X 光片的骨齡評估 (bone age assessment, BAA) 是小兒科醫師用來判讀兒童生長發育的常規檢查。目前最常使用的判讀準則有兩種：G&P 和 TW2 方法。以 G&P 和 TW2 方法做的骨齡比對方法，不僅耗時，且會根據個人主觀意識或人員訓練方法不同而有不同的比對結果。為了排除上述人為因素，使影像判讀有更客觀的結果，我們提出以電腦視覺的技術從 X 光片中擷取腕骨輪廓。 我們的方法包含影像前處理及腕骨分割兩部份。在影像前處理部份，我們利用二值化 (bi-level thresholding) 手部影像的前臂資訊校正腕部方向，並由尺、橈骨交點和掌骨資訊幫助訂定腕骨分析範圍 (CROI)，藉由最小最大濾波器 (min/max filter) 評估 CROI 的影像背景，將原影像減去背景以提昇對比。在腕骨分割部份，我們利用 Osma-Ruiz等人的分水嶺方法、梯度運算、及距離轉換方法做重疊腕骨的分割；其中的分割重點在於腕骨重疊時各腕骨的判定，引進腕骨骨化次序的特性，依序對鉤狀骨、頭狀骨、三角骨、月狀骨、大多角骨、小多角骨、及舟狀骨進行輪廓擷取。此技術不僅適用 7 歲以下未重疊的腕骨，也有助於 7-14 歲重疊腕骨的分割。 Bone age assessment (BAA) using hand radiographs is an important technique of pediatric radiology for doctors assessing the growing situation of children. Two most commonly used methods are the Greulich-Pyle (G&P) method and the Tanner-Whitehouse (TW2) method. However these methods have the disadvantage of time consuming and the results of BAA highly depending upon the assessor expertise and experience. In order to obtain more objective results, an automatic carpal bone feature extraction method is pursued and proposed. The proposed system consists of two main stages: image preprocessing and carpal-bone segmentation. In image preprocessing, we starts with acquiring wrist direction and extracting carpal bone region of interest (CROI) in the binary hand images. Then, the CROI contrast is enhanced by using a min/max filter. In carpal-bone segmentation, the techniques of watershed method proposed by Osma-Ruiz et al., gradient technique, distance transformation, and anatomy knowledge were used to assist the extraction of carpal bone contours. The carpal bones include: trapezium, trapezoid, capitate, hamate, scaphoid, lunate, triquetrum, and pisform. Our system can extract carpal bone contours effectively in 1-14 year old children.
