Follow-up of airway inflammation of asthmatic patients using an Artificial Intelligence driven sputum cell count and classification

<bold>Background:</bold> Asthma is associated with heterogeneous airway inflammation, which can be reliably assessed by inflammatory cell profiles in spontaneous or induced sputum. However, that technique is time-consuming and relies on a subjective interpretation by the lab technician. Therefore, implementation of induced sputum cell count as a routine measure of airway inflammation in clinical care is restricted to highly specialized centers. <bold>Objective:</bold> This project aims to automatize sputum inflammatory cell counting using artificial intelligence (AI) and enrich it by a morphological analysis of the inflammatory cells and an immunocytochemistry detection of key inflammatory biomarkers. <bold>Methods:</bold> 125 sputum slides from University Asthma Clinic of Liege, have been digitalized and uploaded on Cytomine, an artificial intelligence (AI) developing web platform. A counting module ontology for characterization of cellular and non-cellular objects has been defined and all objects within a region of interest (ROI) of the slide have been annotated. <bold>Results:</bold> Of the 125 digitalized slides, 53 were of sufficient quality for analysis. A total of 46003 objects were annotated, of which 22194 immune cells, 496 free eosinophil granules and 45 Charcot-Leyden crystals. Meanwhile, an immunocytochemistry protocol using alkaline phosphatase has been developed to detect IL-5 expression by inflammatory cells. <bold>Conclusion:</bold> Contrast quality of the scans needs to be optimized to increase the dataset to train the AI algorithm. Once developed, this AI tool would enable more accurate and faster diagnosis of asthmatic patients and better prediction of response to expensive therapies.