Skip to main content

Table 1 Segmentation methods in chest X-ray. The datasets, methods, assessment measures, and segmentation results are provided in each column, respectively

From: Computer-aided detection in chest radiography based on artificial intelligence: a survey

  Study Datasets Assessment measures Results
Image progressing based methods Cheng et al. [34] Custom Accuracy  
Armato et al. [35] Custom (600) Subjectively assessed the accuracy and completeness of the contour Up to 79.1% (score 4 or 5) and 8.1% inaccurate (score 1 or 2)
Li et al. [36] Custom (40) Accuracy, sensitivity, specificity Left lung: 95.2% accuracy, 91% sensitivity, 96.5% specificity; right lung: 96% accuracy, 91.1% sensitivity, 97.2% specificity
Iakovidis et al. [37] Custom (24) Accuracy, sensitivity, and specificity 95.3% sensitivity, 94.3% specificity
Wan et al. [38] JSRT Custom (154) Accuracy, overlap scores, precision, sensitivity, specificity, and F score Accuracy, F value, accuracy, sensitivity, and specificity were higher than 90%; the JSRT dataset overlap score was 87%; the overlap rate of the custom datasets (standard machines) was 81% and (mobile machines) is 69%
Van Ginneken et al. [42] Custom (230) Overlap scores Left lung: 0.887 ± 0.114; right lung: 0.929 ± 0.026
Machine learning based methods Mcnittgray et al. [45] Custom (33) Accuracy NN: 76%; LDA: 70%; KNN: 70%
Vittitoe et al. [46] Custom (198) Sensitivity, specificity, and accuracy Sensitivity: 0.907 ± 0.044; specificity: 0.972 ± 0.022; accuracy: 0.948 ± 0.016
Shi et al. [47] JSRT (52) Accuracy 0.978 ± 0.0213
Novikov et al. [51] JSRT Dice coefficient, jaccard coefficient Lung: 97.4%, 95%; collarbone: 92.9%, 86.8%; heart: 93.7%, 88.2%
Dai et al. [52] JSRT, MC Intersection-over-union Both lungs: 94:7% ± 0:4%, heart: 86:6% ± 1:2%
  1. Accuracy: (TP + TN)/(TP + TN + FP + FN); sensitivity: R = TP/(TP + FN); specificity: TN/(TN + FP); overlap scores: TP/(TP + FP + FN); precision (or positive predictive value): P = TP/(TP + FP); F score: 2 × P × R/(P + R); intersection-over-union: IoU = TP/(TP + FP + FN); negative precision: TN/(TN + FN); false accept rate: FAR = FP/(FP + TN); false rejection rate: FRR = FN/(TP + FN); where TP, TN, FP, and FN represent true positive, true negative, false positive, and false negative, respectively
  2. Dice coefficient: DSC = 2 × (|S∩GT|)/(|S| + |GT|, jaccard coefficient of concordance: JS = (|S∩GT|)/(|SGT|), where S is segmentation result, GT is the ground truth