Table 1.
Summary of various state-of-the-art techniques employed for plaque characterization using coronary artery scans.
| Dataset | Preprocessing/ROI Segmentation | Feature Extraction | Feature Reduction/Feature Selection/ Feature Ranking/Organization |
Detection | Classification | Task | Outcomes * | |
|---|---|---|---|---|---|---|---|---|
| [11] | 599 VH-IVUS images of 10 patients | Thresholding + HFCM-kNN model | CLBT + OLBT | SVM with radial basis function (RBF) | Multiclass (PIT, TCFA and CaTCFA) and binary (TCFA and non-TCFA) | For binary: Pqacc.: 81.03 Pqsen.: 84.81 Pqspec.: 84.81 Precision: 84.81 For multiclass Pqavg.acc.: 98.42 + 0.01 Kappa: 0.9198 |
||
| [15] | IVUS images | Neuro Fuzzy | Atheromatous plaque (fibrotic, lipidic, calcified, and normal) | Pqavg.acc.: 98.9 | ||||
| [21] | 300 IVUS frames of 10 patients | Deformable models + Estimation borders by experts | Co-occurrence matrix + LBP+ Mean value + Entropy + Geometrical features | t-test | RF | Multiclass (DC, NC, FT, and FFT) | Pqacc.: 85.65 | |
| [22] | 553 IVUS frames of eight patients | ROI Extraction + Otsu’s automatic thresholding + Pathological tissue detection | CNN | Multiclass (DC, NC, FT, FFT, Media) | Overall accuracy: 93.5Pqacc.:DC: 98.5NC: 88.6FT: 91.1FFT: 90.0Media: 99.4 | |||
| [25] | IVUS images from 11 patients | Manual segmentation by expert | LBP + FOS +GLCM + LEM + Extended GLRLM + Intensity |
PCA | RF | Multiclass(DC, NC, FT, and FFT) | AUC: 0.845, 0.704, 0.783 Pqacc.: 85.1,71.9,77.2 Pqsen.: 82,81.2, 80.6 Pqspec.: 87.1, 59.6, 75.9 (Respectively for Net1: FT/FFT or NC/DC Net2: FT or FFT Net3: NC or DC) |
|
| [27] | 1000 IVOCT images from 47 patients | Anisotropic diffusion + Polar Transformation+ Hough Transform | Intensity + HOG + LBP + FV + k-means clustering | SVM | Multiclass (normal, fibrous plaque, fibro-atheroma, plaque rupture, fibro-calcific plaque) |
Pqavg.acc.: 90 With standard deviation of 0.02 |
||
| [67] | 770 OCT images of 5 patients | ROI Extraction | LBP+GLCM | CNN (U Net) | Multiclass (lipid tissue, fibrous tissue, background) | Pqacc.: 95.8 | ||
| [70] | 435 IVUS images | Polar Transformation + Gaussian filter +Median filter+ Anisotropic Diffusion filter | Haralick’s +Laws’ textural feature | SVM | Two class (fibrotic plaque and normal) | AUC: 0.97 Jaccard Index: 0.85 |
||
| [72] | 6556 OCT images from 49 patients | ROI+ Dynamic programming + Gaussian filter | CNN + Morphological features | Wilcoxon signed rank test | RF | Binary class: (fibro-lipidic and fibrocalcific plaque) | Fibro-lipidic plaque: Pqsen.: 84.8 Pqspec.: 97.8 Fibro-calcific plaque: Pqsen.: 91.4 Pqspec.: 95.7 |
|
| [73] | 18 CTA images Total: 1786 cross sections with Non calcified plaque (NCP): 729, Calcified plaque (CP): 511, Mixed plaques: 546. |
DRLSE | RRS feature vector | SVM (Gaussian kernel) | Multiclass(calcified, non calcified and mixed plaques) | Average precision: 92.6±1.9 Average recall: 94.3±2.1 |
||
| [80] | 60 IVUS images of 7 patients | Anisotropic diffusion filter + Thresholding | Deformable models |
Bayesian | Two class (calcified and non-calcified plaque) | AUC: 0.943 Pqspec.: 98.5 Pqsen.: 92.67 |
||
| [83] | 27 OCT pullbacks of 22 patients | Gaussian filter + Thresholding + k-means | LBP + GLCM | WRP | RF | Multiclass (calcium, lipid pool, fibrous tissue, and mixed Tissue) | Pearson’s correlation coefficient: 0.97 (FT) | |
| [84] | IVOCT images of 11 patients | Gaussian filter + Otsu threshold filtering | Attenuation coefficient + GLCM | SVM (RBF) | Multiclass (fibrous, calcification and lipid tissue) | Pqacc.: 83 | ||
| [88] | IVUS images of 7 patients | Multilevel discrete wavelet frame decomposition | FOS + GLCM + LBP + RL + Wavelet Intensity values | FuzCoC | SVM (RBF) | Multiclass (calcium, necrotic core, fibrous, and fibro-fatty) | Pqavg.acc.: 81 | |
| [91] | In-vivo dataset: VH-IVUS 2263 images of 10 patients Ex-vivo dataset:64 images |
Shadow detection using threshold | NGL + LBP + MRL | SVM and ECOC | Multiclass(calcium, necrotic core, and fibro fatty) | Kappa values: 0.639 (in-vivo) and 0.628 (ex-vivo) | ||
| [92] | 50 OCT images from 3 patients | Co-occurrence matrix + LBP+ Entropy + Mean value | RF | Multiclass (calcium, lipid pool, fibrous tissue, and mixed plaque) | Pqacc.: 80.41 | |||
| [93] | 300 IVUS images of 7 patients | Multilevel discrete wavelet frames decomposition + SOFM | FOS + GLCM + RL + LBP + wavelets + LISA | FaIRLiC | Multiclass (DC, NC, FT, and FFT) | Testing accuracy: 76.16% | ||
| [94] | IVUS images of 7 patients | Border detection + 2-D Kohonen’s self-organizing feature map (SOFM) | FOS + GLCM + WF + RL + LBP | FaIRLiC | Multiclass(calcium, necrotic core, fibrous and fibro lipid) | Average classification Accuracy on each frame: 73.67 | ||
| [95] | 2646 Coronary Tomography Angiography (CTA) images of 73 patients (CP: 28, NCP: 15, Normal: 30) |
Adaptive Histogram Equalization | Gabor Transform + Entropy | ANOVA | SVM (RBF and polynomial kernel) | Multiclass (normal, non calcified and calcified) | Pqacc.: 89.09 PqPPV: 91.70 Pqsen.: 91.83 Pqspec.: 83.70 |
|
| [96] | 316 IVUS images of 26 patients | Thresholding + Polar transformation + Morphological operations | FOS + FD (Box counting) + GLCM + GLRLM + LTE | PCA | Deep belief network | Multiclass (DC, NC, FT, and FFT) | Pqsen.: 92.8 ± 0.1, Pqspec.: 85.1 ± 0.1, Pqacc.: 88.4 ± 0.1, PqPPV: 86 ± 0.1 PqNPV: 91.2 ± 0.1 (p < 0.05). |
|
| [97] | IVUS images of 7 patients | Adjacent pattern algorithm + Color moments of histogram + Statistical features | SVM based CNN | Multiclass (mild, moderate and severe) | Pqacc.: 98.80, 98.80, 97.59 Pqsen.: 100, 100, 100 Pqspec.: 98.70, 98.70, 97.40 Precision: 85.71, 85.71, 75 Recall: 100, 100, 100 F-score: 0.92, 0.92, 0.99 (Respectively for Mild, moderate and severe) |
|||
| [98] | IVUS images from 11 patients | Manual border segmentation | FOS + GLCM + GLRLM + LBP + Intensity + Discrete wavelet features +LTE | Genetic algorithm | Hybrid ensemble classifier(FFNN+ RF+ Ada boost) | Multiclass (DC, NC, FT, and FFT) | Pqacc.: 82.8, 71.6, 77 AUC: 0.832, 0.697, 0.787 Pqsen.: 84.4, 81.9, 74.9 Pqspec.: 81.9, 57.6, 82.4 PqPPV: 71.2, 72.4, 91.7 PqNPV: 90.8, 70.1, 55.9 (Respectively for Net1: FT/FFT or NC/DC Net2: FT or FFT Net3: NC or DC) |
|
| [99] | 2685 IVUS images of 15 patients | ImgTracer software | GLCM + GLRLM + IH + GLDS + NGTDM + IM + Statistical feature matrix | SVM (polynomial kernel 2nd order) | Coronary and carotid plaque | Pqacc.: 94.95 AUC: 0.95 Pqsen.: 92.88 Pqspec.: 96.61 PqPPV: 96.69 |
||
| [100] | 588 VH-IVUS images of 10 patients | Fuzzy c means and k means with particle swarm optimization | LBP + GLCM + MRL | PCA | SVM (RBF) | TCFA and Non-TCFA | Pqacc.: 98.61 | |
| [102] | 4000 IVOCT images from 49 patients | Cartesian Transformation | CNN from ImageNet ResNet50-v2 and DenseNet-121 |
Binary class: plaque (calcified plaque and lipid/fibrous plaque) and no plaque | Pqacc.: 91.7 Pqsen.: 90.9 Pqspec.: 92.4 |
|||
| [103] | CCTA of 150 patients | CNN (U Net + V Net) | Stenosis Detection and Plaque classification (calcified, partially calcified, noncalcified and no plaque) |
Stenosis identification: CCTA AI (p<0.001) AUC: 0.870 Pqacc.: 86 Pqsen.: 88 Pqspec.: 85 PqPPV: 73 PqNPV: 94 Plaque classification: AUC: 0.750 |
||||
| [104] | 12,325 IVUS images from 100 patients | IVUS and OCT registration + ROI segmentation | CNN | Binary (thin cap fibro-atherma and normal) | AUC: 0.911 Pqspec.: 82.81 Pqsen.: 87.31 |
|||
| [105] | 64 IVOCT images from 49 patients | Otsu’s method + morphological operation | Attenuation + Texture | RF | Multiclass (fibrotic, calcified, and lipid rich) | Pqacc.: 81.5 | ||
| [106] | 4469 IOCT images of 48 patients | Edge detection + Gaussian filter | t-test | CNN followed by post processing (Conditional Random Field + Morphological processing) | Multiclass (fibrocalcific, fibro-lipidic and other classes) |
Pqacc.: 77.7 ± 4.1, 86.5 ± 2.3, 85.3 ± 2.5 Pqsen.: 80, 85, 84 Pqspec.: 95, 92, 92 (Respectively for fibrocalcific, fibro-lipidic, other classes) p-value: 0.00027 |
||
| [107] | 700 OCT images of 28 patients | CNN | Multiclass (calcium, lipid tissue, fibrous tissue, mixed tissue, media and no visible tissue) | Pqacc.: 96 | ||||
| [108] | CCTA scans of 131 patients | 3D Recurrent Convolutional Neural Network | Multiclass (no plaque, non-calcified, mixed, calcified) and stenosis (no stenosis, non- significant, significant) | Plaque analysis: Pqacc.: 72, F1 score: 0.61 Cohen’s kappa: 0.60 Stenosis analysis: Pqacc.: 81 F1 score: 0.78 Cohen’s kappa: 0.70 Pqsen.: 61 PqPPV: 83 |
||||
| [109] | CCTA scans of 163 patients | Recurrent convolutional neural network | Multiclass(calcified, non-calcified and mixed) | Plaque detection: Pqacc.: 77 F1 score: 0.61 Cohen’s kappa: 0.61 Stenosis detection: Accuracy: 80% F1 score: 0.75 Cohen’s kappa: 0.68 Pqsen.: 61 PqPPV: 65 |
||||
| [110] | CTA scans from 25 patients | Multiplanar reformation technique | 3D CNN U-Net (Encoder-decoder) |
Multiclass (calcified plaque, non-calcified plaque and mixed calcified plaque) | Dice scores: 0.83, 0.73, 0.68 Pqsen.: 85, 76,72 PqPPV: 82, 69, 62 Respectively for CAP, NCAP, MCAP |
|||
| [111] | 2060 CTA images from 60 patients | Higher-order spectra cumulants | Multiple factor analysis + t-test | SVM(RBF) | Binary (calcified, noncalcified) | Pqacc.: 95.83 Pqsen.: 94.54 Pqspec.: 97.13 PqPPV: 97.05 |
||
* AUC: Area Under Curve, PqPPV (%): Plaque Positive Predictive Value, PqNPV (%): Plaque Negative Predictive Value, Pqsen. (%): Plaque Sensitivity, Pqspec. (%): Plaque Specificity, Pqacc. (%): Plaque accuracy, Pqavg.acc. (%): Plaque average accuracy, Precision (%).