STBi-YOLO: A Real-Time Object Detection Method for Lung Nodule Recognition

Abstract

Lung cancer is the most prevalent and deadly oncological disease in the world, but a timely detection of lung nodules can greatly improve the survival rate of this disease. However, due to the tiny size of lung nodules and inconspicuous edges, lung nodules are not easily distinguished by naked eyes thus medical image diagnosticians are prone to misdiagnosis simply based on their own experiences and subjective judgements. In recent years, the machine-learning-based image processing techniques find their wide applications in the field of medical diagnosis, and have been proved to be an efficient way to aid diagnosticians to accurately identify subtle lesions in images. To accurately recognize lung nodules in CT images, in this paper, we propose an approach, called STBi-YOLO. This approach stems from YOLO-v5, but makes significant improvements from three dimensions—we first use the spatial pyramid pooling network that is based on stochastic-pooling method to modify the basic network structure of YOLO-v5; then apply a bidirectional feature pyramid network to perform multi-scale feature fusion; finally improve the loss function of the YOLO-v5 and adopt the EIoU function to optimize the training model. To evaluate our approach, we compare STBi-YOLO with YOLO-v3, YOLO-v4, YOLO-v5, and multiple leading object detection models, such as Faster R-CNN and SSD. The experiments show that STBi-YOLO achieves an accuracy of 96.1% and a recall rate of 93.3% for the detection of lung nodules, while producing a $4\times $ smaller model size in memory consumption than YOLO-v5 and exhibiting comparable results in terms of mAP and time cost against Faster R-CNN and SSD.