Lesion Segmentation Framework Based on Convolutional Neural Networks with Dual Attention Mechanism

Fei Xie,Tao Jiang,Panpan Zhang,Wei Zhao,Gang Gao,Xuemin Shen,Ziyu Guan,Jiao She,Pengfei Xu

doi:10.3390/electronics10243103

Abstract

Computational intelligence has been widely used in medical information processing. The deep learning methods, especially, have many successful applications in medical image analysis. In this paper, we proposed an end-to-end medical lesion segmentation framework based on convolutional neural networks with a dual attention mechanism, which integrates both fully and weakly supervised segmentation. The weakly supervised segmentation module achieves accurate lesion segmentation by using bounding-box labels of lesion areas, which solves the problem of the high cost of pixel-level labels with lesions in the medical images. In addition, a dual attention mechanism is introduced to enhance the network’s ability for visual feature learning. The dual attention mechanism (channel and spatial attention) can help the network pay attention to feature extraction from important regions. Compared with the current mainstream method of weakly supervised segmentation using pseudo labels, it can greatly reduce the gaps between ground-truth labels and pseudo labels. The final experimental results show that our proposed framework achieved more competitive performances on oral lesion dataset, and our framework further extended to dermatological lesion segmentation.

Highlights

With the rapid development of computer vision, especially the significant improvement of the representation ability of convolutional neural networks [1,2], image segmentation has achieved good performances and laid a solid foundation for the application of medical image segmentation
In the fully supervised segmentation experiment, the ground-truth labels are the binary masks of the original dataset
In the weakly supervised segmentation experiment, the ground-truth labels are the circumscribed rectangles of the binary masks

Summary

Introduction

With the rapid development of computer vision, especially the significant improvement of the representation ability of convolutional neural networks [1,2], image segmentation has achieved good performances and laid a solid foundation for the application of medical image segmentation. Medical images segmentation as an important and difficult task of computer-aided diagnosis, is the key to further obtain diagnostic information. Traditional object location in medical images requires professional doctors to manually identify, which is time-consuming and labor-intensive and vulnerable to subjective factors. While the lesion segmentation results obtained by deep learning methods are becoming a promising method. Compared with ordinary images, clinical diagnosis invokes higher requirements for the accuracy of the segmentation results of medical images. The high variability, the complex morphological structure, 4.0/)

Objectives

Methods

Results

Conclusion