Influence of different image preprocessing methods on bone age prediction

In medical image recognition represented by bone age prediction, image samples need to be preprocessed to improve the quality of image samples and improve the learning efficiency of deep learning. This paper aims to compare the effects of different image preprocessing methods on the performance of the neural network. In this paper, the method of control experiment is used. Without pretreatment, the structure and framework of the neural network are controlled to remain unchanged, to make the conclusion more objective. This paper mainly discusses three pretreatment methods. 1 Conventional image filtering; 2. Use u-net network specially used for biomedical image segmentation to segment hand bones in X-ray; 3. The control group did not undergo image preprocessing. At the same time, this paper proposes to mark the gender of the owner of hand bone X-ray film in the form of a white background mark on the original image and control the gender weight by adjusting the size of the mark. U-net network preprocessing does not significantly improve the accuracy of the neural network, but this method makes the effect of deep neural network and shallow neural network almost the same, so it can be used as an effective method to prevent overfitting of neural networks. The main innovation of this paper is to explore the effectiveness of preprocessing algorithms in preventing the overfitting of medical image models by comparing the bone age prediction under various preprocessing methods.