Table 2 Summary of articles in classification of breast tumour.
References | Topic | Modality | Strategies | Sample Size | Performances | Issues |
|---|---|---|---|---|---|---|
Choi et al.34 | Effect of a deep learning framework-based computer-aided diagnosis system on the diagnostic performance of radiologists in differentiating between malignant and benign masses on breast ultrasonography | US | 1. Deep Learning (GoogLeNet) | 173 benign and 80 malignant breast messes | Sensitivity: 85.0% Specificity: 94.0% | 1, 2 |
Zhou et al.31 | Classification of benign and malignant breast tumours in ultrasound images with posterior acoustic shadowing (PAS) using half-contour features | US | 1. Disk expansion (DE) segmentation 2. Half-contour detection of PAS: tumour circularity (TC) and standard deviation of degree (SDD) | 40 cases with PAS / 10 cases without PAS | Sensitivity: 72% Specificity: 76% AUC: 0.78 | |
Becker et al.23 | Classification of breast cancer in ultrasound imaging using a generic deep learning analysis software | US | 1. Deep learning-based image software (ViDi Suite v 2.0) | 82 patients with malignant / 550 with benign lesions | Sensitivity: 80.4% Specificity: 84.2% AUC: 0.84 | 1 |
Ciritsis et al.32 | Automatic classification of ultrasound breast lesions using a deep convolutional neural network | US | 1. Deep Learning (dCNN) | 40 patients | Sensitivity: 92.1% Specificity:76.0% AUC: 83.8 | 1 |
Byra et al. 38 | Classification of breast lesions using segmented quantitative ultrasound maps of homodyned K distribution parameters | US | 1. Homodyned K distribution maps 2. Markov random field model | 32 malignant breast tumours / 71 benign masses | Sensitivity: 76.8% Specificity: 75.8% AUC: 0.84 | 2, 3, 6 |
Silva et al.29 | Breast tumour classification in ultrasound images using neural networks with improved generalisation methods | US | 1. Neural network 2. 22 mixed features | 50 malignant and 50 benign tumours | Sensitivity: 97% Specificity: 96% AUC: 0.98 | 2, 3 |
Cai et al. 39 | Robust phase-based texture descriptor for classification of breast ultrasound images | US | 1. Phase congruency detection 2. Texture-based feature extraction 3. SVM | 69 benign and 69 malignant cases | Sensitivity: 83.4% Specificity: 85.4% AUC: 0.86 | 2, 3 |
Ara et al.27 | Bimodal multiparameter-based approach for benign–malignant Classification of breast tumours | US / UE | 1. Genetic Algorithm | 170 patients (56 malignant lesions and 145 benign cases) | Sensitivity: 89.3%* Specificity: 80.1%* AUC: 0.91* | 2, 3 |
Moon et al. 40 | The adaptive computer-aided diagnosis system based on tumour sizes for the classification of breast tumours detected at screening ultrasound | US | 1. Quantitative morphological and textural features 2. Linear logistic regression | 156 tumours (78 benign and 78 malignant) | Sensitivity: 85.4% Specificity:77.8% AUC: 0.86 | 2, 3 |
Singh et al. 30 | Fuzzy cluster based neural network classifier for classifying breast tumours in ultrasound images | US | 1. Fuzzy c-means clustering / back-propagation artificial neural network 2. SVM | 178 patients and mixed with open data (88 benign / 90 malignant) | Sensitivity: 94.7% Specificity: 93.6% AUC: 0.94 | 2, 3, 5, 6 |
Jain et al. 28 | Texture ratio vector technique for the classification of breast lesions using SVM | US | 1. Texture feature: inside the lesion (IAI) and upper side of lesion (UAI) 2. Statistical texture: 7 features 3. SVM | 117 images (45 benign and 72 malignant) from open data | Overall classification accuracy: 86.6% | 2, 3, 5, 6 |
