基于 3D FPN 的肺结节检测

肺癌早期发现与诊断对提高肺癌患者生存率至关重要。肺癌主要是由恶性肺结节造成的，通过肺结节早期检测与诊断能够及时发现病情，显著提高肺癌存活率。随着深度学习网络在医学辅助诊断领域应用的迅速发展，现已有很多深度网络被应用于肺结节检测。通过统计检测目标肺结节半径分布情况，发现大部分肺结节半径较小。因此，结合 CT 影像数据的三维特性，提出使用 3D FPN 进行单阶段肺结节检测，能够解决肺小结节检测效果不佳的问题。在公用肺结节数据集 LUNA16 上验证了网络的有效性，CPM 值达到 0.893 2，相比其它肺结节检测网络，检测效果提高了 2%。

3D FPN-based Lung Nodule Detection

Early detection and diagnosis of lung cancer is essential to improve the survival rate of lung cancer patients. Lung cancer is mainly caused by malignant lung nodules. Through the detection and diagnosis of early lung nodules，the disease can be detected in time，and the survival rate of lung cancer is significantly improved. With the rapid development of the application of deep learning networks in the field of medical aided diagnosis，many deep networks have been proposed for the detection of lung nodules. By statistically detecting the radius distribution of the target lung nodules，it is found that most of the lung nodules have a small radius. Aiming at the problem of small detection target and the three-dimensional characteristics of CT image data，this paper proposes to use 3D FPN for single-stage pulmonary nodule detection，which can solve the problem of poor detection of small pulmonary nodules. The effectiveness of the network was verified on the common lung nodule data set LUNA16. The CPM value reached 0.8932. Compared with other lung nodule detection networks，the detection effect was improved by 2%.