基于组合学习算法的肺栓塞诊断模型

Pulmonary Embolism Diagnosis and Treatment Model Based on Combined Learning Algorithm

CT血管造影在肺栓塞临床诊断中有着广泛的应用,然而血管造影大都包含上百张肺部切片,人工筛选这些切片效率低下出错率较高。尝试通过组合学习算法建立自动识别模型来有效提高识别能力,识别模型由三部分构成:数据的不平衡处理,变量选择方法和组合学习模型。通过比较不同不平衡数据处理策略和变量选择方法的基础上,选择Adaboost方法进行分类算法的学习,临床数据的结果表明该方法能较好地辅助实际诊断。

： Computed Tomography Angiography （CTA） has become an accurate diagnostic tool for Pulmonary Emboli （PE）. However, each CTA study consists of hundreds of images, each representing one slice of the lung. Manual reading of these slices is an inefficiency job with high misclassification rate. This paper is aimed to build models which can automatically identify the disease through a combination of learning algorithms. The proposed diagnostic model consists of three parts： the imbalance data processing, variable selection procedure and Boosting algorithm model. After comparing the result of different unbalance data processing strategies and different variable selection methods, the Bootstrap Adaboost learning model is chosen to conduct our classification task. Finally, results in clinical data shows that our model performed well.