The Combine of GLCM and Group, Focuses on the Grayscale of Medical Images

Medical image classification is very important in the diagnosis of hepatocellular carcinoma, which can provide more accurate computer-aided diagnosis, and accurate extraction of key semantic information from medical data is crucial to improve classification performance. However, many studies focus on the feature extraction module of natural image design, and do not carry out targeted design from the perspective of medical images, resulting in limited improvement in classification performance and mediocre performance. Therefore, we designed a plug and play attention module from the perspective of Grey-Level Cooccurrence Matrix and groups to improve the performance of backbone network for medical image classification. The results showed that the Sensitivity, Specificity, ACC and AUC of our method were increased by 0.0394, 0.0184, 5.26% and 0.0399, respectively, for ResNet18. The improvements of 0.0394, -0.0263, 1.75% and 0.0074 on MobileNetv3 indicate that the proposed method has advantages over other methods for medical image classification.