Breast Microcalcification Diagnosis Using Deep Convolutional Neural Network from Digital Mammograms

Table 3

Comparison of performance for MC classification on different sets of features.

Method	Accuracy	Precision	Specificity	AUC	Sensitivity
CNN	0.8768 ± 0.0431	0.8891 ± 0.0349	0.8667 ± 0.0457	0.9336 ± 0.0238	0.8701 ± 0.0144
Morphological	0.8525 ± 0.0203	0.8624 ± 0.0267	0.8311 ± 0.0471	0.9256 ± 0.0211	0.8492 ± 0.0246
CNN + morphological	0.8828 ± 0.0437	0.8911 ± 0.0447	0.8667 ± 0.0602	0.9385 ± 0.0238	0.8761 ± 0.0104
CNN filtered by morphologic	0.8859 ± 0.0363	0.8932 ± 0.0384	0.8689 ± 0.0528	0.9392 ± 0.0240	0.8843 ± 0.0344
Textural	0.7677 ± 0.0634	0.7964 ± 0.0659	0.7511 ± 0.0924	0.8721 ± 0.0530	0.7703 ± 0.0544
CNN + textural	0.8727 ± 0.0500	0.8853 ± 0.0410	0.8622 ± 0.0522	0.9338 ± 0.0248	0.8801 ± 0.0434
CNN filtered by textural	0.8747 ± 0.0387	0.8842 ± 0.0423	0.8578 ± 0.0603	0.9434 ± 0.0220	0.8831 ± 0.0276
Morphological + textural	0.8667 ± 0.0223	0.8768 ± 0.0309	0.8489 ± 0.0511	0.9381 ± 0.0219	0.8601 ± 0.0251
CNN + morphological + textural	0.8818 ± 0.0434	0.8895 ± 0.0457	0.8644 ± 0.0624	0.9379 ± 0.0237	0.8791 ± 0.0124
CNN filtered by morphological + textural	0.8747 ± 0.0376	0.8873 ± 0.0238	0.8644 ± 0.0339	0.9398 ± 0.0242	0.8751 ± 0.0328