A nasal cytology dataset for object detection and deep learning

Nasal cytology is a new and efficient clinical technique for diagnosing rhinitis and allergies that is not widespread due to the time-consuming nature of cell counting. Thus, AI-aided counting could be a turning point for the diffusion of this technique. In this article we present the first dataset of rhino-cytological field images: the Nasal Mucosa Cell Dataset (NMCD), aimed to train and deploy Object Detection models to support physicians and biologists during clinical practice. The real distribution of the cytotypes that populate the nasal mucosa was replicated, by sampling images from slides of clinical patients and manually annotating each detected cell. The corresponding object detection task presents non-trivial issues associated with strong class imbalance, involving the rarest cell types. This work contributes to some of the open challenges by presenting a novel machine learning-based approach to aid in the automated detection and classification of nasal mucosa cells. The DETR, YOLO and Faster R-CNN models have shown good performance in both cell detection and cell classification, revealing great potential to accelerate the work of rhinologists experts. • First public dataset to kickstart research in the field of nasal cytology. • Includes 500 images with 10k+ bounding boxes across 10 cell classes. • Three deep-learning models evaluated as a baseline for nasal cell classification.