Abstract
We evaluated the reproducibility of computer-aided detections (CADs) with a convolutional neural network (CNN) on chest radiographs (CXRs) of abnormal pulmonary patterns in patients, acquired within a short-term interval. Anonymized CXRs (n = 9792) obtained from 2010 to 2016 and comprising five types of disease patterns, including the nodule (N), consolidation (C), interstitial opacity (IO), pleural effusion (PLE), and pneumothorax (PN), were included. The number of normal and abnormal CXRs was 6068 and 3724, respectively. The number of CXRs (region of interests, ROIs) of N, C, IO, PLE, and PN was 944 (1092), 550 (721), 280 (538), 1361 (1661), and 589 (622), respectively. CXRs were randomly allocated to training, tuning, and test sets in 70:10:20 ratios. Two thoracic radiologists labeled and delineated the ROIs of each disease pattern. The CAD system was developed using eDenseYOLO. For the reproducibility evaluation of developed CAD, paired CXRs of various diseases (N = 121, C = 28, IO = 12, PLE = 67, and PN = 20), acquired within a short-term interval from the test sets without any changes confirmed by thoracic radiologists, were used to evaluate CAD reproducibility. Percent positive agreement (PPAs) and Chamberlain’s percent positive agreement (CPPAs) were used to evaluate CAD reproducibility. The figure of merit (FOM) of five classes based on eDenseYOLO showed N-0.72 (0.68–0.75), C-0.41 (0.33–0.43), IO-0.97 (0.96–0.98), PLE-0.94 (0.92–95), and PN-0.87 (0.76–0.93). The PPAs of the five disease patterns including N, C, IO, PLE, and PN were 83.39%, 74.14%, 95.12%, 96.84%, and 84.58%, respectively, whereas the values of CPPAs were 71.70%, 59.13%, 91.16%, 93.91%, and 74.17%, respectively. The reproducibility of abnormal pulmonary patterns from CXRs, based on deep learning-based CAD, showed different results; this is important for assessing the reproducible performance of CAD in clinical settings.
Highlights
Abbreviations AUC Areas under the curve CAD Computer-aided detection convolutional neural network (CNN) Convolutional neural network chamberlain’s percent positive agreement (CPPA) Chamberlain’s percent positive agreement CXR Chest radiograph free-response receiver operating characteristic (FROC) Free-response ROC curve picture archiving and communication system (PACS) Picture archiving and communication system PPA Percent positive agreement ROI Region of interest you only look once (YOLO) You only look once
All CXRs scanned by the computed radiography (CR) or digital radiography (DR) system were downloaded from PACS at ASAN Medical Center
These cut-off thresholds for reproducibility were determined empirically as the number of average false positives 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6 in the FROC curve of the validation set in Figs. 1 and 2
Summary
Abbreviations AUC Areas under the curve CAD Computer-aided detection CNN Convolutional neural network CPPA Chamberlain’s percent positive agreement CXR Chest radiograph FROC Free-response ROC curve PACS Picture archiving and communication system PPA Percent positive agreement ROI Region of interest YOLO You only look once. The performance of CAD has improved significantly, better sensitivity and low false-positive rates are required for its integration into clinical use. Another important aspect of concern for using CAD on CXRs is its reproducibility. Islam[11] reported on the diagnosis of pulmonary abnormalities on CXRs and found that the ensemble method with deep learning provided the highest accuracy for detecting abnormalities These previous studies did not address the reproducibility of CAD in CXRs of same patients within a short-term interval; they reported on the changes in C XRs15. We evaluated the reproducibility of CAD of multiple lesions on CXRs using paired images acquired within a short-term interval from the test sets and in those where no changes were reported by expert thoracic radiologists of our institution
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have