A Study of Hybrid Features and YOLO model in COVID’19

Abstract

Nowadays, Covid ’19 disease threatens people’s health and life. Accurate detection of Covid’19 is necessary to control its spread. In critical diagnosis, its better to analyze CT imaging than Lab test for fast recovery. Traditional deep learning method has the learning problem which affects the accuracy. In this paper, a new methodology based on YOLOv3 deep learning model proposes to detect Covid’19 in low dose computed tomography lung images. The feature vectors Haar like, HOG, LBP are added to calculate shape and texture of the object. The efficacy of hybrid computational YOLOv3, Haar-like, HOG and LBP features are compared. The sole usage of Haar-like, LBP is less effective, where fast access YOLOv3 features are most promising in multitask schemes. The automatic assessment between the low dose lung CT contents and the clinical semantic terms may help to retrieve reports and medical images from medical database for better diagnostic decision support system. This leads improvement of retrieval accuracy in medical databases.