大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【计算机科学】【2016】基于卷积神经网络的前列腺癌分类

已有 256 次阅读 2020-7-22 17:01 |系统分类:科研笔记|文章来源:转载

本文为瑞典隆德大学(作者:Anna Gummeson)的硕士论文,共54页。

 

2012年,前列腺癌已经成为第二常见的男性癌症诊断病例。病理学家对前列腺活检进行目镜检查、确认诊断,并根据Gleason分级系统对标本进行分类。本文的主要目的是利用卷积神经网络(CNN)实现分类的自动化。随着卷积神经网络的引入,在模式识别领域的应用越来越广。传统手工特征设计和提取的分类方法与让计算机自动决定哪些特征是重要的有很大的不同,新的分类方法是由CNN实现的。这与基准图像集的开创性成果一起,使CNN成为模式识别中一种很好的方法。在本论文中,我们利用具有动量的随机梯度下降,从零开始训练具有小卷积滤波器的CNNCNN的错误率为7.3%,明显优于以前使用相同数据集的工作。尽管数据集很小,但仍然取得了很好的结果,因此CNN是解决这一问题的一种很有前途的方法。

 

In 2012 prostate cancer was the second mostcommon cancer diagnose for men. The diagnosis is confirmed by pathologistsdoing ocular inspection of prostate biopsies and the specimens are classifiedaccording to the Gleason grading system. The main goal of this thesis is toautomate the classification using Convolutional Neural Networks (CNN). With theintroduction of Convolutional Neural Networks the field of pattern recognitionbroadened. The classical way of designing and extracting hand-made features forclassification is substantially different to letting the computer itself decidewhich features are of importance, the new approach was enabled by CNNs. Thistogether with groundbreaking results on benchmark image sets has made CNNs awell-used method in pattern recognition. In this thesis a CNN with smallconvolutional filters has been trained from scratch using stochastic gradientdescent with momentum. The error rate for the CNN is 7.3%, which issignificantly better than previous works using the same data set. Since goodresults were obtained even though the data set were rather small, theconclusion is that CNNs are a promising method for this problem.

 

1. 引言

2. 人工神经网络理论

3. 格里森评分

4. 数据集与相关材料

5. 研究方法

6. 检测特征的可视化

7. 结果

8. 讨论


更多精彩文章请关注公众号:205328s611i1aqxbbgxv19.jpg




http://wap.sciencenet.cn/blog-69686-1243133.html

上一篇:[转载]【信息技术】【2003.03】视觉监控应用中人体跟踪算法的设计与实现
下一篇:[转载]【统计学】【2018.05】【含源码】时间序列:以密度预测评价方法为中心的预测与评价方法

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2020-10-21 18:08

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部