本篇論文探討指骨X光，經過切割處理，尋找影像特徵參數和預估骨齡等流程，來預估其骨頭的年齡。在切割處理中，使用鄰近差異處理法來找到骨頭的部位；而在抽取特徵參數上，觀察指關節的的形狀和面積，依據不同指節抽取10個或8個特徵參數，作為之後的特徵分析預估骨齡。為方便處理，在分類之前將10 個特徵參數先經過PCA處理，篩選特徵成8個參數，用來提高分類正確率。
對一張左手的X光影像分別圈出食指、中指、無名指，每一個手指取出三個指節，共九個指節。將這九個指節，分別進入綜合骨齡判讀系統後，會得到九個預估的結果，再由這些結果取眾數後跟真實年齡做比較，達到自動化判讀骨齡。
進行分類之前，由於會先比較過多重分類和二分法，發現多重分類的正確率較低，所以選擇二分法來使用。接著結合Matlab 的Tree 函式進行分類，會發現所建構的樹狀圖深度各不相同。為了改善這個問題，參考完全二元樹狀圖的結構，設計了一個修正完全二元樹狀圖，因結果為16種類別用完全二元樹狀圖得到分類深度為4的樹狀圖，並且比較九個指節分開執行或一起執行的差異，結果發現一起執行的正確率較低，所以選擇將九個指節分開來做。
在分類器選擇上，採用KNN分類器進行估測，在沒有經過PCA篩選特徵的分析中，藉由樹狀圖通過分類深度為4的樹狀圖往下分類，因為輸入的資料為1歲到16歲每一歲為一個間隔，所以在樹狀圖的分類把這十六種類別，用一次分兩類的方式來分類，一共分四個切割回合就能夠處理完成。並得到預估的歲數。
參考Leave-one-out ("一次挑一個"辨識率)想法，在樹狀圖的分類節點當中，用來預估測試影像的骨齡。全部樣本為男生女生各160張原始X光影像，借用Leave-one-out的觀念一次挑一個出來當做測試資料，剩下的159筆資料當做訓練資料，來看這張測試資料的分類準確率。

This research investigates the hand radiogram with knuckles which were processed by the segmentation, feature extraction and bone age assessment. In the segmentation phase, we apply the difference in strength to segment the region of epiphyseal and metaphyseal. For the feature extraction, we extract 10 and 8 features according to the different knuckles. To simplify the analysis, we introduce PCA technique to reduce the dimensionality from 10 features to 8 features, for the improvement of the classification accuracy.
In the beginning, a left hand radiogram with its index finger, middle finger and ring finger, corresponding to the distal, middle and proximal phalanges, were located and extracted. Next, the nine knuckles were estimated individually by bone age assessment to generate nine ages. The nine ages were involved in the comparison with the chronological age for building an automatic bone age assessment.
Since the classification performance of the binary decision tree is better than the multiple decision tree, the binary decision tree was chosen in the following analysis. Subsequently, the tree function of Matlab was involved and modified to be able to produce the same layers for each branch. In our study, a modified binary decision tree was evaluated their difference between the two analyses of each 9 knuckles and the integration of 9 knuckles. The results were shown the lower accuracy for the integration of 9 knuckles, so we decide to analyze the 9 knuckles separately.
The KNN classification was selected to analyze the PCA features for each knuckle. Then, the binary decision tree was set 4 layers and the sum of 16 outcomes was obtained to correspond with the chronological ages ranged from 1 to 16.
The Leave-one-out cross validation is a useful test for the stability and accuracy, and we integrate the Leave-one-out with binary decision tree for assessing the bone age. The radiograms include 160 boys and 160 girls.

[1] Greulich W. W., Pyle S. I., Radiographic Atlas of Skeletal Development the and Wrist, 2nd edition, Stanford, CA: Stanford University Press, 1959.
[2] Tanner J. M., Whitehouse R. H., Marshall W. A., Assessment of SkeletalMaturity and Prediction of Adult Height (TW2 Method), New York: AcademicPress, 1975.
[3] 蔣新土, 呂岳, "基於改進的加權 Hausdorff 距離的圖像匹配" 計算機應用研究, 2007 年第 24 卷第 4 期, 2007/4.
[4] Todd T.W., St. Louis, C.V. Mosby Co, Atlas of Skeletal Maturation (Hand),1937.
[5] Bolboaca S., Jäntschi L., Achimas Cadariu A., "PHP and MySQL MedicalApplication Based on Tanner Whitehouse Standard," Proceeding of theUNITECH'03 International Scientific Conference, Vol. 2003, pp.304-308, 2003.
[6] Milner G.R., Levick R.K., Kay R., “Assessment of bone age: a comparison of the Greulich and Pyle, and the Tanner and Whitehouse methods,” Clinical Radiology, vol. 37, pp.119-121, Mar 1986.
[7] Xu C.Y., Prince J.L., "Snakes, shapes, and gradient vector flow,” IEEE Transactions on Image Processing, vol. 7, pp. 359-369, Mar 1998.
[8] 黃耀仟, 一個自動指骨影像切割系統之研製, 國立清華大學電機系碩士論文, 2010/7.
[9] Liu C.C. and Chung P.C., “Objects extraction algorithm of color image using adaptive forecasting” IJICIC,vol.7, pp.5771-5781, 2011.
[10] Lin H.H., Shu S.G., Kuo S.W., Wang C.H., Chan Y.P., Yu S.S., “Alpha-gamma equalization-enhanced hand radiographic image segmentation scheme,” Optical Engineering, vol.48, pp.107001-9, 2009.
[11] Pappas T.N., "An adaptive clustering-algorithm for image segmentation,” IEEE Transactions on Signal Processing, vol. 40, pp. 901-914, Apr 1992.
[12] Li C.M., Xu C.Y., Gui C.F., Martin D. Fox, " distance regularized level set evolution and its application to image segmentation,” IEEE Transactions on Image Processing, vol. 19, pp. 3243-3254, December 2010.
[13] Nassir S., Jordan “Image Segmentation Based on Watershed and Edge Detection Techniques” Int. Arab J. Inf. Technol., vol.3(2), pp.104-110, 2006.
[14] Lin H.H., Shu S.G., Lin Y. H., Yu S.S., “Bone age cluster assessment and feature clustering analysis based on phalangeal image rough segmentation,” Pattern Recognition, vol.45, pp.322-332, 2012.
[15] Alkin M.C., and Christie C. A., “An evaluation theory tree,” An evaluation theory tree. Evaluation roots: Tracing theorists’ views and influences, pp.12-65, 2004.
[16] Ye J., Janardan R., & Li Q., "Two-dimensional linear discriminant analysis," In Advances in Neural Information Processing Systems, pp. 1569-1576, 2004.
[17] Loh W.Y.,"Classification and Regression Tree Methods’ In Encyclopedia of Statistics in Quality and Reliability," Ruggeri, Kenett and Faltin (eds.), 315–323, Wiley, 2008.
[18] Dučinskas K., & Šaltytė, J., “Quadratic Discriminant Analysis of Spatially Correlated Data,” Nonlinear Analysis, vol.6(2), pp.15-28, 2001.
[19] Refaeilzadeh P., Tang L., Liu H., “Cross-Validation,” Encyclopedia of Database Systems, 5.
[20] Lowd D., Domingos P., ”Naive Bayes Models for Probability Estimation” , In Proceedings of the 22nd international conference on Machine learning, pp. 529-536, 2005.
[21] Nayak J., Routray A.R., & Adhikary M.C., “Classification of Trees by Pattern Recognition,” Classification of Trees by Pattern Recognition. International Journal of Computer Applications, vol.19, 2011.
[22] Buja A., & Lee Y.S., "Data mining criteria for tree-based regression and classification," In Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 27-36, 2011.
[23] Chen X., Wang M., & Zhang H., "The use of classification trees for bioinformatics," Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, vol.1, pp.55-63, 2011.
[24] Anyanwu M.N., & Shiva S.G., "Comparative analysis of serial decision tree classification algorithms," International Journal of Computer Science and Security, vol.3, pp.230-240, 2009.
[25] Song Y., Huang J., Zhou D., Zha, H., & Giles C. L. Iknn: Informative k-nearest neighbor pattern classification. In Knowledge Discovery in Databases: PKDD 2007 (pp. 248-264). Springer Berlin Heidelberg, 2007.
[26] 邱繼鴻, 基於指骨X 光影像之電腦輔助骨齡判讀研究, 國立清華大學電機系碩士論文, 2011/7.
[27] 楊嘉棟, α-γ增強的前處理對指骨影像切割的影響, 國立清華大學電機系碩士論文, 2012/7.
[28] Statistics Toolbox in Matlab with KNN ‘http://www.mathworks.com/help/bioinfo/ref/knnclassify.html’
[29] Tax D.M., & Duin R.P., "Using two-class classifiers for multiclass classification," In Pattern Recognition, 2002. Proceedings. 16th International Conference on IEEE, vol. 2, pp. 124-127, 2002.
[30] Lee J.S., & Oh I.S., "Binary Classification Trees for Multi-class Classification Problems," In ICDAR, vol. 3, pp. 770-774, 2003.
[31] Li C., Xu C., Gui C., & Fox M.D., "Distance regularized level set evolution and its application to image segmentation," Image Processing, IEEE Transactions on, vol.19(12), pp.3243-3254, 2010.