AB0015 ANTINUCLEAR ANTIBODIES PATTERN CLASSIFIER WITH CONVOLUTIONAL NEURAL NETWORKS

Background To detect antinuclear antibodies (ANA) on the blood’s patients is a mostly used technique to diagnose immunologic diseases. The Indirect Immunofluorescence method (IIF) generates images with ANA patterns them and commonly by immunology laboratories around world. Those are documented in International Consensus of Patterns (ICAP) there 29 at present, split into three main groups: Nuclear, Cytoplasmic Mitotic. Classifying these subjective task that has strong dependence physician’s experience training, thus professionals need second reader reach successful classification patterns. support professionals, we developed Machine Learning model (ML), based Convolutional Neural Networks (CNN) able classify between positive negative then classifying only samples groups mentione. Objectives Developing machine learning help physicians pattern without train entry-level different themselves. Methods Using public database 2079 defined limitations our model; preprocessed dataset made data augmentation process avoid overfitting issues. objectives define two parts, first one was CNN each sample negative, another from last step before mentioned. Following this idea, proved 17 pre-trained compared their results metrics learning: accuracy, precision, recall F1. prove final it mentioned augmented test private 445 Inmuno21 Laboratory Caracas, Venezuela. Results Finally, reached acc: 93%, pre: recall: 93% F1: positives negatives, improving state art models presented as solution for problem1. On other hand, shows an 75%, 85%, 92% 87% stage, Mitotic Cytoplasmic, becoming be multilabel capability positives-negatives previous discrimination. Conclusion It’s necessary make efforts create bigger datasets images, better more label, since major problem found when we’ve project poor had not were available, lower amount than ones or no groups. Thus augment front Despite this, high performance works, can training partner new professionals. Reference [1]Cascio, D.; Taormina, V.; Raso, G. Deep Network HEp-2 Fluorescence Intensity Classification. Appl. Sci. 2019 , 9 408. https://doi.org/10.3390/app9030408 Acknowledgements: NIL. Disclosure Interests None Declared.