基于RepMLP的肺部CT图像分割

Lung CT image segmentation based on REPMLP

针对现存的临床辅助诊断的实时性需求,本文在传统PSPNet模型中引入了RepMLP模块来实现快速的肺部CT图像分割。该模型以PSPNet网络作为基础,使用加入了RepMLP模块的ResNet来替换传统的ResNet作为PSPNet的主干网络提取特征,并在RepMLP-ResNet中运用空洞卷积增加感受野。实验结果表明:在mIou相似的情况下加入RepMLP模块的PSPNet比传统PSPNet推理速度提升了约55%,较之于以MobileNetv2为特征提取网络的PSPNet虽然速度慢了约56%但mIou提升了约5.32%。该实验模型在保证较高的检测准确率的情况下达到了快速分割的目的,满足实时检测的要求。

In order to meet the real-time requirements of clinical auxiliary diagnosis,this paper introduced the REPMLP module into the traditional PSPNET model to achieve rapid lung CT image segmentation.The model is based on PSPNET network,and USES REPMLP module to replace the traditional REPNET as the backbone network of PSPNET to extract features,and uses void convolution in REPMLP-RESNET to increase the receptive field.The experimental results show that the PSPNET with the REPMLP module in similar MIOU conditions is about 55% faster than the traditional PSPNET reasoning speed.Compared with the PSPNET with the MobileNetV2 feature extraction network,although the PSPNET speed is about 56% slower,MIOU is about 5.32% faster.The experimental model achieves the purpose of rapid segmentation with high detection accuracy and meets the requirement of real-time detection.