統計分析應用在骨質疏鬆症上的X光影像

Abstract

本文主要是從人體第一節脊椎骨的X光影像萃取出特徵, 並且預測骨頭強度值的分群. 在骨質疏鬆症的臨床實驗中, 這研究提供了一種電腦輔助診斷的工具. 在這個研究中, 我們總共收集了36筆資料, 包含骨頭的X光影像和其相對應的骨頭強度值. 如何利用骨頭強度值來判斷是否有骨質疏鬆症? 我們將使用無母數的核密度估計來決定分群的數目和其分佈. 然後利用有母數的有限混合模型來將骨頭強度值做分群. 型態學中的骨骼化方法是從X光的影像中萃取出骨架的部分. 從所得到的影像, 我們選用了三種參數. 我們可以利用sliced inverse regression 這個降維度的方法, 將選出的特徵投影到較低的維度以便視覺化和分群. 並且利用分類樹將投影過後的特徵進行分群預測. 接下來, 我們用leave-one-out cross validation的方法來計算預測錯誤率, 其預測錯誤率是19.4%.

This study is aimed to extract features from X-ray images of the first lumbar vertebra and predict the clusters of measured bone strengths. This provides a tool for the computer-aided diagnosis for detecting osteoporosis in clinical practice. There are 36 digital X-ray images and their measured bone strengths collected in this study. How to decide osteoporosis by the measured bone strengths? We will use the nonparametric method based on kernel density estimation to determine the cluster number and the shapes of the distribution. Then, the parametric model of finite mixtures is used to separate the bone strengths into clusters. The skeleton operation in morphology is used to extract skeletal images from X-ray image. From skeletal images, three kinds of features are extracted. The dimension reduction technique of sliced inverse regression is used to project features into low dimension spaces for visualization and classification. Classification trees of the projected features are applied to predict the clusters. The leave-one-out cross validation is performed to evaluate the prediction errors and the overall error rate is 19.4%.