BEATing CILIAted cells

Nasal cytology is a medicine field that focuses on the examination of nasal mucosa cells with the objective of recognizing changes in the epithelium, which is frequently subjected to acute or chronic irritation and inflammation caused by viruses, bacteria, or fungi; in the last decade nasal cytology is becoming increasingly critical in diagnosing nasal conditions. Among the cell population of the nasal mucosa, ciliated cells make up 80% of the epithelium in the upper airways. These cells are characterized by the presence of cilia that normally beat in a consistent, fluid pattern, easing effective mucociliary clearance. This cytotype results quite important since its study allows for the diagnosis of primary ciliary dyskinesia (PCD), a rare pathology that is usually associated with other severe pathologies such as situs inversus, cardiac diseases and male infertility. The assessment of ciliary function through the measurement of the ciliary beating frequency (CBF).In addition, the CBF decay of the ciliated cells in a dead body may help to trace the time of death. These motivations have fueled the research seeking an architecture that supports such long, complex estimations, and feels necessary in medical andforensics contexts, that’s why we developed Deep Cilia, a completely autonomous CBFdecay estimation system. We also collected a new dataset containing videos of beating ciliated cells, made available to the Scientific Community on IEEE Dataport. The videos were acquired by an expert cytologist using a microscope (with 1000x magnification) equipped with a digital camera. The dataset contains 119 videos of beating ciliated cells and each video exhibits a variable number of ciliated cells, varying from 1 to more than 5 (in contexts where many cells are amassed). The average video duration in the dataset is 93 seconds, with an upper bound of 352 seconds and the shortest being only 3 seconds long. As underlined, the cardinality of ciliated cells recorded in each video can vary. From the 119 videos a total of 246 clearly visible strias were reported, labeled and manually extracted by the authors on each frame of the videos. In addition to the number of cells and video duration, the videos differ from each other in terms of frame rate with an average of 23.63 Fps and an standard deviation of ± 2.54 Fps and in terms of resolution. The video frames are also already extracted, augmented and splitted in three partitions to train, validate and test eventual object detection systems, alongside the bounding boxes for the cells manually extracted by the authors on each frame.