Sternal Fracture Recognition Based on EfficientNetV2 Fusion Spatial and Channel Features

Abstract

Fractures are one of the most common injuries in medicine. A fracture is a partial or complete break in the continuity of bone structure. Fractures can occur for many reasons. For example, a certain part of the bone is fractured due to the direct action of violence on the part, resulting in a fracture of the injured part, which is often accompanied by varying degrees of soft tissue damage; fracture. The sternum is the most dense and complex part of human fractures, and it is very difficult to identify manually. The recognition of sternal fractures has great research significance in the field of fracture recognition, so this paper proposes an image recognition method based on a balanced improved convolutional neural network, and introduces the SA module combined with the SE module to design the model of the network MESCNet (fusion space and channel features), the network can help us identify sternal fractures very well. We used a dataset of 1227 sternum X-rays containing fractures from a dataset collected by the Radiology Department of Xi’an Red Cross Hospital. The images are not processed in any way and have different resolutions. We preprocessed the data and transformed it into a coloured database of sternal fractures. The recognition accuracy rate reached 78.12%. Therefore, it is proved that MESCNet can accurately identify sternum fractures and has broad application prospects.