Medical Image Recognition Using a Novel Neural Network Construction Controlled by a Kernel Method Encoder

The recognition and diagnosis of medical images is one of the important topics in artificial intelligence, such as colonic polyp and cataract classification. The appearance, shape, and size of pathological features in medical images significantly influence the accuracy of identification. Improving recognition accuracy is crucial due to the difficulty in distinguishing pathological features accurately. This paper presents a medical image recognition method utilizing a neural network controlled by a kernel method encoder, which presents the originality of this research. Medical images are enhanced to extract various features. Using the kernel method encoder, a high-dimensional distinguishable feature map is generated. This mitigates the problem of insufficient features in a single image. Image classification is achieved by optimizing kernel method encoder parameters and training the enhanced neural network classification model. We conducted experiments on colonography computed tomography images and colour fundus images. Initially, we conducted classification experiments on the colonography computed tomography image dataset. The experimental results demonstrate that the proposed model improves classification accuracy and shows good performance. Subsequently, we conducted the same classification experiment on colour fundus images using the proposed method. The experimental results indicate that this method improves the classification accuracy. The accuracy of this method on two medical image datasets is 92.1% and 97.4%, respectively. The proposed method performs well across both medical image datasets. This further demonstrates the robustness of the multi-feature and enhanced neural network model classification method.