Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (7): 243-249.DOI: 10.3778/j.issn.1002-8331.2009-0506

• Graphics and Image Processing • Previous Articles     Next Articles

Gleason Grading of Prostate Cancer Based on Improved Convolution Neural Network

SHAN Yiqing, HUANG Mengxing, ZHANG Yu, LI Yuchun, ZHANG Xinhua, FENG Siling, CHEN Jing   

  1. 1.School of Computer Science and Cyberspace Security, Hainan University, Haikou 570228, China
    2.School of Information and Communication Engineering, Hainan University, Haikou 570228, China
    3.Department of Radiology, Haikou People’s Hospital, Haikou 570228, China
  • Online:2022-04-01 Published:2022-04-01



  1. 1.海南大学 计算机与网络空间安全学院,海口 570228
    2.海南大学 信息与通信工程学院,海口 570228
    3.海口市人民医院 放射科,海口 570228

Abstract: Prostate cancer is one of the most common cancers in the world, second only to lung cancer. In the diagnosis of prostate cancer, the most commonly used method is pathological experts to observe the stained biopsy tissue through the microscope, and get the Gleason score of tissue microarray image. In a large number of tissue microarray images, it is very time-consuming for pathological experts to use Gleason pattern to score prostate cancer tissue microarray, which is easily affected by subjective factors among different observers, and has low repeatability. The development of deep learning and computer vision makes the computer-aided diagnosis system of pathology more objective and repeatable. U-Net is the most widely used network in the field of medical image segmentation, which is different from the classifier used in previous studies. A region segmentation model based on improved net network is proposed, which combines the deep and shallow features through dense connection blocks and supervises the features of each scale at the same time. It can reduce the network parameters and improve the calculation efficiency, and verify the effectiveness of the method on the annotated complete dataset.

Key words: U-Net network, Gleason pattern, image segmentation

摘要: 前列腺癌是全球范围内男性最常见的癌症之一,仅次于肺癌。在前列腺癌的诊断过程中最常用的方法是病理学专家通过显微镜对染色活检组织进行观察,得出组织微阵列图像的Gleason评分。在大量的组织微阵列图像下,病理学专家使用Gleason模式对前列腺癌组织微阵列进行评分非常耗时,易受到不同观察者之间主观因素的影响,且可重复性低。深度学习和计算机视觉的发展使得病理学计算机辅助诊断系统更具有客观性和可重复性。U-Net是医学影像分割领域应用最广泛的的网络,不同于以往研究中使用分类器,提出了一种基于改进的U-Net网络的区域分割模型,通过密集连接块来融合深层和浅层特征的同时对各个尺度的特征进行监督。可以减少网络参数,提高计算效率,并在标注完整的数据集上验证了方法有效性。

关键词: U-Net网络, Gleason模式, 图像分割