基于張量分析的腦部醫(yī)學圖像識別
發(fā)布時間：2018-10-11 19:10
【摘要】：隨著計算機領域的不斷創(chuàng)新和發(fā)展,醫(yī)學成像技術也在不斷地提高。醫(yī)學圖像識別作為診療病情的關鍵技術手段,在醫(yī)學研究和臨床試驗方面需求龐大,發(fā)展迅速。作為是人體內最復雜也是最重要的器官,關于大腦的相關醫(yī)學研究非常依賴醫(yī)學圖像識別。腦部醫(yī)學圖像識別的基礎包括成像技術,腦部結構特征提取和分類等,因此腦部醫(yī)學圖像識別作為多學科交叉的領域,具有非常高的研究價值和意義。由于腦部醫(yī)學圖像的數據相對于一般圖像數據而言,其數據本質是三維空間結構的體素數據,傳統上對圖像特征提取和分類算法的研究習慣于從向量的角度出發(fā)來考慮問題,然而這樣卻忽略了圖像結構上的特點,從而破壞了圖像的高階信息,這種圖像高階結構信息的損失不僅導致了圖像識別率的損失,還造成了居高不下的計算復雜度。如何能夠在保存空間結構信息的同時對高維空間結構數據進行特征提取和分類,成為現如今醫(yī)學圖像識別領域的一個問題。本文結合高維空間結構數據,以數據張量化為研究重點,幫助腦部醫(yī)學圖像數據進行整體的特征提取提出改進,結合基于循環(huán)卷積的張量模型以及張量主成分分析(簡稱TPCA)的理論知識以及相關的基本概念,能夠廣泛的對各種醫(yī)學圖像數據進行張量數據分析。同時,本文以時下流行的SBD數據集,結合基于張量模型的主成分分析對數據張量化后的數據集進行提取圖像特征,用所提取的圖像特征對SBD醫(yī)學圖像進行識別分類,驗證并分析最后的處理結果。實驗證明,數據張量化方法在提取腦部醫(yī)學圖像這樣的高維空間結構數據的圖像特征方面具有良好的適用性,數據張量化后基于張量模型的算法比基于向量的算法要好。
[Abstract]:With the innovation and development of computer field, medical imaging technology is improving. Medical image recognition, as a key technique in diagnosis and treatment, has a huge demand for medical research and clinical trials. As one of the most complex and important organs in the human body, medical research on the brain relies heavily on medical image recognition. The basis of brain medical image recognition includes imaging technology, brain structure feature extraction and classification. Therefore, as a multidisciplinary field, brain medical image recognition is of great value and significance. Because the data of brain medical image is essentially voxel data with three-dimensional structure compared with general image data, the traditional research on image feature extraction and classification algorithm is used to consider the problem from the point of view of vector. However, the characteristics of the image structure are ignored and the higher-order information of the image is destroyed. The loss of the higher-order structure information of the image not only leads to the loss of the recognition rate of the image, but also results in a high computational complexity. How to extract and classify high-dimensional spatial structure data while preserving spatial structure information has become a problem in the field of medical image recognition nowadays. Based on high-dimensional spatial structure data and data tensor quantization, this paper proposes an improved method to help the whole feature extraction of brain medical image data. Combined with Zhang Liang model based on cyclic convolution and the theoretical knowledge and related basic concepts of Zhang Liang principal component analysis (TPCA), it can be widely used to analyze all kinds of medical image data. At the same time, using the popular SBD data set and principal component analysis (PCA) based on Zhang Liang model, this paper extracts the image feature of the tensor data set, and classifies the SBD medical image with the extracted image feature. Verify and analyze the final processing results. Experiments show that data tensor quantization method has good applicability in extracting image features of high-dimensional spatial structure data such as brain medical image. The algorithm based on Zhang Liang model after data tensor is better than that based on vector.
【學位授予單位】：中原工學院
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41
本文編號：2264940
[Abstract]:With the innovation and development of computer field, medical imaging technology is improving. Medical image recognition, as a key technique in diagnosis and treatment, has a huge demand for medical research and clinical trials. As one of the most complex and important organs in the human body, medical research on the brain relies heavily on medical image recognition. The basis of brain medical image recognition includes imaging technology, brain structure feature extraction and classification. Therefore, as a multidisciplinary field, brain medical image recognition is of great value and significance. Because the data of brain medical image is essentially voxel data with three-dimensional structure compared with general image data, the traditional research on image feature extraction and classification algorithm is used to consider the problem from the point of view of vector. However, the characteristics of the image structure are ignored and the higher-order information of the image is destroyed. The loss of the higher-order structure information of the image not only leads to the loss of the recognition rate of the image, but also results in a high computational complexity. How to extract and classify high-dimensional spatial structure data while preserving spatial structure information has become a problem in the field of medical image recognition nowadays. Based on high-dimensional spatial structure data and data tensor quantization, this paper proposes an improved method to help the whole feature extraction of brain medical image data. Combined with Zhang Liang model based on cyclic convolution and the theoretical knowledge and related basic concepts of Zhang Liang principal component analysis (TPCA), it can be widely used to analyze all kinds of medical image data. At the same time, using the popular SBD data set and principal component analysis (PCA) based on Zhang Liang model, this paper extracts the image feature of the tensor data set, and classifies the SBD medical image with the extracted image feature. Verify and analyze the final processing results. Experiments show that data tensor quantization method has good applicability in extracting image features of high-dimensional spatial structure data such as brain medical image. The algorithm based on Zhang Liang model after data tensor is better than that based on vector.
【學位授予單位】：中原工學院
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41
【參考文獻】
相關期刊論文 前4條
1 廖亮;楊程凱;;基于圖像邊緣位移的有監(jiān)督的稀疏表示分類方法[J];成都大學學報(自然科學版);2016年04期
2 黃京;郭水霞;;基于醫(yī)學影像數據的大腦功能連接方法研究[J];應用概率統計;2014年02期
3 陳玉林;吳力源;胡曉云;陸艷艷;翟曉東;夏震;陳靜雯;方向明;陳宏偉;;刀鋒偽影校正技術在顱腦高場MRI中的臨床應用價值[J];中國現代醫(yī)學雜志;2011年23期
4 孟麗艷;;功能磁共振成像技術及其在腦科學研究中的應用[J];物理通報;2001年06期
，本文編號：2264940
本文鏈接：http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2264940.html
