基于CBAM Faster R-CNN的食道癌检测

Detection of esophageal cancer based on CBAM Faster R-CNN

为解决钡餐造影图像经Faster R-CNN的骨干网络得出的特征图中的食管区域不明显以及背景区域占比较大等问题,提出了CBAM Faster R-CNN,它是在原Faster R-CNN模型的基础上添加卷积注意力模块CBAM,以提升特征图中食管区域的特征的显著性.采用CBAM Faster R-CNN模型对数据增强后的训练集进行了训练,并采用Recall值、Precision值、AP值进行了评价分析.实验结果表明:在不同的IOU阈值下,CBAM Faster R-CNN模型的检测效果明显优于Faster R-CNN,且对实际的临床研究具有一定的意义.

To solve the problems that the esophagus area in the feature map obtained by the Faster R-CNN backbone network of the barium meal angiography is not obvious and the background area is relatively large,CBAM Faster R-CNN is proposed.It is based on the original Faster R-CNN model by adding a convolutional attention module CBAM to improve the feature saliency of the esophagus region in the feature map.The CBAM Faster R-CNN model is used to train the training set after data augmentation,and the Recall value,Precision value,and AP value are used for evaluation and analysis.The experimental results show that under different IOU thresholds,the detection effect of the CBAM Faster R-CNN model is significantly better than the Faster R-CNN,and it has certain significance for practical clinical research.