An Ad Hoc Random Initialization Deep Neural Network Architecture for Discriminating Malignant Breast Cancer Lesions in Mammographic Images

Table 1

Summary of the annotations available for each image in the CBIS-DDSM dataset. As all these annotations are derived from the image, none of these features were imputed into our classifier.

Patient_id	Anonymous alphanumeric code
Breast_density	4 (153), 2 (757), 3 (449), 1 (337)
Left or right breast	Left (817), right (879)
Image view	CC(784), MLO(912)
Abnormality id	1 (1570), 2 (84), 4 (10), 3 (28), 5 (2), 6 (2) (integer index used to label multiple lesions within the same image)
Abnormality type	Mass (1696)
Mass shape	Irregular (526), round (169), lobulated (399), oval (423), architectural_distortion(158), asymmetric_breast_tissue(26), lymph_node(45)
Mass margins	Focal_asymmetric_density (27), n/a (4), spiculated (407), circumscribed (455), ill_defined (472), obscured (308), microlobulated (143), n/a (60)
Assessment	5 (374), 4 (702), 0 (162), 3 (364), 2 (91), 1 (3)
Pathology	Malignant (784), benign (771), benign_without_callback (141)
Subtlety	5 (687), 4 (453), 2 (141), 3 (358), 1 (55), 0 (2)