Detection of Clustered Microcalcifications on Mammograms Using Surrounding Region Dependence Method and Artificial NeuralNetwork

  • Authors:

  • Jong Kook Kim;Jeong Mi Park;Koun Sik Song;Hyun Wook Park

  • Affiliations:

  • Department of Information and Communication Engineering, Korea Advanced Institute of Science and Technology, 207-43, Cheongryangri, Dongdaemungu, Seoul 130-012, Korea;Department of Diagnostic Radiology, Asan Medical Center, University of Ulsan College of Medicine, 388-1, Poongnap-Dong, Songpagu, Seoul 138-040, Korea;Department of Diagnostic Radiology, Asan Medical Center, University of Ulsan College of Medicine, 388-1, Poongnap-Dong, Songpagu, Seoul 138-040, Korea;Department of Information and Communication Engineering, Korea Advanced Institute of Science and Technology, 207-43, Cheongryangri, Dongdaemungu, Seoul 130-012, Korea

  • Venue:
  • Journal of VLSI Signal Processing Systems - special issue on applications of neural networks in biomedical image processing
  • Year:
  • 1998

Quantified Score

Hi-index 0.01

Visualization

Abstract

Clustered microcalcifications on X-ray mammograms are an importantsign in the detection of breast cancer. A statistical texture analysismethod, called the surrounding region dependence method (SRDM), is proposedfor the detection of clustered microcalcifications on digitized mammograms.The SRDM is based on the second-order histogram in two surrounding regions.This method defines four textural features to classify region of interests(ROIs) into positive ROIs containing clustered microcalcifications andnegative ROIs of normal tissues. The database is composed of 64 positive and76 negative ROI images, which are selected from digitized mammograms with apixel size of 100 × 100 μm^2 and 12 bits per pixel.An ROI is selected as an area of 128 × 128 pixels on the digitizedmammograms. In order to classify ROIs into the two types, a three-layerbackpropagation neural network is employed as a classifier. A segmentationof individual microcalcifications is also proposed to show theirmorphologies. The classification performance of the proposed method isevaluated by using the round-robin method and a free-response receiveroperating-characteristics (FROC) analysis. A receiveroperating-characteristics (ROC) analysis is employed to present the resultsof the round-robin testing for the case of several hidden neurons. The areaunder the ROC curve, A_z, is 0.997, which is achieved in thecase of 4 hidden neurons. The FROC analysis is performed on 20 croppedimages. A cropped image is selected as an area of 512 × 512 pixels onthe digitized mammograms. In terms of the FROC, a sensitivity of more than90% is obtained with a low false-positive (FP) detection rate of0.67 per cropped image.