CAD for CT Nodules in Lung Cancer Detection

描述（由申请人提供）：拟议研究的目标是开发计算机辅助诊断（CAD）方案，用于检测和表征计算机断层扫描（CT）肺部图像中的肺结节。计算机输出将被用作“第二意见”，以帮助放射科医生解释CT肺结节，以早期发现肺癌。将通过结合多模板匹配技术和大规模训练人工神经网络（MTANN）开发一种先进的CAD方案，以实现80-90%的高灵敏度，同时低剂量螺旋CT（LDCT）图像的每部分的假阳性率约为0.1或更低。一种新的减影CT技术将被开发用于通过抑制包括肺血管在内的正常背景肺结构来增强细微肺结节。将研究减影CT技术与常规CT图像相结合在提高肺结节计算机检测的整体性能方面的有用性。除了检测任务外，还将开发一种用于表征结节的CAD方案，以区分良性和恶性病变。该表征任务旨在根据计算机和/或放射科医生检测到的结节图像特征的定量分析，提供恶性肿瘤的估计可能性。一个新的心理物理措施将确定基于人工神经网络的客观图像特征和主观评级对相似的图像，并将用于选择一组良性和恶性结节从一个大型数据库，这将是类似于一个未知的结节的问题，以协助放射科医生的图像解释。随着高水平的检测性能，我们期望实现，原型CAD工作站将开发和观察员的性能研究将进行检查的潜在有用性的CAD方案的检测和分类的肺结节的CT图像。这些CAD方案将为放射科医生提供高度可疑病变的位置和/或良性或恶性结节的定量测量，并有可能提高肺癌早期检测的诊断准确性，从而改善患者的预后。

DESCRIPTION (provided by applicant): The goal of the proposed research is to develop computer-aided diagnostic (CAD) schemes for detection and characterization of pulmonary nodules in computed tomography (CT) lung images. The computer output will be used as a "second opinion" to assist radiologists in their interpretation of CT lung nodules for early detection of lung cancer. An advanced CAD scheme will be developed by incorporating a multiple-template matching technique and also a massive training artificial neural network (MTANN) in order to achieve a high sensitivity of 80-90% with a low false positive rate of approximately 0.1 or less per section of low-dose helical CT (LDCT) images. A novel subtraction CT technique will be developed for enhancing subtle lung nodules by suppressing the normal background lung structure including pulmonary vessels. The usefulness of the subtraction CT technique in combination with the conventional CT image will be investigated in improving the overall performance in the computerized detection of lung nodules. In addition to the detection task, a CAD scheme for characterization of nodules will be developed in order to distinguish between benign and malignant lesions. This characterization task will be designed to provide the estimated likelihood of malignancy based on quantitative analysis of image features of nodules detected by computer and/or radiologists. A new psychophysical measure will be determined based on ANN by use of both objective image features and subjective ratings on pairs of similar images, and will be used to select a set of benign and malignant nodules from a large database, which would be similar to an unknown nodule inquestion, in order to assist radiologists' image interpretation. With the high level of detection performance that we expect to achieve, a prototype CAD workstation will be developed and observer performance studies will be carried out to examine the potential usefulness of CAD schemes on detection and classification of pulmonary nodules in CT images. These CAD schemes will provide the radiologists with the location of highly suspected lesions and/or quantitative measures for benign or malignant nodules, and have the potential to improve diagnostic accuracy in the early detection of lung cancer, which may lead to improved prognosis of patients.