Estimation of Architectural Distortion in Mammograms using Fractal Features

Abstract

Moving towards accurate breast cancer detection, X-ray mammography is the gold standard in medical imaging for its efficiency and reliability. Abnormalities often encountered in mammograms are in the form of benign or malignant masses, calcifications, asymmetry and architectural distortion. Other than masses and calcifications, architectural distortion should not be overlooked, since it is often the major cause of nonpalpable cancer. However, due to the appearance variability and subtle differences of the abnormalities from the tissues, the radiologists face ambiguity to detect and differentiate the malignant one from the benign one. Due to the existence of irregular and ill-defined structure in architecturally distorted areas, fractal features namely fractal dimension and lacunarity are considered in our work to differentiate the malignant architectural distortion from the benign architectural distortion. Our study can provide a second opinion to the radiologists in decision making. The performance of the proposed system has been evaluated with a dataset of total 19 mammograms with architectural distortions from the mini-MIAS database. The Mann Whitney Wilcoxon nonparametric test shows the statistical significance of fractal features in differentiating the abnormal mammograms from the benign ones. Based on the experimental results, we found that the combination of fractal dimension and lacunarity feature gives a prediction accuracy of 90%.