BMLDD: Breast mass lesion detection by using Deep Learning

Early detection of breast cancer can increase treatment efficiency and therefore decline the high rate of moralities inpatients with breast cancers. Computer-aided detection systems can help radiologists to detect breast mass lesions ina faster and more efficient way. This study proposed BMLDD, a new pipeline for breast mass lesion detection by adeep learning method using mam-mographic images. The proposed method uses histogram stretching,morphological operations, color-map, and a CNN with VGG۱۶ architecture as its fundamental steps. This pipelinecan efficiently improve the image condition, remove artifacts, extract breast region from the back-ground, highlightthe differences in images, extracting useful features, and finally detect the breast masses. The performanceevaluation on INbreast mammographic images has shown that BMLDD can yield promising results with ۹۶%Sensitivity, ۹۶% Precision, ۹۶% F۱-score, ۹۶% Accuracy and ۹۹% AUC. Moreover, the comparisons verified thatBMLDD performed better than other state-of-the-art methods. Consequently, BMLDD is an efficient method forbreast mass lesion detection.

Keywords:

Breast cancer mass , lesion detection , Deep learning , Morphological operation Mass detection , Convolutional neural network