Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (2): 161-169.DOI: 10.3778/j.issn.1002-8331.2107-0235

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

ConA-FPN Based Algorithm for Liver Tumor Detection

MA Jinlin, MAO Kaiji, MA Ziping, DENG Yuanyuan, OUYANG Ke, CHEN Yong   

  1. 1.School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, China
    2.Key Laboratory of the National Ethnic Affairs Commission for Intelligent Processing of Image and Graphics, Yinchuan 750021, China
    3.School of Mathematics and Information Science, North Minzu University, Yinchuan 750021, China
    4.Department of Interventional Radiology, General Hospital of Ningxia Medical University, Yinchuan 750004, China
  • Online:2023-01-15 Published:2023-01-15



  1. 1.北方民族大学 计算机科学与工程学院,银川 750021
    2.图像图形智能处理国家民委重点实验室,银川 750021
    3.北方民族大学 数学与信息科学学院,银川 750021
    4.宁夏医科大学总医院 放射介入科,银川 750004

Abstract: Deep learning methods are widely used in lesion detection tasks, but the problems of small liver tumors and small samples lead to the inability to achieve the accuracy requirements for assisted diagnosis. To address the above problems, this paper proposes a liver tumor detection algorithm based on ConA-FPN, and the specific process is as follows:first, the feature extraction network in Faster R-CNN is replaced by using a feature pyramid fused with ResNet and attention mechanism; then, the fused features are used to solve the problem of information loss of higher-level module channels in the feature pyramid, the CAG attention mechanism is added to solve the  problem that feature fusion brings the feature conflation, and enhances the detection ability of the model for small tumors; finally, transfer learning and data augmentation are used to enhance the detection ability and generalization ability of the model on small data sets. The experimental results show that the average accuracy of ConA-FPN on LITS2017 and 3D-IRCADB reaches 87.43%, which is significantly better than the mainstream detection models.

Key words: ConA-FPN, liver tumor, feature fusion, small targets, small datasets

摘要: 深度学习方法在病灶检测任务中被广泛应用,但因肝脏肿瘤较小和样本较少的问题,导致无法达到辅助诊断的准确率要求。针对以上问题,提出基于ConA-FPN的肝脏肿瘤检测算法,具体过程为:使用融合ResNet和注意力机制的特征金字塔替换Faster R-CNN中的特征提取网络;使用融合特征解决特征金字塔中的高层模块通道信息损失问题,通过添加CAG注意力机制解决了特征融合带来的特征混叠问题,增强模型对小肿瘤的检测能力;使用迁移学习和数据增强提升模型在小数据集上的检测能力和泛化能力。实验结果表明,ConA-FPN在LITS2017和3D-IRCADB上的平均精度达到87.43%,明显优于主流检测模型。

关键词: ConA-FPN, 肝脏肿瘤, 特征融合, 小目标, 小数据集