Multi-scale high and low feature fusion attention network for intestinal image classification

• Published in: Signal, image and video processing Vol. 17; no. 6; pp. 2877 - 2886
• Main Authors: Li, Sheng, Zhu, Beibei, Guo, Xinran, Ye, Shufang, Ye, Jietong, Zhuang, Yongwei, He, Xiongxiong
• Format: Journal Article
• Language: English
• Published: London Springer London 01.09.2023; Springer Nature B.V
• Subjects: Classification; Computer Imaging; Computer networks; Computer Science; Datasets; Feature extraction; Image classification; Image Processing and Computer Vision; Machine learning; Medical imaging; Modules; Multimedia Information Systems; Original Paper; Pattern Recognition and Graphics; Signal,Image and Speech Processing; Vision; Deep learning; Feature fusion; Intestinal image classification; Convolutional neural network
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-023-02507-0

Accurate abnormality classification in intestinal images is critical for the diagnosis and treatment of early-stage intestinal cancers, but remains challenging due to the large intra-lesion variability and high similarity between abnormal lesions and normal tissue. To solve the above problems, a multi-scale high and low feature fusion attention network is proposed to efficiently utilize the features extracted by the backbone network. First, a multi-scale feature extraction module is used to extract features, and then a detail capture attention module and a dense sampling fusion module are used to focus on and fuse the lesion information hidden in the superficial and deep layers. Finally, classification effect is obtained by fusing the lesion information. Experiments show that the method achieves 98% classification accuracy on the Kvasir dataset with three types of normal, polyp and ulcer. In addition, the classification accuracy in the private dataset reaches 97.52% and the experimental effect is better than other existing deep learning classification methods.