YOLO Based Segmentation and CNN Based Classification Framework for Epithelial and Pus Cell Detection.

Identifying the cells, such as pus and epithelial cells, from microscopic images is one of the important steps in medical diagnostics. Microscopic examination by hand is labor-intensive and unreliable. Therefore, it is helpful to have an automated approach for classifying these cells to enable quick and accurate diagnosis. Creating a model for automated cell identification is challenging because of the numerous variable parameters such as various stains and magnifications and cell overlapping. This paper offers a robust object detection model that detects the pus and epithelial cells images obtained from the microscopic analysis of direct samples of Gram-stained patient samples such as pus and sputum. This paper also presents a novel classifier that addresses the overlapping issues present in the cells. The proposed methodology offers an mAP of 0.87 and a classification accuracy of 94.5%.