Digital chest x-rays using radiomics

Dataset wherein a unique sliding window segmentation method was developed to eliminate the difficult and time-consuming task of accurate Pulmonary tuberculosis (PTB) disease segmentation from planar images. It was applied as a secondary segmentation, superimposed on a primary automatic lung segmentation, that divided the entire lung region into uniform windows that overlapped while sliding over the  chest x-ray (CXR) in both image dimensions. When radiomic features were extracted from each sliding window, it allowed the distribution of the features across the lung region to be evaluated.
Three different outcomes were achieved when radiomic feature extraction was applied to chest x-rays using the sliding window segmentation. Firstly a model was developed that can automatically differentiate normal CXR from CXR with PTB cavities, which could improve the accuracy of CXR reporting currently regaining prominence as a high-volume screening tool. Secondly, signature parameter maps that showed a strong correlation to the lung pathology were constructed. This might be valuable as a quantitative supplementary indicator in the management of PTB disease and further increase the acceptance of CXR as a tool for assessing the Tuberculosis (TB) response in medical research and clinical practice. Finally, a radiomics score was constructed that was able to quantify the change in the disease characteristics as seen from digital CXR of patients diagnosed with PTB. This radiomic score analysis of serial x-rays taken while patients receive TB therapy has the potential to be a quantitative monitoring tool of response to therapy. Radiomics was therefore successfully applied in this study to quantify the characteristics of PTB from chest x-rays.

本数据集研发了一种独特的滑动窗口分割（sliding window segmentation）方法，以解决从平面影像中精准分割肺结核（Pulmonary tuberculosis, PTB）病灶这一耗时且极具挑战性的任务。该方法作为二级分割任务，叠加于一级自动肺分割（lung segmentation）之上：将整个肺区域划分为带有重叠区域的均匀窗口，并在胸部X线（chest X-ray, CXR）的两个图像维度上进行滑动扫描。通过从每个滑动窗口中提取放射组学特征（radiomic features），可实现肺区域内特征分布的全面评估。
采用该滑动窗口分割方法对胸部X线进行放射组学特征提取后，本研究实现了三项不同的研究成果：其一，构建了可自动区分正常胸部X线与伴肺结核空洞的胸部X线的模型。当前胸部X线作为大规模筛查工具重新受到重视，该模型可有效提升胸部X线报告的精准度。其二，构建了与肺部病理具有强相关性的特征参数图谱，其可作为肺结核管理中的定量辅助指标，同时有望进一步推动胸部X线在医学研究与临床实践中作为评估结核病（Tuberculosis, TB）治疗应答工具的应用普及。其三，针对确诊肺结核患者的数字化胸部X线影像，构建了可量化疾病特征变化的放射组学评分。对患者接受结核病治疗期间获取的系列胸部X线进行该放射组学评分分析，有望成为评估治疗应答的定量监测工具。综上，本研究成功将放射组学（radiomics）技术应用于从胸部X线影像中量化肺结核的特征。