Jk. Kim et al., DETECTION OF CLUSTERED MICROCALCIFICATIONS ON MAMMOGRAMS USING SURROUNDING REGION DEPENDENCE METHOD AND ARTIFICIAL NEURAL-NETWORK, Journal of VLSI signal processing systems for signal, image, and video technology, 18(3), 1998, pp. 251-262
Citations number
21
Categorie Soggetti
Computer Science Information Systems","Engineering, Eletrical & Electronic","Computer Science Information Systems
Clustered microcalcifications on X-ray mammograms are an important sig
n in the detection of breast cancer. A statistical texture analysis me
thod, called the surrounding region dependence method (SRDM), is propo
sed for the detection of clustered microcalcifications on digitized ma
mmograms. The SRDM is based on the second-order histogram in two surro
unding regions. This method defines four textural features to classify
region of interests (ROIs) into positive ROIs containing clustered mi
crocalcifications and negative ROIs of normal tissues. The database is
composed of 64 positive and 76 negative ROI images, which are selecte
d from digitized mammograms with a pixel size of 100 x 100 mu m(2) and
12 bits per pixel. An ROI is selected as an area of 128 x 128 pixels
on the digitized mammograms. In order to classify ROIs into the two ty
pes, a three-layer backpropagation neural network is employed as a cla
ssifier. A segmentation of individual microcalcifications is also prop
osed to show their morphologies. The classification performance of the
proposed method is evaluated by using the round-robin method and a fr
ee-response receiver operating-characteristics (FROC) analysis. A rece
iver operating-characteristics (ROC) analysis is employed to present t
he results of the round-robin testing for the case of several hidden n
eurons. The area under the ROC curve, A(z), is 0.997, which is achieve
d in the case of 4 hidden neurons. The FROG analysis is performed on 2
0 cropped images. A cropped image is selected as an area of 512 x 512
pixels on the digitized mammograms. In terms of the FROG, a sensitivit
y of more than 90% is obtained with a low false-positive (FP) detectio
n rate of 0.67 per cropped image.