Computerized detection of masses on mammograms by entropy maximization thresholding

In many cases, masses in X-ray mammograms are subtle and their detection can benefit from an automated system serving as a diagnostic aid. It is to this end that the authors propose in this paper, a new computer aided mass detection for breast cancer diagnosis. The first step focuses on wavelet filters enhancement which removes bright background due to dense breast tissues and some film artifacts while preserving features and patterns related to the masses. In the second step, enhanced image is computed by Entropy Maximization Thresholding (EMT) to obtain segmented masses. The efficiency of 98,181% is achieved by analyzing a database of 84 mammograms previously marked by radiologists and digitized at a pixel size of 343μmm x 343μ mm. The segmentation results, in terms of size of detected masses, give a relative error on mass area that is less than 8%. The performance of the proposed method has also been evaluated by means of the receiver operating-characteristics (ROC) analysis. This yielded respectively, an area (Az) of 0.9224 and 0.9295 under the ROC curve whether enhancement step is applied or not. Furthermore, we observe that the EMT yields excellent segmentation results compared to those found in literature. (author)