【摘要】 结肠癌的鉴别在医学诊断中具有重要意义.实时、客观、准确的检查结果有利于医疗人员及时对症治疗.然而，现有的方法严重依赖专业医师对病理图像进行手工特征提取和分析，不仅检测周期较长，检测结果会存在不同程度的误差.据此，本文提出了一种基于注意力机制的卷积神经网络识别结肠癌的组织病理学图像.该网络采用轻量化的卷积网络结构，并嵌入注意力机制后，对结肠腺癌上皮(TUM)和正常结肠黏膜(NORM)的识别精度与F-1分值分别达到97%和0.971 8.假阴性率和假阳性率分别降低到2.3%和3.3%.该网络的优越性能为实时、客观和准确诊断癌症提供了全新的视角.更多还原

【Abstract】 Identification of colon cancer is of great significance in medical diagnosis. Real-time, objective and accurate inspection are facilitating medical professionals to explore symptomatic treatment. However, the existing methods mainly rely on hand-crafted features, which require professional expertise and long inspection cycle. Therefore, we propose a convolutional neural network based on the attention mechanism to identify histopathological images of colon cancer. Using the light weighted convolutional network structure and embedded attention mechanism, the proposed network can achieve the identification accuracy of 97% and F-1 score of 0.971 8 for colorectal adenocarcinoma epithelium(TUM) and normal colonic mucosa(NORM). The false negative rate and false positive rate can be reduced to 2.3% and 3.3%, respectively. The superior performance of the network provides a new perspective for real-time, objective and accurate cancer diagnosis.更多还原