Automatic Semi-quantitative Histological Assessment of Tissue Traits Using a SmartWeb Application

Asmart web application suitable for classifying goblet cell hyperplasia and level of mucus production in stained lung tissues from mice with experimentally induced allergic asthma. Multiple trainer-model approaches are investigated and proposed in this manuscript, based on machine learning techniques, which provide a technological evolution in the analysis of traits of biomedical imaging. Several schemes, which consist of pre-trained image classifiers on ImageNet, are analyzed and compared each other. Lung tissue images of mice with allergic asthma, depicting mucus-containing periodic acid-Schiff (PAS) positive bronchial cells, are fed as input datasets. The performance of each model is evaluated, based on a variety of metrics: accuracy, recall, precision, cross entropy, f1-score, confusion matrix. Such a web tool could contribute to biomedical research by providing an automated standardized way to determine phenotypic severity of histological traits based on a semi-quantitative scoring scale.