Selected article for: "lesion level and training set"
Author: Min Zhou; Yong Chen; Dexiang Wang; Yanping Xu; Weiwu Yao; Jingwen Huang; Xiaoyan Jin; Zilai Pan; Jingwen Tan; Lan Wang; Yihan Xia; Longkuan Zou; Xin Xu; Jingqi Wei; Mingxin Guan; Jianxing Feng; Huan Zhang; Jieming Qu
Title: Improved deep learning model for differentiating novel coronavirus pneumonia and influenza pneumonia
  • Document date: 2020_3_30
    • ID: ilc2bzkx_31
    Snippet: Detailed information was presented on the Result E2, Table E3 and Figure Figure E5 . The results showed that the detection performance was not sensitive to the confidence score as long as the cutoff for confidence score was in a reasonable range (Table E4 ). The detection model achieved F1 score 0.742 under confidence cutoff 0.1. We further used the annotated lesions to train and evaluate the model. Trinary scheme (with AUC 0.95) performed better.....
    Document: Detailed information was presented on the Result E2, Table E3 and Figure Figure E5 . The results showed that the detection performance was not sensitive to the confidence score as long as the cutoff for confidence score was in a reasonable range (Table E4 ). The detection model achieved F1 score 0.742 under confidence cutoff 0.1. We further used the annotated lesions to train and evaluate the model. Trinary scheme (with AUC 0.95) performed better than the Plain scheme (with AUC 0.93) ( Figure E6 ). More performance measures can be found in Table E5 . Two experienced specialists classified the lesions on which two schemes made very different predictions (with probability difference no less than 0.5) ( Figure E7 ). 366 (or 174) out of 540 NCP lesions were identified by Trinary (or Plain) scheme correctly. Detailed analysis showed that the Plain scheme tend to yield unreasonably high or low probability of lesion predictions depending on the lesions from centers in the training set or not. The results indicated the Trinary scheme was more consistent with specialists than Plain scheme on the lesion level classification.

    Search related documents:
    Co phrase search for related documents
    • AUC Plain scheme and Plain scheme: 1, 2, 3
    • AUC Plain scheme and training set: 1, 2
    • AUC Plain scheme and trinary scheme: 1, 2, 3
    • AUC trinary scheme and lesion level: 1
    • AUC trinary scheme and Plain scheme: 1, 2, 3
    • AUC trinary scheme and training set: 1, 2
    • AUC trinary scheme and trinary scheme: 1, 2, 3
    • confidence score and detection model: 1, 2, 3
    • confidence score and detection performance: 1
    • confidence score and F1 score: 1, 2, 3
    • confidence score and model evaluate: 1
    • confidence score and performance measure: 1
    • confidence score and probability difference: 1
    • confidence score and training set: 1