Segmentation Of Microcalcifications In Mammogram Images Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

An intutive segmentation technique to separate the microcalcification regions from the mammogram is given in this article. This mechanism, first enhances the input image using an image-dependent threshold value and binarizes it to obtain an enhanced image. Then the pixels constituting the edges of microcalcification regions are grown in the enhanced image, with respect to the neighbourhood pixels. Lastly, the edge intensities of the enhanced image are remapped into the original image, using which the regions of interest are segmented. The results of the present work are compared with the ground realities of the sample images obtained from MIAS database.

Keywords-Microcalcification; Mammogram; Threshold; Region-Growing; Segmentation; Enhancemnet.


Digital Image Processing (DIP) is a technique of processing any digital image, which is a collection of discrete intensity values, each being identified by a unique pair of planar coordinates. The key objective of this technique is to obtain application-specific information from the images pertaining to Medicine, Defence, Industries, Law enforcement, Remote and Satellite Sensing, etc. [1].

It is evident from the literature that in the recent past, the potentials of DIP have exponentially scaled up the application domains beyond our prediction. The prospects and promises of automatic image processing, especially for Medical diagnosis to identify the severity of any disorder, have substantially increased the multi-dimensional perspectives of image processing. Segmentation is the process of partitioning an image into multiple sets of pixels, in order to simplify the representation of an image in a more meaningful way [1].

Breast Cancer continues to be a worrying global health issue of women and prognosis is deemed as the foremost remedy to combat it. Microcalcifications are the tiny specks of calcium present in breast, which may appear as a single mass or a clusters, whereas macrocalcifications are in the form of coarse deposits which may probably change into malignant (cancerous cells) over a period or may remain as benign (Non- cancerous cells) [2, 3].

©978-1-4673-2907-1/13/$31.00 ©2013 IEEE

Digital Mammogram image has emerged as a reliable source for the prognosis of breast cancer by diagnosing and locating the microcalcifications, pre-cancerous lesions, etc. A positive screening mammogram is classified as a true positive if breast cancer diagnosis is confirmed with biopsy. A false positive screening mammogram is defined as true negative if the mammogram and/or ultrasound of the breast and/or biopsy are excluded the diagnosis breast cancer [ 2, 3].

Image enhancement in the form of contrast / brightness enhancement plays a significant role in identifying the edges of a region. Threshold-based enhancement is categorized into two as global threshold-based and local threshold-based enhancement. The former technique enhances the entire image with reference to a single threshold, while the latter enhances the image using the threshold values, which are dynamically obtained from the subimages of the input image [4].

The segmentation technique presented in this paper has two intutive phases namely global threshold-based enhancement and an intensity-directed region-growing mechanism for segmentation.

In this article, section II describes the methodology of the proposed technique. The results and discussions are given in section III and the conclusions drawn from this present work are summarized in section IV.

Methodology of the Proposed Method

The proposed work focuses on the identification of the microcalcifications present in the mammogram images, by tracing the intensities of the pixels. The algorithmic description of the present work is given below.

The input grayscale image is transformed into a binary image, based on an optimal threshold value, which is the average of the maximum and minimum intensity of the image.

Algorithm for the Proposed Segmentation Technique

Phase I: Image Enhancement

Input : A digital mammogram (M)

Output : Enhanced Image (ME)

Read the input image M of size M X N

Compute the Mmax and Mmin of image M

For the entire image of size M ï‚´ N do

Read M(x,y) /*M(x,y)  M */

Initialize the threshold (MT) value using

(Mmax + Mmin)/2

If M(x,y) > MT and M(x,y)<Mmax then

ME(x,y) = Mmax

elseif M(x,y) < MT and M(x,y) > Mmax then

ME(x,y) = Mmin

Phase II: Image Segmentation

Input : Enhanced Image (ME) and original image (M)

Output : Segmented Image (MS)

For the entire image ME

Read the input image ME (ME(x,y)  M)

if ME(x,y) > Mmax AND ME(x,y) < MT then

if ME(x,y) ≠ Mmin

{ ME(x,y)  N4( ME(x,y)} then

/* N4: Four Neighbourhood */

ME(x,y) = Mmin

{ ME(x,y)  N8( M(x,y)}

/* N8: Eight Neighbourhood */

Remap the intensities of edges plotted in ME into the input image M.

Extract the regions obtained in M.


This algorithm in Phase I uses a global threshold (MT) value which is obtained as an average of the minimum and the maximum intensity of the input mammogram. The image is bisected into two, with respect to the intensity criteria given in Step 3 (iii) that results in the enhanced image ME, using which the target regions are segmented in the original image M. The phase II of this technique uses the intensity profile of ME as per step 4 (ii). The regions are grown using the pixels of N4 or N8, which satisfy the devised criterion.

In this algorithm a global threshold value, which remains unchanged throughout the process of enhancement and segmentation is used. Initially, the digital mammogram image (M) is taken as the input image, from which the maximum (Mmax) and minimum (Mmin) intensity of the image is computed. The average of Mmax and Mmin forms the threshold value MT, using which the entire image is transformed into bi-level image ME. From the bi-level image, the threshold image (MT) is obtained, by separating the foreground and background of the image. The four neighbourhoods (N4) of the pixels of ME whose intensity value is greater than MT and less than Mmax are set to Mmin and otherwise the eight neighbourhoods (N8) are set to Mmin. The N4 or N8 pixels set to Mmin constitute the edge pixels, which are correspondingly remapped from ME to the original image M.

Results and Discussion

The algorithm presented in this paper is developed in MATLAB 7.8. The performance of the algorithm presented in this paper is evaluated on more than fifty sample images obtained from Mammographic Image Analysis Society (MIAS) database [11]. For illustrative purpose, the results obtained for five of those images [mdb010, mdb013, mdb019, mdb023 and mdb032] are presented in Figure 1 (a) to (e). These figures depict the process of segmentation in two stages as enhanced images plotted with the prospective edge pixels using Figure 1 (f) to (j) and the region-grown images are given in Figure 1 (k) to (o).

It is visually evident that the devised region-growing technique has accurately segmented the regions of varying intensities indicating the presence of disorders of different sizes in various locations. These segmented images may further be subjected to intensity classification using which the features of microcalcification can be studied and intensity-based inferences be drawn.

The obtained results are qualitatively validated by comparing them with the ground reality descriptions provided in the MIAS dataset. In the MIAS description the first column denotes the feature of background tissue as F - Fatty, G - Fatty-glandular and D - Dense-glandular. The second column specifies the class of abnormalities present in the image as CALC - Calcification, CIRC - Well-defined / circumscribed masses, SPIC - Spiculated masses, MISC - Other, ill-defined masses, ARCH - Architectural distortion, ASYM - Asymmetry and NORM - Normal. The third column specifies the severity of the abnormality as B - Benign and M - Malignant. The fourth and fifth column present the co-ordinate pair of centre of abnormalities (x,y) and the sixth column specifies about the approximate radius (in pixel) of a circle enclosing the abnormality [11].

The initial and final co-ordinate points of the experimental results are obtained, by sampling the input images in Adobe Photoshop CS3 version.

The comparative description between MIAS specifications and the obtained results of the proposed method are furnished in Table I.

Spatial Comparision of MIAS Datasets and the obtained Results

Image Reference

MIAS Description

Co-ordinates of segmented region

Initial position

Final position


F CIRC B 525 425 33

473 389

576 454


G MISC B 667 365 31

607 311

698 412


G CIRC B 653 477 49

591 456

684 491


CIRC M 538 681 29

488 623

567 715


MISC B 388 742 66

321 694

421 796

It is apparent from the comparative analysis that the obtained results highly comply with the MIAS database descriptions, which is an added merit of the algorithm presented in this paper. The results obtained out of the segmentation process are depicted in Figure 1.
















(a) - (e): Original digital mammogram images (mdb010, mdb013, mdb019, mdb023 & mdb032); (f) - (j): Region - grown images of (a) - (e); (k) - (o): Segmented regions of (a) - (e)

Figures 1 (a) - (e) are sample images processed by the proposed technique. The respective images obtained after enhancement, region-growing and intensity remapping are shown in Figure 1 (f) - (j). The resultant segmented images are given in Figure 1 (k) - (o). It is evident that the regions of microcalcifications identified by the present work agree well with the specifications given in the MIAS database.


This paper presents a novel technique to segment the microcalcifications in the mammogram. This mechanism is found to be robust on the digital mammogram images of varying intensities, orientations and dimensions. The qualitative results of this technique endorse its merits, as the precision of segmentation converges to the regions of dissimilar intensities and closely agree with MIAS specifications. Hence, this algorithm provides greater scope in prognosis and early screening of breast cancer among the subjects. The outcome of this algorithm would serve as a supplementary data, to validate the diagnostic inferences of the physicians.