perkinelmer-728×90-20191127-1227

Digital morphology is the next chapter in hematology

Digital morphology is the next chapter in hematology

An important issue ahead is the standardization of digital-image quality in hematology. With further technological advances in digital imaging/morphology, diagnostic applications and tele-hematology could become feasible in the future.

Laboratory automation has been for long limited to clinical chemistry and immunochemistry platforms. The major hurdle encountered in developing models of automation for laboratory hematology is represented by the peculiar sample type, which is whole blood anticoagulated with dipotassium ethylenediaminetetraacetic acid (EDTA). The presence of EDTA in the sample, which irreversibly sequestrates ionized calcium and many other metal ions, makes EDTA plasma an unsuitable sample matrix for clinical chemistry and even for coagulation testing, thus generating considerable obstacles for consolidation of laboratory hematology with other branches of laboratory medicine.

Nevertheless, a number of technological solutions have been developed for laboratory hematology in recent times. These basically include commercialization of modular, high-throughput, and versatile analyzers, which can be easily interconnected by means of sample conveyers, and can fit the organization of small, medium, and large facilities; integration of pre-analytical workstations, which can be identical to those included in models of total laboratory automation, or can be specifically designed to suit hematological testing; connection with automated slide strainers, which help improving the entire slide making process; as well as integration of automated image-analysis systems.

Throughout the relatively long history of laboratory hematology, the only reliable means for identifying, enumerating, and sizing blood cells has been for long represented by optical microscopy of peripheral blood smear. This practice carries many drawbacks, since it is inevitably time-consuming, is vulnerable to high inter- and intra-observer inaccuracy and imprecision, needs specific education and training of microscopists, and is poorly suited for rapid diagnostics as otherwise needed in patients with many acute hematological disorders.

Digital imaging

Digital imaging/morphology makes use of digital images and software algorithms to classify hematological cells, such as leukocytes and red blood cells. For a subset of leukocytes, digital system classification correlates well with the manual microscope method, the gold standard. Digital imaging thereby leads to a faster, more efficient, and more standardized way of performing a morphological analysis of a peripheral blood smear. Future-possible applications include the morphological analysis of other cell categories such as red blood cells and thrombocytes and digital analysis of other materials such as bone marrow samples. Furthermore, digital imaging can be used as an excellent learning tool. Independently organized quality surveys need to be implemented by manufacturers of digital microscopy systems. Integration of digital imaging with basic cell-counter results could lead to a faster detection and higher sensitivity and specificity in the detection of hematological malignancies.

Morphological analysis of the peripheral blood smear (PBS) is an essential element of hematological diagnostics. Traditionally, the analysis of a PBS has been performed by using the manual microscope. This method, however, is labor-intensive, requires continuous training of personnel, and is subject to relatively large inter-observer variability. The development of digital microscope systems, capable of using digital images of leukocytes, erythrocytes, and thrombocytes for classification, has been ongoing for more than a decade.

Companies have generated digital microscopy systems, especially in the field of pathology. However, these systems are still in the experimental phase with regard to implementation in routine (hematological) diagnostics, and are currently more used for diagnostic remote and scholar use. In essence, they are not now used as pathology filters in the diagnostic work-up. Some systems have, however, found a central place in the routine hematological peripheral blood smear screening.

The digital workflow in detail. The digital microscope (DM) makes use of a digital camera, which is capable of generating high-resolution images of leukocytes, red blood cells, and thrombocytes. The systems are equipped with software capable of not only detecting these cells, but also subsequently classifying them into the correct cell classes. This software application was developed using an artificial neural network, and it considers a large number of features, such as size, roundness, and size and shape of the nucleus for the morphological classification of leukocytes. The number of cells counted by the system can be set by the operator, and the DM will present the results in both absolute numbers and percentages. The classification results can then be judged by an experienced morphological research technician. The expert will validate the results and send them to the laboratory information system (LIS) and hospital information system (HIS).

Before digital microscopy systems can be implemented in daily hematological routine with regard to PBS analysis, these systems should be validated thoroughly. Currently, the manual microscope method is considered the standard. To date, various papers have described accuracy and precision studies of digital imaging when compared to manual assessment. It has been shown that digital microscopy shows good correlation with the standard with regard to the five-part differential, namely, segmented, eosinophilic, and basophilic granulocytes; lymphocytes; and monocytes. Nucleated red blood cells (NRBC) also display a good correlation. Moreover, digital imaging leads to reproducible results when comparing the results of independently operated digital microscope system of the same type and brand. Pre-classification of immature granulocytes and myeloid progenitor cells shows inadequate correlation with the manual method, requiring manual supervision to ensure proper results. Therefore, the mathematical algorithm to classify these classes should be further improved. However, one could argue that in a clinical situation the sole description of left-shift is sufficient for diagnostics and does not necessarily need specific sub-classification of individual myeloid progenitor cells. Combining this with state-of-the art immature granulocyte functions on cell counters could help with this issue.

Lymphocyte pathology. Correct sub-classification of specific lineage subset pathology, e.g., follicular lymphoma (FCL) or mantle cell lymphoma (MCL) in the case of abnormal malignant lymphocyte classes present, is currently still lacking. However, recent advances have been made in this specific area. Scientists have developed algorithms to discriminate hairy cells and chronic lymphocytic leukemia (CLL) cells from normal lymphocytes using digital imaging analysis. It seems likely that further improvement of these algorithms will facilitate the detection and classification of pathological cells in the nearby future.

Red blood cell morphology. Recent advances have also been made in the field of red blood cell morphology using digital imaging. The advanced RBC module generates between 2000 and 4000 individual images of red blood cells. The systems subsequently classify the cells on the basis of various morphological algorithms, comparable to the method for five-part leukocyte classification. The first steps toward standardization of the morphological analysis of the RBC lineage by use of digital imaging have recently been made. Recent studies using a morphological red blood cell module show lower within-run, between-run, and between-observer coefficients of variation when counting schistocytes compared to the coefficients observed for manual assessment. Digital imaging overestimates the percentage of schistocytes in the pre-classification, resulting in a high sensitivity. The subsequent low specificity, however, underlines the need for post-classification/manual supervision. Overestimation is also seen in the pre-classification of teardrop cells, for instance, seen in myeloproliferative diseases (i.e., myelofibrosis). Detection of inclusions has also been shown to be possible with digital imaging. These latter, specific applications of digital imaging, however, are currently not FDA-approved, and the underlying detection and decision algorithms need further improvement before these modules can be implemented safely in routine diagnostics.

Integration in daily practice. The introduction of digital microscopes implemented in total lab automation tracks (i.e., coupled with cell-counter and slide-maker stainers) paves the way for the implementation of clinical decision modules by coupling cell counter results with digital imaging results. This will undoubtedly lead to a higher sensitivity and specificity in the detection and classification of hematological diseases. Several requirements will have to be met, however, before one reaches this state of development. One of these requirements is continuous sample preparation.

Good-quality slide/sample preparation plays a pivotal role in good detection and subsequent classification of both leukocytes and red blood cells, using digital imaging. For instance, the digital imaging systems have difficulty separating neutrophilic granulocytes from basophilic granulocytes when the staining is of insufficient quality and gets too dark. Manually prepared slides are of insufficient quality to allow for a good and reproducible classification. Semi-automated and fully automated slide preparation and staining systems are the preferred methods of choice.

Way forward

To summarize, digital imaging seems to work for several cell classes in classifying both leukocytes and red blood cells in peripheral blood smears. The automated analysis of body fluids has also been investigated. Results from both cerebrospinal fluids and other body fluids (including abdominal fluids/ascites and CAPD) show good and clinically acceptable accuracy when comparing automated morphological analysis using digital imaging with the manual method. Even the pre-classification is good for most body fluid categories. A big drawback for the automated digital analysis of body fluids, however, is the lack of recognition of certain defined categories. For instance, the system is currently incapable of detecting and classifying mesothelial cells, and possible tumor cells present are frequently found in the category other (subcategory present in body fluid module). Another disadvantage is the analysis time; automated digital analysis of body fluids still requires significant manual correction and, therefore, takes longer than a full classification done manually. Improvement of the current digital body-fluid applications is needed before this module can be fully integrated in the routine body-fluid diagnostic process.

Digital images were already being used for scholarly purposes long before the introduction of digital microscopy systems. With the introduction of such systems, the application of digital imaging in educational settings has grown tremendously. Cell atlases and even mobile apps have been generated to teach students various normal and pathological cell classes that can be found in both PBS and body fluids. Pictures of classified or pre-classified cells can be sent to colleagues all over the world for a second opinion or to share rarely seen diseases. One can even imagine that imaging modules that are capable of pre-classification could send their pre-classification results to a central location for post-classification. This would be very helpful for hospital locations with low-volume PBS amounts, which struggle to maintain morphological expertise.

Share this:

Related Post

Stay Updated on Medical Equipment and Devices industry.
Receive our Daily Newsletter.