AI-enhanced evaluation software for immunofluorescence microscopy

Artificial intelligence (AI) can significantly enhance the evaluation of fluorescence patterns in indirect immunofluorescence assay (IFA) analyses. The EUROPattern Classifier software from EUROIMMUN utilizes deep convolutional neural networks to provide accurate IFA result proposals for a wide variety of autoantibodies.

Challenges of IFA evaluation

The indirect immunofluorescence assay (IFA) is a core method in autoantibody diagnostics due to its high sensitivity and specificity and its broad range of application. Traditional manual evaluation of the fluorescence results is, however, time-consuming and labour-intensive. Moreover, the subjective interpretation of fluorescence patterns and intensities can lead to intra- and inter-laboratory variability in result reporting.

A particular challenge is the presence of mixed patterns, which occur when more than one autoantibody is present in a sample. A shortage of technicians or specialists with sufficient experience in evaluating IFA may lead to limitations in the availability and quality of IFA testing services. Automation of the evaluation using advanced software can address many of these challenges and increase the accuracy, efficiency and reproducibility of IFA diagnostics.

AI-enhanced evaluation software

The EUROPattern Classifier is a state-of-the-art IFA evaluation software, which has been trained during its development using AI processes. Using defined datasets, the software has been taught to recognise specific features and patterns and to apply this knowledge to new data. The artificial networks are organized into multiple layers known as deep convolutional neural networks, which enable the evaluation of complex data.

The EUROPattern Classifier can thus accurately distinguish positive and negative IFA results, classify fluorescence patterns and determine titers. A high degree of overall agreement is achieved between the software-generated result proposals and traditional visual microscopy (refs 1 to 5). Results should nevertheless always be approved by a specialist before release. The EUROPattern Classifier is already established for a range of applications on cell substrates and has recently been extended to include more complex tissue substrates for diagnostics of organ-specific autoimmune diseases.

ANA classification

The detection of anti-nuclear antibodies (ANA) on HEp-2 cells is among the core applications of the EUROPattern Classifier. The software recognises nine fluorescence patterns in line with the International Consensus on Antinuclear Antibody Patterns (ICAP), as well as any combination of these patterns. It can also recognise patterns on HEp-20-10 cells, which are optimized to present more mitotic-stage cells. Further, the software supports evaluation of antibodies against double-stranded DNA (dsDNA) on Crithidia luciliae, accurately discriminating positive and negative samples by identifying the specific kinetoplast fluorescence.

ANCA classification

In the detection of anti-neutrophil granulocyte antibodies (ANCA) on ethanol and formalin-fixed granulocytes, the software distinguishes positive and negative reactions and identifies the main patterns of cytoplasmic and perinuclear ANCA as well as atypical ANCA. Analysis of a supplementary substrate of HEp-2 cells coated with granulocytes enables immediate differentiation between ANCA and ANA. Further, microdot substrates of the purified antigens myeloperoxidase (MPO) and proteinase 3 (PR3) can be analyzed in parallel for monospecific confirmation of the ANCA.

Transfected cells

Recombinant-cell IFA utilises transfected human cells expressing defined target antigens or relevant epitopes for monospecific detection of the corresponding antibodies. The technique is especially useful for detection of antibodies against fragile, conformation-dependent antigens. The EUROPattern Classifier supports evaluation of antigen-expressing cells for detection of diagnostically critical markers in autoimmune encephalitis, neuromyelitis optica spectrum disorders or primary membranous nephropathy.

Tissues

Computer-aided IFA evaluation using the EUROPattern Classifier has recently been extended to tissue substrates, which can be difficult to evaluate manually due to their complex structures. The Classifier supports evaluation of autoantibodies that occur in autoimmune hepatitis types 1 and 2 and primary biliary cholangitis on tissue sections of rat liver, kidney and stomach.

The software provides positive-negative classification of relevant ANA, anti-mitochondrial antibodies (AMA) and anti-smooth muscle antibodies (ASMA), as well as identification of an anti-liver kidney microsome (LKM) pattern. An additional substrate of cultured cells enables detection of antibodies against the smooth muscle antigen F-actin.

The EUROPattern Classifier also performs positive-negative discrimination of IgA antibodies against endomysium (EmA) on liver and oesophagus substrates, which serves as an important confirmatory test in coeliac disease diagnostics.

Optimized IFA workflows

The EUROPattern Classifier* is an integral part of EUROIMMUN’s solutions for IFA workflow optimisation. It is used in the EUROPattern series of automated microscopes, which combine rapid image acquisition with convenient on-screen viewing and computer-aided result evaluation. The EUROPattern Microscope* has a high processing capacity of up to 500 fields per run. The compact EUROPattern Microscope Live* combines live microscopy with ultrafast image processing in as little as two seconds per image and is ideal for small- to medium-throughput laboratories. A large range of computer-aided evaluation applications has also recently been incorporated into the UNIQO 160*, an all-in-one device that provides walkaway IFA automation from primary sample to result proposal.

The EUROPattern Classifier is connected to the EUROLabOffice 4.0 laboratory management software, which consolidates all results for the patient into one place, including the patient’s complete history with all test results and microscopy images. The management software communicates bidirectionally with the LIS, ensuring secure, traceable and rapid data exchange.

Further applications with the EUROPattern Classifier will be available soon.

* The availability of EUROPattern Classifier applications is dependent on the device. Regulatory status and product availability must be verified for the user‘s individual jurisdiction.

References

1. Gerlach S, Affeldt K, Pototzki L, Krause C, Voigt J, Fraune J, Fechner K. Automated evaluation of Crithidia luciliae based indirect immunofluorescence tests: A novel application of the EUROPattern-Suite technology. J Immunol Res Article 2015:742402 (2015).
2. Krause C, Ens K, Fechner K, Voigt J, Fraune J, Rohwäder E, Hahn M, Danckwardt M, Feirer C, Barth E, Martinetz T, Stoecker W. EUROPattern-Suite technology for computer-aided immunofluorescence microscopy in autoantibody diagnostics. Lupus 24: 516-529 (2015).
3. Fraune J, Gerlach S, Renztsch K, Teegen B, Lederer S, Affeldt K, Fechner K, Danckwardt M, Voigt J, Probst C, Komorowski L, Stöcker W. Multiparametric serological testing in autoimmune encephalitis using computer-aided immunofluorescence microscopy (CAIFM). Autoimmun Rev 15(10):937-942 (2016).
4. Hormann W, Hahn M, Gerlach S, Hochstrate N, Affeldt K, Giesen J, Fechner K, Damoiseaux J. Performance analysis of automated evaluation of Crithidia luciliae-based indirect immunofluorescence tests in a routine setting – strengths and weaknesses. Clin Chem Lab Med 56(1):86-93 (2017).
5. Van Beers JJBC, Hahn M, Fraune J, Mallet K, Krause C, Hormann W, Fechner K, Damoiseaux JGMC. Performance analysis of automated evaluation of antinuclear antibody indirect immunofluorescent tests in a routine setting. Auto Immun Highlights 9(1):8 (2018).

Want the latest science news straight to your inbox? Become a SelectScience member for free today>>