Selected article for: "abnormal finding and local region"
Author: Xueyan Mei; Hao-Chih Lee; Kaiyue Diao; Mingqian Huang; Bin Lin; Chenyu Liu; Zongyu Xie; Yixuan Ma; Philip M. Robson; Michael Chung; Adam Bernheim; Venkatesh Mani; Claudia Calcagno; Kunwei Li; Shaolin Li; Hong Shan; Jian Lv; Tongtong Zhao; Junli Xia; Qihua Long; Sharon Steinberger; Adam Jacobi; Timothy Deyer; Marta Luksza; Fang Liu; Brent P. Little; Zahi A. Fayad; Yang Yang
Title: Artificial intelligence for rapid identification of the coronavirus disease 2019 (COVID-19)
  • Document date: 2020_4_17
    • ID: 79tozwzq_61
    Snippet: A max pooling layer that outputs log probability was used at the last layer, instead of the standard design that uses an average pooling layer at the last layer. The rationale of this design is that, given the abnormal finding is usually localized in a subregion of a CT image, we would like to predict whether a small region is abnormal due to SARS-CoV-2 28 . The CNN model can then be seen as a classifier that reports whether its receptive field i.....
    Document: A max pooling layer that outputs log probability was used at the last layer, instead of the standard design that uses an average pooling layer at the last layer. The rationale of this design is that, given the abnormal finding is usually localized in a subregion of a CT image, we would like to predict whether a small region is abnormal due to SARS-CoV-2 28 . The CNN model can then be seen as a classifier that reports whether its receptive field is SARS-CoV-2 (+). The label of an image is then predicted by combining all predictions of every local region over the whole image. Max pooling serves as an "OR" gate that labels an image as SARS-CoV-2 (+) if there is any subregion in it that is SARS-CoV-2 (+). Patient level prediction was set as the average of image-level prediction of a patient's 10 most abnormal images. To visualize the CNN's prediction, we up-sampled the CNN's outputs, without applying the max pooling layer, to the original image size. The lung mask was applied to up-sampled outputs for clear visualization.

    Search related documents:
    Co phrase search for related documents
    • abnormal image and CT image: 1, 2
    • average pooling layer and CNN model: 1
    • CNN model and CT image: 1, 2, 3, 4, 5, 6, 7, 8
    • CNN model and image level: 1, 2, 3, 4
    • CNN output and CT image: 1
    • CNN prediction and CT image: 1, 2
    • CT image and image label: 1
    • CT image and image level: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10