Microcalcification Segmentation from Mammograms: A Morphological Approach

Abstract

This publication presents a computer method for segmenting microcalcifications in mammograms. It makes use of morphological transformations and is composed of two parts. The first part detects microcalcifications morphologically, thus allowing the approximate area of their occurrence to be determined, the contrast to be improved, and noise to be reduced in the mammograms. In the second part, a watershed segmentation of microcalcifications is carried out. This study was carried out on a test set containing 200 ROIs 512 × 512 pixels in size, taken from mammograms from the Digital Database for Screening Mammography (DDSM), including 100 cases showing malignant lesions and 100 cases showing benign ones. The experiments carried out yielded the following average values of the measured indices: 80.5% (similarity index), 75.7% (overlap fraction), 70.8% (overlap value), and 19.8% (extra fraction). The average time of executing all steps of the methods used for a single ROI amounted to 0.83 s.

Keywords: Breast cancer; Image processing; Mammography; Mathematical morphology; Microcalcification; Segmentation.

Fig. 1

Cranio-caudal view of an example mammogram A_1553_1.LEFT_CC. a Mammogram A_1553_1.LEFT_CC with a white rectangular ROI marked. b Enlarged ROI 512 × 512 pixels in size extracted from item (a)

Fig. 2

Two examples of patches with microcalcifications: benign (upper row, based on the image A_1551_1.LEFT_MLO), malignant (lower row, based on the image A_1214_LEFT_MLO). First column: mammographic image patches. Second column: the image inverted in gray levels. Third column: segmentation—all microcalcifications have been marked

Fig. 3

An illustration of detecting microcalcifications in an image 512 × 512 pixels in size, extracted from mammogram A_1131_1.RIGHT_MLO. a The result of microcalcifications detection in stage 1. b, c The result of using the detector based on Eq. (1) at the second and third levels of the morphological pyramid, respectively. d Extended maximum emax. e An image showing the marker for reconstructing microcalcifications detected at the second level of the morphological pyramid—an intersection of images from items (b) and (d). f The result of the reconstruction by dilation of the mask presented in item (d) and the marker presented in item (e), i.e., an image presenting microcalcifications detected at the second level of the morphological pyramid. g An image showing the marker for reconstructing microcalcifications detected at the third level of the morphological pyramid—an intersection of the images from items (c) and (d). h The result of the reconstruction by dilation of the mask presented in item (d) and the marker presented in item (g), i.e., an image presenting microcalcifications detected at the third level of the morphological pyramid. i The sum of images from items (f) and (h), using the OR operator, as the result of extracting microcalcifications using the morphological pyramid. The image has been subjected to additional “cleaning” operations

Fig. 4

An illustration of extracting the microcalcifications shape. Part 1. a Image A_1131_1.RIGHT. b The filtered and inverted image from item (a). c The result of extracting microcalcifications using the morphological pyramid from Fig. 3. d Regional minima obtained for the image from item (b). e The intersection of images from items (c) and (d). f The gradient obtained for the image from item (b)

Fig. 5

An illustration of extracting the microcalcifications shape. Part 2. a The result of the watershed segmentation, i.e., the external marker that has been overlaid on the image from Fig. 4b. b The internal marker. c The dilation of the external marker from item (a). d The logical sum of the external and internal markers. e The result of the minima imposition operation for the gradient image from Fig. 4f and the complete marker from Fig. 5d. f Output watershed lines overlaid on the image from Fig. 4b

Fig. 6

Graphs of the mean value and the standard deviation based on measurements of four indices: SI, OF, OV, and EF for the applied method, compared to the contours drawn by the radiologist based on the data from Table 2

Fig. 7

A graph of mean sensitivity values and standard deviations for microcalcifications detected in 200 mammograms from the DDSM database (100 malignant and 100 benign cases), depending on false-positive (FP) examples per image detected. Data based on Table 3

Fig. 8

Benign cases—example results of microcalcification segmentation for selected mammograms from then DDSM database, together with the GTA contours marked. The images have been inverted in the gray scale to better bring out individual microcalcifications. First column: the GTA contour marked. Second column: contours of individual microcalcifications traced by a radiologist. Third column: results of segmenting microcalcifications using the computer method, with the calculated indices. a–c Image A_1480_1.LEFT_MLO. d–f A_1551_1.LEFT_MLO. g–i A_1553_1.LEFT_MLO

Fig. 9

Malignant cases—example results of microcalcification segmentation for selected mammograms from the DDSM database, together with the GTA contours marked. The images have been inverted in the gray scale to better bring out individual microcalcifications. First column: the GTA contour marked. Second column: contours of individual microcalcifications traced by a radiologist. Third column: results of segmenting microcalcifications using the computer method, with the calculated indices. a–c Image A_1131_1.RIGHT_MLO. d–f A_1201_1.RIGHT MLO. g–i A_1214_1.LEFT MLO