Table 4.
Overview of published works regarding conventional methodologies for the segmentation of lung CT images (2014–2021).
| Authors | Year | Dataset | Methods | Performance Results (%) |
|---|---|---|---|---|
| Lai and Wei [148] | 2014 | Private (10 patients) | Filtering process + morphological operations (threshold, region filling, closing) | TPR = 97.0 TNR = 99.0 AAE = 1.58 |
| Li et al. [147] | 2015 | Private (15 patients) | Edge-based recursive geometric active contour (GAC) model | OV = 98.0 |
| Shi et al. [149] | 2016 | Private (23 patients) | Histogram thresholding + region growing and random walk | OR = 1.87 UR = 2.36 ABD = 0.620 mm |
| Zhang et al. [150] | 2017 | LIDC-IDRI | Region- and edge-based GAC (REGAC) method | DSC = 97.7 HD-95 = 2.50 mm |
| Rebouças Filho et al. [151] | 2017 | Private (40 patients) | 3D ACACM | F-score = 99.2 (ACACM), 97.6 (RG), 97.4 (OsiriX), 97.2 (LSCPM) |
| Oliveira et al. [153] | 2018 | VISCERAL Anatomy3 | Multi-atlas alignment + label fusion (voting and statistical selection) | DSC = 97.4 (LL), 97.9 (RL) HD-95 = 4.65 mm (LL), 2.81 mm (RL) |
| Chen et al. [152] | 2021 | LOLA11 Private (65 patients) | Random walker | (Private) DSC = 98.6 (LL), 98.5 (RL) (LOLA11) DSC = 97.4 |
AAE: average area error; ABD: absolute border distance; ACM: active contour method; DSC: Sørensen–Dice coefficient; LL: left lung; LSCPM: level-set based on coherent propagation method; HD: Hausdorff distance; OR: over-segmentation rate; OV: overlap volume; RG: region growing; RL: right lung; TPR: true positive rate; TNR: true negative rate; UR: under-segmentation rate.