2018/09/05 by Vivek Kumar Singh, Hatem A. Rashwan, Singh, Vivek Kumar +18 · 2 citations
Computer Science · Medicine · #AI in cancer detection #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Infrared Thermography in Medicine #Radiomics and Machine Learning in Medical Imaging
paper · pdf · doi:10.48550/arxiv.1809.01687
openalex publication_date 2018/09/05 · openalex created_date 2022/08/03 · openalex updated_date 2026/07/28
Mammogram inspection in search of breast tumors is a tough assignment that\nradiologists must carry out frequently. Therefore, image analysis methods are\nneeded for the detection and delineation of breast masses, which portray\ncrucial morphological information that will support reliable diagnosis. In this\npaper, we proposed a conditional Generative Adversarial Network (cGAN) devised\nto segment a breast mass within a region of interest (ROI) in a mammogram. The\ngenerative network learns to recognize the breast mass area and to create the\nbinary mask that outlines the breast mass. In turn, the adversarial network\nlearns to distinguish between real (ground truth) and synthetic segmentations,\nthus enforcing the generative network to create binary masks as realistic as\npossible. The cGAN works well even when the number of training samples are\nlimited. Therefore, the proposed method outperforms several state-of-the-art\napproaches. This hypothesis is corroborated by diverse experiments performed on\ntwo datasets, the public INbreast and a private in-house dataset. The proposed\nsegmentation model provides a high Dice coefficient and Intersection over Union\n(IoU) of 94% and 87%, respectively. In addition, a shape descriptor based on a\nConvolutional Neural Network (CNN) is proposed to classify the generated masks\ninto four mass shapes: irregular, lobular, oval and round. The proposed shape\ndescriptor was trained on Digital Database for Screening Mammography (DDSM)\nyielding an overall accuracy of 80%, which outperforms the current\nstate-of-the-art.\n