臺灣博碩士論文加值系統

本論文的目的在於建立一個辨識甲狀腺超音波影像中各種常見病變組織的系統，本系統分為三個部份，第一個部份首先使用16×16的區塊對於不同大小的病理組織取出其紋理特徵，接下來使用Relief-F演算去選出對於各個組織具有代表性的特徵集合，第二個部份使用兩個監督式類神經網路分類器(倒傳遞類神經網路與徑向基底函數類神經網路)，並且分別使用單一分類器與多分類器，單一分類器直接對六種病理組織進行分類，而多分類器是將各個相似性高的組織在第一個分類器歸為三類，第二、三個分類器在將其細分出來，第三部份是使用已訓練好的分類器直接對影像進行辨識各種病理組織。

結果顯示使用單一分類器容易發生錯誤，因此先將相似性高的組織歸為一類，並且採用階層式分類器的架構，在第一級的分類器中先分三類，接著在第二級的兩個分類器再細分為六個類別。首先用Relief-F演算法針對三個分類器分別找出適當的部分類別的特徵集合，再分別經由倒傳遞類神經網路與徑向基底函數類神經網路作辨識。結果顯示階層式分類器的確比單一分類器的結果好，而以三個倒傳遞類神經網路分類器的結果為最好，整體的辨識率為84.5%。本研究最後的影像辨識結果均由專業醫師確認，證實本系統確實有助於醫生在臨床甲狀腺組織辨識的診斷，並對於往後甲狀腺病理組織的研究有相當的幫助。

The purpose of this thesis is to build a system which is capable of pathological tissues in the ultrasonic images of thyroid gland. This system is divided into three parts. In the first part, 16×16 block image are used to extracted textural features, and then a Relief-F algorithm is used to select representative features for different tissue type. In the part, two supervising neural network classifiers(back-propagation and radial basis function neural networks), are used to realize a single classifier and a hierarchical classifier architectures. The single classifier architecture distinguishes six different tissues directly. On the other side, the hierarchical classifier architecture using a single classifier to separate the six tissue types into three groups in the first level. And then use two other classifiers to further divide them into six types. In the third part, the trained classifiers are employed to discriminate six tissues ine the ultrasonic image.

The experimental results show that single classifier is apt to make false decision. Therefore , we adopt the hierarchical classifier architecture in our system. After the representative features have been selected by the Relief-F algorithm. The back-propagation and radial basis function neural networks are used in hierarchical architecture to classify six tissue types. The result shows that the hierarchical architecture has better accuracy than single classifier. The best result is obtained by using BPNN classifier to achieve 84.5% of accuracy. The tissue recognition results are confirmed by professional doctors. The results discriminate that help the clinical professional in the diagnosis of thyroid gland tissues, and prove the capability of the proposed scheme in future research on.

第一章 緒論 1
1.1: 研究背景與動機 1
1.1.1: 簡介甲狀腺弁?1
1.1.2: 簡介超音波影像 2
1.1.3: 甲狀腺超音波檢查 3
1.2: 相關文獻回顧 3
1.3: 研究目的 4
1.4: 論文架構 5
第二章 紋理特徵擷取法 6
2.1: 空間性灰階共生紋理特徵 6
2.2: Law紋理能量特徵 8
2.3: 灰階長度紋理特徵 9
2.4: 區域性灰階相依紋理特徵 11
2.5: 區域性灰階差異紋理特徵 13
2.6: 基於區域性的傅立葉特徵 15
2.7: 小波特徵 16
第三章 特徵選取法 17
3.1: Relief-F特徵選擇演算法 17
3.2: 主成分分析 19
第四章 類神經網路分類器 23
4.1: 倒傳遞類神經網路 23
4.2: 徑向基底函數類神經網路 28
第五章 實驗方法與結果 33
5.1: 本研究所使用的影像資料與實驗環境 33
5.2: 辨識系統的架構與流程 33
5.3: 實驗結果與討論 39
5.3.1: 類神經網路分類器的結果 39
5.3.2: 影像辨識的結果 41
5.3.3: 結果與討論 48
第六章 結論與未來發展 50
6.1: 結論 50
6.2: 未來發展 52
參考文獻 53

[1]. Lale OZYILMZ , Tulay ULDIRIM , “diagnosis of thyroid disease using artificial neural network methods”Proceedings of the 9th International Conference on Neural Information Processing (ICONIP'OZ) , Vol. 4

[2]. Leung, C.C.; Chan, F.H.Y.; Lam, K.Y.; Kwok, P.C.K.; Chen, W.F.; , “Thyroid cancer cells boundary location by a fuzzy edge detection method” , Pattern Recognition , 2000. Processings. 15th international Conference on Publication Publication Date: 3-7 Sept. 2000 Volume: 4

[3]. Maroulis, D.E.; Savelonas, M.A.; Karkanis, S.A.; Iakovidis, D.K.; Dimitropoulos, N.; , “Computer-aided thyroid nodule detection in ultrasound images” , Computer-based Medical Systems , 2005. Processings. 18th IEEE Symposium on Publication Date: 23-24 June 2005 On page(s): 271 – 276

[4] Ablameyko, S.; Kirillov, V.; Lagunovsky, D.; Patsko, O.; Paramonova, N.; Petrou, M.; Tchij, O.; , “From cell image segmentation to differential diagnosis of thyroid cancer” , Pattern Recognition , 2002. Processings. 16th international Conference on Publication Date: 11-15 Aug. 2002 Volume: 1 On page(s): 763 - 766 vol.1

[5] Robert . M . Haralick , Linda .G. Shapiro , “computer and robot vision”, volume 1 Wesley 1992

[6] Mougiakakou, S.Gr.; Golemati, S.; Gousias, I.; Nikita, K.S.; Nicolaides, A.N.;” Computer-aided diagnosis of carotid atherosclerosis using laws' texture features and a hybrid trained neural network” , Engineering in Medical and Biology Socity,2003. Processsings of the 25th Annual International Conference of IEEE Publication Date: 17-21 Sept. 2003
Volume: 2 On page(s): 1248 - 1251 Vol.2 Number of Pages: 4295

[7] Chung-Ming Wu; Yung-Chang Chen; Kai-Sheng Hsieh;” Texture features for classification of ultrasonic liver images”, Medicak Imaging , IEEE transactions on Publication Date: June 1992
Volume: 11 , Issue: 2 On page(s): 141 - 152

[8]Mary M.Galloway , “Testure Analysis Using Gray Level Run Lengths”, computer graphics and image processing volume 4 , 172-179(1975)

[9]Chengjun Sun , william G . Wee , “Neighboring Gray Level Dependence Matrix for Texture classification”, computer vision , graphics and image processing volume 23 341-352

[10] Moses Amadasun.,Robert King , “texural features corresponding to texural properties” , IEEE transactions on systems, man, and cybernetics, VOL. 19, NO. 5 ,

[11] Feng Zhou; Ju Fu Feng; Qing Yun Shi; , “Texture feature based on local Fourier transform”, Image Processing,2001.Proceedings , 2001 . 2001. International Conference on Publication Date: 7-10 Oct. 2001 Volume: 2 On page(s): 610 - 613 vol.2

[12] S. Arivazhagan and L. Ganesan , “Texture segmentation using wavelet transform ”, Pattern Recognition Letters, Volume 24, Issue 16, December 2003, Pages 3197-3203

[13] S. Arivazhagan and L. Ganesan , “Texture classification using wavelet transform ”, Pattern Recognition Letters, Volume 24, Issues 9-10, June 2003, Pages 1513-1521

[14]施泰宇著 , ”一個使用紋理特徵及羅吉斯迴歸法分辨超音波乳房腫塊良惡性的方法”,中正大學電機所碩士論文, 2005年7月

[15] Marko Robnik-ˇ Sikonja , Igor Kononenko , “Theoretical and Empirical Analysis of ReliefF and RReliefF”, Machine Learning Journal (2003) 53:23-69

[16] Lindsay I Smith , “A tutorial on Principal Components Analysis ”

[17] Chen, S.; Cowan, C.F.N.; Grant, P.M.;, “Orthogonal least squares learning algorithm for radial basis function networks”, Neural Networks , IEEE Transactions on Publication Date: March 1991 Volume: 2 , Issue: 2 On page(s): 302 - 309