Channel Fitting Network for Retinal Lesion Segmentation from OCT Images.

Journal: Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:

Abstract

Retinal lesion is a cause of age-related macular degeneration that poses a big threat to elderly population. The accurate detection and segmentation of retinal lesions benefits the early diagnosis of age-related macular degeneration and monitoring of disease progression. However, due to the large variant of imaging characteristics among different subtypes of diseases, it is challenging for a model to precisely segment retinal lesions. On the other hand, with the usage of deep neural networks (DNN), the accuracy of classifying subtypes of the retinal disease is high, which proves that the imaging features extracted from the classification models are more accurate and useful than the features obtained from segmentation models. Inspired by it, we propose a channel fitting module to enhance feature maps of segmentation model, with guidance of classification model. After the channel fitting, those feature maps that incorrectly emphasize irrelevant regions are filtered out. Therefore, our segmentations are more accurate to reflect true lesion regions. We conducted five-fold cross validation on 2633 retinal images scanned from 164 patients, proving the effectiveness and superiority of the proposed model.

Authors

  • Zhiyu Ning
    Sydney Polytechnic Institute, Sydney, Australia.
  • Yupeng Xu
  • Changyang Li
    School of Information Technologies, University of Sydney, Australia; Institute of Biomedical Engineering and Technology, University of Sydney, Australia. Electronic address: changyang.li@sydney.edu.au.
  • Yichao Hao
  • Zhiyuan Ning
  • Cong Liu
    Department of Bioengineering, University of Illinois at Chicago, 851 S Morgan St, Chicago, IL, 60607, USA.
  • Ke Yan
    Department of Biostatistics, Medical College of Wisconsin, Milwaukee, Wis.