Secondary Pulmonary Tuberculosis Lesions Detection Based on Improved YOLOv5 Networks.

Tuberculosis (TB) is an infectious disease caused mainly by Mycobacterium tuberculosis. It has been reported that the mortality rate of TB is extremely high if not timely detected and diagnosed in early stages. Currently, the diagnosis of secondary pulmonary tuberculosis (PTB) relies heavily on subjective analysis by specialized radiologists, which is time-consuming and inefficient. Thus, the application of neural networks augmented with intelligent algorithm rules holds significant value in detecting secondary PTB. This paper proposes an algorithm based on an attention mechanism and a loss function-improved YOLOv5 neural network to accurately detect secondary PTB lesions with four specific features. Furthermore, a multi-scale data augmentation method is proposed to expand the lesion dataset, enhancing the generalization ability and robustness of the trained model. Experimental results demonstrate that our proposed method can effectively improve the detection accuracy and speed of secondary PTB lesions, achieving accurate recognition of secondary PTB lesions with the four specific features.