Discrimination of Breast Cancer with Microcalcifications on Mammography by Deep Learning

Abstract

Microcalcification is an effective indicator of early breast cancer. To improve the diagnostic accuracy of microcalcifications, this study evaluates the performance of deep learning-based models on large datasets for its discrimination. A semi-automated segmentation method was used to characterize all microcalcifications. A discrimination classifier model was constructed to assess the accuracies of microcalcifications and breast masses, either in isolation or combination, for classifying breast lesions. Performances were compared to benchmark models. Our deep learning model achieved a discriminative accuracy of 87.3% if microcalcifications were characterized alone, compared to 85.8% with a support vector machine. The accuracies were 61.3% for both methods with masses alone and improved to 89.7% and 85.8% after the combined analysis with microcalcifications. Image segmentation with our deep learning model yielded 15, 26 and 41 features for the three scenarios, respectively. Overall, deep learning based on large datasets was superior to standard methods for the discrimination of microcalcifications. Accuracy was increased by adopting a combinatorial approach to detect microcalcifications and masses simultaneously. This may have clinical value for early detection and treatment of breast cancer.

Figure 1. An illustrative example showing segmentation of microcalcifications and breast mass in a mammogram of the left breast of a 60-year-old patient with invasive ductal carcinoma.

(a) The craniocaudal (CC) view shows focal clustered microcalcifications (indicated by thin arrows) and an irregular circumscribed mass (indicated by a thick arrow). (b) The suspicious mass is automatically delineated within the red curve. (c) the segmented microcalcifications detected in (b) are used to characterize the features.

Figure 2. An illustrative example showing segmentation of microcalcifications in a mammogram of the left breast of a 56-year-old patient with ductal carcinoma in situ.

(a) The mediolateral oblique (MLO) view shows clustered coarse and low density microcalcifications (indicated by thin arrows). (b) The image shows the region of suspicious microcalcifications(indicated by thin arrows). (c) The segmented microcalcifications from (b) are used to characterize the features.

Figure 3. An illustrative example showing segmentation of microcalcifications in a mammogram of the right breast of a 49-year-old patient with fibrocystic changes.

(a) The focal microcalcifications (indicated by thin arrows) appear low contrast compared with the dense background in the mediolateral oblique (MLO) view. (b) The region of suspicious microcalcifications is indicated by thin arrows. (c) A zoomed-in view of (b) highlights the segmented microcalcifications.

Figure 4. ROC curves for selected microcalcification features.

The ROC curves compare the discriminative performances of individual features versus combinations of features.

Figure 5. ROC curves comparing the discriminative performances of the four classification models.

The three scenarios show the ROC curves based on the following classifications: (a) Microcalcifications alone (15 segmentation features). (b) Breast masses alone (26 segmentation features). (c) Microcalcifications plus breast masses (41 segmentation features).