COMPUTER AIDED DIAGNOSIS IN CHEST RADIOGRAPHY

拟议研究的目标是开发计算机辅助诊断 (CAD)检测肺结节、间质浸润物和 数字胸部图像中的气胸。我们计划开发先进的 用于提高灵敏度的计算机化方案和软件， 具体和效率，以便执行和评价这类计划 在受控的临床环境中。我们认为这些计算机辅助的 诊断方案，为放射科医生提供位置和/或 对高度可疑的损伤进行定量测量，有可能 通过减少人类活动，提高癌症检测的诊断准确率 与放射学诊断相关的错误。 具体地说，我们计划(L)制定一个自动化的改进方案 利用(A)线性和非线性的组合检测肺结节 基于差分图像方法的形态滤波技术 增强和抑制肺结节，(B)减少假性 通过详细分析胸部图像特征进行阳性检测 放射科医生以及人工神经网络的使用，(C)分析 用于减少假阳性的后肋骨，(D)应用 提高灵敏度的小波变换，和(E)观测器 优化使用CAD方法的性能研究；(2)开发一种 通过(A)设计用于自动肺纹理分析的改进方案 自动采样多个感兴趣区域(ROI)的技术 肺野，(B)基于以下方面的新纹理措施的调查 能量谱形状和各向异性的分析 肺纹理，以及(C)人工神经网络在 间质浸润物的检测和分类；(3)开发一种 气胸的自动检测方案通过(A)应用 霍夫变换与边缘增强技术相结合 细微曲线的检测，以及(B)放射科医生的ROC分析 用于评估CAD方案的有用性的绩效；以及(4) 高分辨率地实施和评估CAD方案。高速 由(A)开发的智能图像处理系统原型 具有高效算法和高效人机的工作站 接口，以及(B)开展临床评估的试点研究 我们的胸部CAD计划与传统的读数相比， 与肺结节、间质相关的三种异常 渗出物和气胸。

The goal of the proposed research is to develop computer-aided diagnostic (CAD) schemes for detection of lung nodules, interstitial infiltrates, and pneumothoraces in digital chest images. We plan to develop advanced computerized schemes and software for improvements in sensitivity, specificity and efficiency in order to implement and evaluate such schemes in a controlled clinical environment. We believe that these computer-aided diagnostic schemes, which provide the radiologist with the location and/or quantitative measures of highly suspected lesions, have the potential to improve diagnostic accuracy in the detection of cancer by reducing human errors associated with radiologic diagnoses. Specifically, we plan to (l) develop an improved scheme for automated detection of lung nodules by (a) combinations of linear and nonlinear morphological filtering techniques based on a difference image method for enhancement and suppression of lung nodules, (b) reduction of false positive detections by detailed analysis of image features by chest radiologists and also use of artificial neural networks, (c) analysis of posterior ribs for reduction of false positives, (d) application of wavelet transform for increasing the sensitivity, and (e) observer performance studies for optimal use of CAD methods; (2) develop an improved scheme for automated lung texture analysis by (a) devising an automated technique for sampling numerous regions of interest (ROIs) in the lung fields, (b) investigation of new texture measures based on analysis of the shape and anisotropic properties of the power spectrum of lung textures, and (c) application of artificial neural networks for detection and classification of interstitial infiltrates; (3) develop an automated scheme for detection of pneumothorax by (a) application of the Hough transform in conjunction with an edge enhancement technique for detection of subtle curved lines, and (b) ROC analysis of radiologists' performances for evaluation of the usefulness of the CAD scheme; and (4) implement and evaluate the CAD schemes in a high-resolution. high-speed image processing system by (a) development of a prototype intelligent workstation with efficient algorithms and efficient man-machine interfaces, and (b) carrying out pilot studies on clinical evaluation of our chest CAD schemes in comparison with conventional readings in terms of the three types of abnormalities related to lung nodules, interstitial infiltrates, and pneumothoraces.