Convolutional Neural Networks Applied to Lung Image Detection: Model Comparison and Optimal Selection

Abstract

The topic of pulmonary health has become a prevalent concern among the public, making monitoring lung health a crucial subject. Numerous researchers have proposed models or methods for detecting pulmonary images, and their effectiveness has been confirmed. However, in the domain of pulmonary image detection, there is a systematic comparison to identify the most suitable and efficient method. Therefore, the theme of this study is to discover, through comparative analysis, the model method most fitting for current pulmonary image detection. The research methodology is outlined as follows: firstly, introduce popular model methods in the current context; secondly, summarize and record the accuracy of each model in pulmonary image recognition; and finally, conduct a horizontal comparison to derive conclusions regarding the optimal model method. This study reveals that Feature Fusion exhibits an outstanding accuracy of 97.79% in pulmonary image recognition, securing the top position. In contrast, Transformer has the lowest accuracy, yet it still reaches 78.9%. Therefore, Feature Fusion emerges as the most promising model method currently deserving widespread promotion in the field of pulmonary image detection.