell infiltration in interstitial cystitis. Neurourol Urodyn. 2018;37:650–657.
47. Hartman DJ, Ahmad F, Ferris RL, et al. Utility of CD8 score by automated quantitative image analysis in head and neck squamous cell carcinoma. Oral Oncol. 2018;86:278–287.
48. Le DT, Uram JN, Wang H, et al. PD-1 blockade in tumors with mismatch-repair deficiency. N Engl J Med. 2015;372:2509–2520.
49. Koelzer VH, Sirinukunwattana K, Rittscher J, et al. Precision immunoprofiling by image analysis and artificial intelligence. Virchows Arch. 2019;474:511–522.
50. Wang S, Yang DM, Rong R, et al. Pathology image analysis using segmentation deep learning algorithms. Am J Pathol. 2019;189:1686–1698.
51. Xing F, Cornish TC, Bennett T, et al. Pixel-to-pixel learning with weak supervision for single-stage nucleus recognition in Ki67 images. IEEE Trans Biomed Eng. 2019;66:3088–3097.
52. Akbar S, Jordan LB, Purdie CA, et al. Comparing computer-generated and pathologist-generated tumour segmentations for immunohistochemical scoring of breast tissue microarrays. Br J Cancer. 2015;113:1075–1080.
53. Couture HD, Williams LA, Geradts J, et al. Image analysis with deep learning to predict breast cancer grade, ER status, histologic subtype, and intrinsic subtype. NPJ Breast Cancer. 2018;4:30.
54. Beck AH, Sangoi AR, Leung S, et al. Systematic analysis of breast cancer morphology uncovers stromal features associated with survival. Sci Transl Med. 2011;3:108–113.
55. Shamai G, Binenbaum Y, Slossberg R, et al. Artificial intelligence algorithms to assess hormonal status from tissue microarrays in patients with breast cancer. JAMA Netw Open. 2019;2:e197700.
Keywords:

image analysis; biomarkers; digital pathology; Ki-67; HER2/neu; estrogen receptor; progesterone receptor

Supplemental Digital Content

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.