Detection of Microcalcification in Digitized Mammogram Images Using LVQ

Proceedings of UIBL International Conference 2012, Jakarta, pages 148-152

Abstract :

The presence of tight clusters of microcalcifications in a mammogram image indicates the early breast cancer and the analysis of the mammogram image is usually done by a doctor or radiologist. Microcalcifications are calcium deposits that may be found in the breast gland, appear as white dots on mammogram image. However, the analysis is often constrained by the subjectivity due to experience level, fatigue and other human factors. This is because the analysis tasks are repetitive and time consuming. The existence of device or automated tools will greatly assist the work. The main objective of this research is to detect microcalcifications in digitized mammogram images with the aid of Matlab software tools. In this research, the detection of microcalcifications in digitized mammogram process is designed as a pattern recognition system using Artificial Neural Network (ANN), i.e. LVQ (Learning Vector Quantization) classifier. The microcalcification detection is formulated as a supervised-learning problem and classification was based on six-feature input given to the ANN. The system detects microcalcifications in three steps. Firstly, a tophat filtering is applied on the images, and then six features of the image are extracted. Finally, image classification is carried out in recognizing the tissue containing microcalcifications. The results of the classification are used to locate microcalcifications in the image. Mammogram images that have been identified will be compared with analysis results done by doctor or radiologist. This benchmarking process aims to determine the amount of TP (True Positive) and FN (False Negative). The final results show that the detection system sensitivity is 72.22%, and the average number of FN is 0.625 per image. 

Keywords: microcalcifications, mammograms, detection, LVQ, True Positive (TP), False Negative (FN)

Full text : Download PDF  (873)

Back ...