Detección de microcalcificaciones mamarias agrupadas

Abstract

Breast microcalcifications are non-palpable lesions present in approximately 55% of breast cancer cases. It is a common find in mammograms and can be an indicator of this disease in early stages. We implemented two enhancement methods based on multiresolution analysis and wavelet transform. Then, used thresholding method to image segmentation. The threshold was determined using statistics parameters of coefficient distribution. Later, the objects were classified by length and through the density based clustering algorithm were detected clustered microcalcifications candidates. In this work we compared microcalcifications enhancement and image segmentation were used different functions of wavelets families: symlet, daubechies and coiflet. The function that presented the best result was daubechies order 16, with a specificity of 68% sensitivity of 77% and a positive predictive value of 72 %. Coiflet had the worst result. Algorithm detects breast clustered microcalcifications in most of cases. The sensitivity, specificity and positive predictive value tests do not have significant variation between wavelet functions. However, the daubechies and symlet wavelet families present the best results.