A coarse-to-fine goblet cell recognition algorithm for whole slide imaging of the lung tissue in a rat model of asthma

• We provide a novel approach that combines coarse segmentation and fine recognition to identify goblet cell. • The proposed method shows a precision of 84.9 ± 4.2 % and the F1-score of 82.8 ± 4.0 % for identifying goblet cells. • This work can contribute to determining the severity of asthma disease from histopathological slides. The number of goblet cells is an important biomarker for the diagnosis and prognosis of asthma disease. However, visual assessment of a plethora of goblet cells is laborious and time-consuming, making it difficult to precisely identify and count the number of the cells in pathohistological slides. We developed an automated method for identifying goblet cells. It was a combination of coarse and refined recognition, including extraction of the cell region and separation of the adjacent cells. The goblet cell region was extracted with the superpixel method. Initial segmentation results were obtained by merging over-segmented superpixels based on their average intensities in the CIE L*a*b* color space. After that, in refining recognition process, the iCut algorithm was applied to segment touching cells and identify individual goblet cells. The proposed method was evaluated on the images of the lung tissue in a rat model of asthma. With a precision of 84.9 ± 4.2 % and the F1-score of 82.8 ± 4.0 %, the proposed method surpassed the other four conventional cell recognition algorithms on the both metrics with relatively higher computational cost. This work demonstrated precise identification of goblet cells based on histopathological slides, which is crucial for evaluating the extent of asthma, enabling investigation into the disease’s origin and the efficacy of related medications. It can also be flexibly extended to cell segmentation in other biomedical images.