The recognition of breast abnormality, which mainly consists of mass and micro-calcification, plays a critical role in the detection of breast cancer. To facilitate the procedure of making decisions on suspicious regions in mammograms, we proposed a light-weighted deep convolutional neural network (CNN) recognition system termed DisepNet, a light-weighted fully convolutional network that shows promising performance on the detection task of breast abnormality. In the proposed DisepNet, novel blocks feature extraction are designed, which are termed as Disep block and Incep-L block respectively. We evaluated the proposed model by 5-fold cross-validation on a combined dataset, which comes from two public databases MINI-MIAS and INbreast. The final accuracy of our proposed model achieved a mean accuracy at 95.60% while the sensitivity and specificity reached 93.71% and 97.44% respectively. A comparison between our model and existing methods shows that our model provides the best performance.

乳房异常的识别主要由肿块和微钙化组成，在乳腺癌的检测中起着至关重要的作用。为了促进对乳房X线照片中的可疑区域做出决策的过程，我们提出了一种轻量化的深度卷积神经网络（CNN）识别系统，称为DisepNet，这是一种轻量化的全卷积网络，在乳腺异常检测任务中显示出令人鼓舞的性能。在提出的DisepNet中，设计了新颖的块特征提取，分别称为Disep块和Incep-L块。我们通过来自两个公共数据库MINI-MIAS和INbreast的组合数据集上的5倍交叉验证对提出的模型进行了评估。我们提出的模型的最终精度达到了95的平均精度。60％，而敏感性和特异性分别达到93.71％和97.44％。我们的模型与现有方法之间的比较表明，我们的模型提供了最佳性能。