臺灣博碩士論文加值系統

本篇論文利用腕骨面積比特徵提出一個全自動化的骨齡評估研究，內容包含前處理、特徵抽取和骨齡判讀。前處理步驟包含切割左手影像、去除背景和腕骨感興趣區域(Region of Interest, ROI)萃取。利用X光影像底部中點為圓心進行逆時針掃描，計算各角度掃描線上非零灰階值數量，以最小數量的掃描線找出左手影像右邊界來切割左手影像。透過左手影像像素灰階值的平均當作threshold將左手影像二值化，利用保留最大群集可將不需要的標籤、右手和背景去除。透過尺骨跟橈骨資訊可萃取可靠的腕骨ROI。在特徵抽取部分，藉由像素的座標位置之灰階值累積分配函數當作權重來抑制不必要的軟組織和紋理，利用膨脹、填滿封閉區間和侵蝕萃取腕骨。以腕骨面積比為特徵進行骨齡判讀，依各骨齡的面積比特性分為三個時期，依照不同時期使用不同的方法來進行階層式的骨齡判讀。判讀出來的骨齡除了可以輔助醫師在臨床上的診斷，其腕骨ROI影像亦可提供醫師和放射學研究者來進行Tanner-Whitehouse (TW)檢驗法。我們所採用的資料庫影像超過3500張，其前處理的成功率達到將近九成，整體骨齡判讀誤差1.5歲以內成功率接近八成。而在骨齡較小的時期，骨齡判讀誤差1.5歲以內成功率接近九成，其實驗結果符合臨床上的研究，利用腕骨來進行骨齡判讀的年齡通常比較小，當超過9-12歲之後的利用指骨資訊來進行骨齡判讀較為可靠。

In this thesis, a fully automatic bone age development using the area ratio of carpals has been proposed, including preprocess, feature extraction, bone age assessment. Preprocess consists of cropping left hand image, removing background, and extraction of carpals' region of interest (ROI). Using radiograms bottom center as the center of a circle to do counter clock scan. Calculate the magnitude of nonzero gray level on scan lines with all angles. Find the right boundary of left hand image by smallest magnitude of scan lines to cropping left hand image. Transform the left image to binary through using mean value of left hand image pixel’s gray level as the threshold. Eliminate the unneeded label, right hand, and background by reserving the maximum group. Through information of ulna and radius, we can extract reliable carpal’s ROI. In feature extraction section, restrain unneeded soft tissue and texture by using gray level cumulative distribution function of pixel location as the weight. By dilating, filling airtight region, and eroding to extract carpals. Use the area ratio of carpals as the characteristic to accomplish the bone age assessment. Separate the carpal’s area ratio to three stages according to characteristics of the bone ages. Use different methods to accomplish classified bone age assessment at different stages. In addition to assist physician's clinical diagnosis with the assessed bone age, the carpals ROI image can provide physician and radiologist with Tanner-Whitehouse (TW) method. The dataset images which we adopted are beyond 3500. The successful rate of preprocess reaches almost ninety percent. The successful rate which the error of bone age is under 1.5 years old reaches nearly eighty percent. At stage of the youngest bone age, the successful rate which the error of bone age is under 1.5 years old reaches nearly ninety percent. The experiment result conforms to clinical research. The age is usually younger by accomplishing bone age assessment with carpals. After the age is beyond 9-12, accomplishing bone age assessment with phalanx is more reliable.

摘要 i
Abstract ii
致謝 iv
目錄 v
圖目錄 vi
表目錄 viii
第一章 簡介 1
1.1 前言 1
1.2 研究動機 2
1.3 研究目的 3
1.4 論文架構 3
第二章 腕骨前處理 4
2.2 去除背景 7
2.3 腕骨ROI萃取 10
第三章 腕骨特徵抽取與骨齡判讀 16
3.1 去除軟組織 16
3.2 腕骨特徵抽取 19
3.3 骨齡判讀 21
第四章 實驗結果與討論 25
4.1 資料庫 25
4.2 前處理成功率 26
4.3 特徵處理驗證(Bone Age) 27
4 .4 討論 33
第五章 結論 35
5.1 結論 35
5.2 未來展望 36
參考文獻 37

[1] Greulich WW, Pyle SI (1959) Radiographic atlas of skeletal development of the hand and wrist. Stanford University Press, Stanford, California.
[2] Mora S, Boechat MI, Pietka E, Huang HK, Gilsanz V (2001) Skeletal age determinations in children of European and African descent: applicability of the Greulich and Pyle standards. Pedia Res 50: 624-628.
[3] Gilli G (1996) The assessment of skeletal maturation. Hormone Res 45: 49-52 Suppl. 2.
[4] E. Pietka, L. Kaabi, M. Kuo, and H. Huang, Feature extraction in carpalbone analysis, IEEE Trans. Med. Imag., vol. 12, no. 1, pp. 44–49, Mar. 1993.
[5] Molinari L, Gasser T, Largo RH (2004) Bone age: RUS/CB and gender differences of percentiles for score and score increments. Ann Hum Biol 31: 421-435.
[6] Aja-Fernandez S, de Luis-Garcia R, Martin-Fernandez MA, Alberola-Lopez C (2004) A computational TW3 classifier for skeletal maturity assessment. A computing with words approach. J Biomed Inf 37 (2): 99-107.
[7] Malina RM, Beunen GP(2002) Assessment of skeletal maturity and prediction of adult height (TW3 method). Am J Hum Biol 14 (6): 788-789.
[8] Ashizawa K, Kumakura C, Zhou X, Fin F, Cao J (2005) RUS skeletal maturity of children in Beijing. Ann Hum Bio 32 (3): 316-325.
[9] Tristan A, Arribas JI (2005) A radius and ulna skeletal age assessment syst em. Machine Learning for Signal Processing, IEEE Workshop: 221-226.
[10] Proos LA, Lonnerholm T, Jonsson B, Tuvemo T (2006) Can bone age determination provide criteria for growth hormone treatment in adopted girls with early puberty? A comparison of the Greulich-Pyle and the Tanner-Whitehouse 2 methods. Ups J Med Sci 111: 117-129.
[11] Milner GR, Levick RK, Kay R (1986) Assessment of bone ages: a comparison of the Greulich and Pyle, and the Tanner and Whitehouse methods. Clin Radiol 37:119-212.
[12] Frisch H, Riedl S, Waldhor T (1996) Computer-aided estimation of skeletal age and comparison with bone age evaluations by the method of Greulich-Pyle and Tanner-Whitehouse. Pedia Radiol 26 (3): 226-231.
[13] King DG, Steventon DM, Osullivan MP, Cook AM, Hornsby VPL, Jefferson IG, King PR (1994) Reproducibility of bone ages when performed by radiology registrars- an audit of Tanner-and-Whitehouse-II versus Greulich-and Pyle methods. Bri J Radiol 67 (801): 848-851.
[14] Ryan PJ, Spector TP, Blake GM, Doyle DV, Fogelman I (1993) A comparison of reference bone-mineral density-measurements derived from 2 sources- referred and population-based. Bri J Radiol 66 (792): 1138-1141.
[15] Vignolo M, Milani S, Cerbrllo G (1992) FELS, Greulich-Pyle, and Tanner-Whitehouse bone-age assessments in a group of Italian children and adolescents. Am J Hum Biol 4 (4): 493-500.
[16] Aicardi G, Vignolo M, Milani S, Naselli A, Magliano P, Garzia P (2000) Assessment of skeletal maturity of the hand-wrist and knee: a comparison among methods. Am J Hum Biol 12 (5): 610-615.
[17] Bulla RK, Edwardsa PD, Kempb PM, Fryc S, Hughesc IA (1999) Bone age assessment: a large scale comparison of the Greulich and Pyle, and Tanner and Whitehouse (TW2) methods. Arch Dis Child 81: 172-173.
[18] Sato K, Mito T, Mitani H (2001) An accurate method of predicting mandibular growth potential based on bone maturity. Am J Orthod Dentofac Orthop 120: 286-293.
[19] Malina RM, Reyes MEP, Eisenmann JC, Horta L, Rodrigues J, Miller R (2000) Height, mass and skeletal maturity of elite Portuguese soccer players aged 11-16 years. J Sports Sci 18 (9): 685-693.
[20] Michael DJ, Nelson AC (1989) HANDX: a model-based system for automatic segmentation of bones from digital hand radiographs. IEEE T Med Imaging 8: 64-69.
[21] Pietka E, McNitt-Gray M, Kuo M.K, Huang HK (1991) Computer-assisted phalangeal analysis in skeletal age assessment. IEEE T Med Imaging 10: 616–629.
[22] Pietka E, Kaabi L, Kuo ML, Huang HK (1993) Feature extraction in carpal-bone analysis. IEEE T Med Imaging 12: 616-620.
[23] Hsieh CW, Jong TL, Chou YH, Tiu CM (2007) Computerized carpal bone geometric feature for bone age estimation. Chin Med J (Accepted in Nov., 2006)
[24] Chu BC (2005) An investigation of computerized automatic bone age assessment system based on carpal and phalanx. Master Thesis, EE NTHU.
[25] Huang MJ (2006) The construction of a carpal ossified model and its applications to carpal bone age assessment and bone growth estimation based on fuzzy criterion and statistics analysis. Master Thesis, EE NTHU.
[26] Fan BC, Hsieh CW, Jong TL, Tiu CM (2001) Automatic bone age estimation based on carpal bone images—a preliminary report. Chin Med J 64: 203-208.
[27] Johnston, Jahina (1965) The contribution of the carpal bones to the assessment of skeletal age. Amer. J, Phys. Anrhrop 23: 349-354.

[28] Chang CH (2003) An investigation of computerized automatic bone age assessment system based on phalangeal image features. Master Thesis, EE NTHU.
[29] Lim, Jae S (1990) Two-dimensional signal and image processing, Englewood Cliffs, NJ, Prentice Hall, pp. 478-488.
[30] Canny, John (1986) A computational approach to edge detection IEEE T Pattern Anal and Machine Intelligence 6: 679-698.