基于残差混合扩张卷积的深度编解码人类精子头部分割网络  

A deep encoder-decoder network for human sperm head segmentation based on residual hybrid dilated convolution

作　　者：吕琪贤 范朝刚 詹曙[1] LüQi-xian;FAN Chao-gang;ZHAN Shu(School of Computer and Information,Hefei University of Technology,Hefei 231009;Eastern Theater General Hospital,Nanjing 210000,China)

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥231009 [2]东部战区总医院,江苏南京210000

出　　处：《计算机工程与科学》2021年第4期721-728,共8页Computer Engineering & Science

基　　金：国家自然科学基金(61371156)。

摘　　要：精子头部形状是精子形态分析中的一个重要指标,对诊断男性不育十分重要,因此准确高效地分割出精子头部至关重要。基于此,在残差网络的基础上融合扩张卷积与堆叠残差结构,构建了一个新型编解码分割网络。建立了一个用于分割人类精子头部的数据集,其中包含1207幅图像,并利用它来训练测试网络。所提出的网络能在多精子、无染色原图中获得优良的分割结果,在验证集上得到了96.06%的Dice系数。实验结果表明,堆叠残差模块和残差混合扩张卷积模块对分割效果有着显著提升作用。此外,本文网络处理的是呈现出精子真实形态的图像,其分割出的精准结果有利于医生临床诊断。Sperm head shape is an important indicator in the analysis of sperm morphology,which is very important for diagnosing male infertility.Therefore,it is very important to segment the sperm head accurately and efficiently.Based on this,this paper builds a new encoder-decoder segmentation network that combines stacked residual block and residual hybrid dilated convolution.Firstly,we build a dataset for segmenting the head of human sperm,which contains 1207 images,and then use it to train and test our network.The proposed network is able to achieve excellent segmentation results in low-quality images that are unstained and contain multiple sperms,and obtain a Dice coefficient of 96.06%on the validation set.The experimental results show that the stacked residual module and the residual hybrid dilated convolution module significantly improve the segmentation performance.In addition,the proposed network processes the images that show the original true state of the sperm,and the accurate segmentation results are very helpful for the doctor’s clinical diagnosis.

关 键 词：人类精子头部分割 精子畸形 深度学习 残差结构 混合扩张卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]