Table 8. A comparative view of several mass detection methods based on different DCNN architectures and datasets, including the newly proposed method.
Numbers in red indicate the best values between the several techniques.
| Reference | Contribution | Data set | AUC | Accuracy |
|---|---|---|---|---|
| Sahiner et al. (1996) | CNN to classify mass and normal breast | Dataset obtained by radiologists | 0.87 | – |
| Jain & Levy (2016) | DCNN—AlexNet classify benign and malignant masses | DDSM | – | 66% |
| Huynh & Giger (2016) | DCNN—AlexNet features to classify benign and malignant tumors | University of Chicago Medical Centre | 0.81 | – |
| Jiang (2017) | DCNN—GoogLeNet DCNN –AlexNet | BCDR-F03 | 0.88 0.83 | – |
| Duraisamy & Emperumal (2017) | DCNN –Vgg | MIAS and BCDR | 0.9 | 85% |
| The proposed CAD system | DCNN-SVM –AlexNet –cropping ROI manually –classify benign and malignant masses | DDSM | 0.88 | 79.1% |
| DCNN-SVM –AlexNet –threshold and region based –classify benign and malignant masses | 0.88 | 80.9% | ||
| DCNN-SVM –AlexNet –classify benign and malignant masses | CBIS-DDSM | 0.94 | 87.2% |