Abstract
To diagnose COVID and Non-COVID Pneumonia in CT scans and X-RAY images researchers have developed intelligent and robust classifiers owing to disease recognition using artificial intelligence-based solutions. However, recognition of the disease itself does not suffice the purpose. In order to develop robust systems, there exists a dire need for COVID and Non-COVID Pneumonia detectors that can perform both classification and detection by drawing bounding boxes in the infected images. To develop a lightweight and robust COVID and Non-COVID Pneumonia detector, in this work we have proposed the RYOLO v4-tiny detector which is developed by integrating the residual network in the existing YOLO v4-tiny feature extraction network. In order to train and test the detector, we have created a richly annotated dataset consisting of images for COVID and Non-COVID Pneumonia in CT scans and X-RAY images. On the created dataset, the proposed RYOLO v4-tiny detector achieved a mAP (Mean Average Precision) value of 88.18 % which was 7.91 % higher as compared to the baseline YOLO v4-tiny detector. Furthermore, the proposed detector achieved 10 % higher precision, 14 % higher recall, and 11 % higher F1 Score as compared to the YOLO v4-tiny detector. The proposed RYOLO v4-tiny detector is a useful tool for medical practitioners for autonomous diagnosis and detection of COVID and Non-COVID Pneumonia in CT scans and X-RAY images.
【저자키워드】 deep learning, Radiography, COVID pneumonia, YOLO v4-tiny, 【초록키워드】 Diagnosis, COVID, CT scan, dataset, YOLO, recall, disease, score, diagnose, Precision, Classifier, researcher, Practitioner, robust, develop, the disease, autonomous, baseline, bounding boxe, 【제목키워드】 COVID, CT scan,