基于卷積神經(jīng)網(wǎng)絡(luò)的醫(yī)學(xué)圖像分類的研究
發(fā)布時(shí)間：2018-11-28 16:36
【摘要】：現(xiàn)代醫(yī)院每天都會(huì)產(chǎn)出大量的醫(yī)學(xué)圖像,這些醫(yī)學(xué)圖像數(shù)據(jù)都會(huì)被傳入醫(yī)學(xué)云影像中心。由于云影像中心中的醫(yī)學(xué)圖像是雜亂無(wú)章的,所以在這些圖像數(shù)據(jù)應(yīng)用到實(shí)際的挖掘工作之前首先應(yīng)對(duì)其進(jìn)行清洗,得出適合挖掘的醫(yī)學(xué)圖像數(shù)據(jù)。隨著數(shù)據(jù)挖掘概念的提出,很多優(yōu)秀的數(shù)據(jù)挖掘方法由于其強(qiáng)大的分類能力也被應(yīng)用到醫(yī)學(xué)圖像分類中,但是其中大部分都是先對(duì)其進(jìn)行特征提取,即提取醫(yī)學(xué)圖像數(shù)據(jù)的統(tǒng)計(jì)學(xué)特征進(jìn)而在得到的特征數(shù)據(jù)集上對(duì)其進(jìn)行分析研究,利用一些比較好的統(tǒng)計(jì)學(xué)習(xí)方法進(jìn)行分類。而近幾年隨著深度學(xué)習(xí)方法的研究取得重大的進(jìn)展,一些較好的深度學(xué)習(xí)方法也自然而然的應(yīng)用到醫(yī)學(xué)圖像分析領(lǐng)域,其中的典型代表就是卷積神經(jīng)網(wǎng)絡(luò)。利用卷積神經(jīng)網(wǎng)絡(luò)對(duì)圖像進(jìn)行分類不僅提高了圖像分類的準(zhǔn)確率,而且還可以省去傳統(tǒng)統(tǒng)計(jì)學(xué)習(xí)方法特征工程部分,大大提高了圖像分類的效率。因此本文主要對(duì)利用卷積神經(jīng)網(wǎng)絡(luò)對(duì)醫(yī)學(xué)圖像分類的方法以及利用卷積神經(jīng)網(wǎng)絡(luò)提取圖像特征進(jìn)行了研究。本文首先回顧了國(guó)內(nèi)外在圖像分類領(lǐng)域的研究現(xiàn)狀,接下來(lái)介紹了傳統(tǒng)的統(tǒng)計(jì)學(xué)習(xí)方法中應(yīng)用在醫(yī)學(xué)圖像分類領(lǐng)域較為優(yōu)越的詞袋模型以及圖像領(lǐng)域表征性較強(qiáng)的SIFT特征,并且詳細(xì)介紹詞袋模型的基礎(chǔ)理論和應(yīng)用領(lǐng)域以及SIFT的基礎(chǔ)原理和應(yīng)用。然后講述了深度學(xué)習(xí)以及卷積神經(jīng)網(wǎng)絡(luò)的基本理論以及其在圖像分類領(lǐng)域的應(yīng)用。最后針對(duì)傳統(tǒng)統(tǒng)計(jì)學(xué)習(xí)的分類方法和卷積神經(jīng)網(wǎng)絡(luò)方法各自的特點(diǎn),進(jìn)行了取其各自所長(zhǎng)將兩者結(jié)合起來(lái)的探索。在最后的通過(guò)實(shí)驗(yàn)結(jié)果進(jìn)行驗(yàn)證部分,我們首先對(duì)利用卷積神經(jīng)網(wǎng)絡(luò)與利用詞袋模型對(duì)醫(yī)學(xué)圖像分類的實(shí)驗(yàn)結(jié)果進(jìn)行對(duì)比分析,說(shuō)明基于深度學(xué)習(xí)方法的卷積神經(jīng)網(wǎng)絡(luò)在醫(yī)學(xué)圖像分析方面不僅可以省去人工特征工程的工作,而且分類效果比傳統(tǒng)統(tǒng)計(jì)學(xué)習(xí)方法更好;然后通過(guò)將卷積神經(jīng)網(wǎng)絡(luò)自動(dòng)提取特征以及傳統(tǒng)分類方法的分類能力相結(jié)合進(jìn)而對(duì)醫(yī)學(xué)圖像進(jìn)行分類與前兩種分類方法進(jìn)行實(shí)驗(yàn)分析比較,驗(yàn)證了將基于深度學(xué)習(xí)的卷積神經(jīng)網(wǎng)絡(luò)與傳統(tǒng)統(tǒng)計(jì)學(xué)習(xí)方法相結(jié)合的分類方法在醫(yī)學(xué)圖像分類領(lǐng)域較有很好的優(yōu)越性。
[Abstract]:Modern hospitals produce a large number of medical images every day, which are passed into the medical cloud image center. Because the medical images in the cloud image center are chaotic, the medical image data suitable for mining should be cleaned first before they are applied to the actual mining work. With the development of the concept of data mining, many excellent data mining methods have been applied to medical image classification because of their powerful classification ability. That is to extract the statistical features of medical image data and then analyze them on the acquired feature data set and classify them by using some better statistical learning methods. In recent years, with the great progress in the research of deep learning methods, some better depth learning methods are naturally applied to the field of medical image analysis, the typical representative of which is convolution neural network. Using convolution neural network to classify images not only improves the accuracy of image classification but also saves the feature engineering of traditional statistical learning method and greatly improves the efficiency of image classification. In this paper, the methods of medical image classification using convolution neural network and image feature extraction by convolution neural network are studied in this paper. Firstly, this paper reviews the research status of image classification at home and abroad, then introduces the word bag model which is used in the field of medical image classification in traditional statistical learning methods and the SIFT feature with strong representativeness in image field. The basic theory and application field of word bag model and the basic principle and application of SIFT are introduced in detail. Then the basic theory of deep learning and convolution neural network and its application in image classification are described. Finally, according to the characteristics of the traditional statistical learning classification method and convolution neural network method, the author explores the combination of the two methods. In the last part, we compare and analyze the experimental results of medical image classification using convolution neural network and word bag model. It shows that the convolutional neural network based on the deep learning method can not only save the work of artificial feature engineering, but also has better classification effect than the traditional statistical learning method in medical image analysis. Then, by combining the automatic feature extraction of convolution neural network and the classification ability of traditional classification methods, the medical image classification is analyzed and compared with the first two classification methods. It is verified that the classification method which combines convolution neural network based on deep learning with traditional statistical learning method has better superiority in the field of medical image classification.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP183
本文編號(hào)：2363537
[Abstract]:Modern hospitals produce a large number of medical images every day, which are passed into the medical cloud image center. Because the medical images in the cloud image center are chaotic, the medical image data suitable for mining should be cleaned first before they are applied to the actual mining work. With the development of the concept of data mining, many excellent data mining methods have been applied to medical image classification because of their powerful classification ability. That is to extract the statistical features of medical image data and then analyze them on the acquired feature data set and classify them by using some better statistical learning methods. In recent years, with the great progress in the research of deep learning methods, some better depth learning methods are naturally applied to the field of medical image analysis, the typical representative of which is convolution neural network. Using convolution neural network to classify images not only improves the accuracy of image classification but also saves the feature engineering of traditional statistical learning method and greatly improves the efficiency of image classification. In this paper, the methods of medical image classification using convolution neural network and image feature extraction by convolution neural network are studied in this paper. Firstly, this paper reviews the research status of image classification at home and abroad, then introduces the word bag model which is used in the field of medical image classification in traditional statistical learning methods and the SIFT feature with strong representativeness in image field. The basic theory and application field of word bag model and the basic principle and application of SIFT are introduced in detail. Then the basic theory of deep learning and convolution neural network and its application in image classification are described. Finally, according to the characteristics of the traditional statistical learning classification method and convolution neural network method, the author explores the combination of the two methods. In the last part, we compare and analyze the experimental results of medical image classification using convolution neural network and word bag model. It shows that the convolutional neural network based on the deep learning method can not only save the work of artificial feature engineering, but also has better classification effect than the traditional statistical learning method in medical image analysis. Then, by combining the automatic feature extraction of convolution neural network and the classification ability of traditional classification methods, the medical image classification is analyzed and compared with the first two classification methods. It is verified that the classification method which combines convolution neural network based on deep learning with traditional statistical learning method has better superiority in the field of medical image classification.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP183
【參考文獻(xiàn)】
相關(guān)期刊論文 前2條
1 易堯華,梅天燦,秦前清,龔健雅;高光譜影像中人工目標(biāo)非監(jiān)督提取的投影尋蹤方法[J];測(cè)繪通報(bào);2004年02期
2 呂硯山,趙正琦;BP神經(jīng)網(wǎng)絡(luò)的優(yōu)化及應(yīng)用研究[J];北京化工大學(xué)學(xué)報(bào)(自然科學(xué)版);2001年01期
相關(guān)碩士學(xué)位論文 前1條
1 劉岳;基于詞袋模型的醫(yī)學(xué)影像分類的研究與實(shí)現(xiàn)[D];東北大學(xué);2012年
，本文編號(hào)：2363537
本文鏈接：http://www.sikaile.net/kejilunwen/ruanjiangongchenglunwen/2363537.html
最近更新
教材專著
