Conditional Generative Adversarial and Convolutional Networks for X-ray Breast Mass Segmentation and Shape Classification
This paper proposes a novel approach for breast mass segmentation in mammography image crops containing mass ROI,based on conditional Generative Adversarial Networks, as its structure is well suited for accurately outlining mass borders, especially when there are few training samples available: the generative network learns intrinsic features of tumors while the adversarial networks enforces segmentation to be similar to ground truth samples. Experiments performed on dozens of malignant tumors extracted from the public DDSM dataset and from our in-house dataset confirm our hypothesis with very high Jaccard and Dice coefficients (>89 the scores obtained by other state-of-the-art approaches. Furthermore, to check that the generated binary masks portray significant morphological features of the tumor, we have also designed a specific Convolutional Neural Network for classifying the tumor area into four types (round, oval, lobular or irregular), which provides around 72
READ FULL TEXT