Small Cell Lung Cancer Semantic Segmentation in Pathological Section Images Based on Multi-Scale Fusion Network

EasyChair Preprint 5897

Abstract

Lung cancer is a serious threat to human health, accounting for 26% of the total number of cancer deaths. In order to develop treatment plans for patients with small cell lung cancer, doctors are required to evaluate their pathological materials, which requires a high level of experience and knowledge and is time-consuming and labor. In this paper, we propose an automatic segmentation model of small cell lung cancer tumor from pathological section images. A multi-scale convolutional network is trained on pathological images using ground truth ROIs that were manually delineated by pathologists for 31 patients. We use high-resolution to low-resolution parallel convolution to extract multi-scale features in the first three down-sampling stages of the network's encoding path. Use mIOU(mean intersection over union) as the evaluation index to compare the training results of this network with U-Net, and the results showed that the segmentation results of this network were significantly improved compared with U-Net. Code and models are available at https://github.com/QinchengZhang/PathologySegmentation.

Keyphrases: Convolutional Neural Networks, Medical image segmentation, multiscale

@booklet{EasyChair:5897,
  author    = {Qincheng Zhang and Junhai Xu and Ke Zheng and Zhiwen Zhang and Jiangyong Yu},
  title     = {Small Cell Lung Cancer Semantic Segmentation in Pathological Section Images Based on Multi-Scale Fusion Network},
  howpublished = {EasyChair Preprint 5897},
  year      = {EasyChair, 2021}}