3D Massive Training ANN for CAD for Colon Cancer in CT Colonography

描述（由申请人提供）：拟议研究的目标是开发一个12维的连续训练人工神经网络（3D MTANN），用于计算机辅助诊断（CAD）方案，用于计算机断层结肠成像（CTC）中检测结直肠息肉。CAD输出将被用作“第二意见”，以帮助放射科医生检测息肉，从而早期发现结直肠癌。我们将开发一种CAD方案，其中包含3D MTANN，用于区分息肉和非息肉（假阳性），以尽可能减少假阳性的数量，同时保持高灵敏度水平。3D MTANN是一种基于人工神经网络的3D体积处理技术，能够直接对图像数据进行操作。通过输入CTC体积和相应的教学体积，可以训练3D MTANN以增强息肉和抑制非息肉。我们计划开发一种多3D MTANN方案（多3D MTANN），由几个专家3D MTANN组成，用于减少各种类型的假阳性，包括褶皱、粪便、回盲瓣和直肠管。通过对3D MTANN的输出体积应用评分方法，息肉候选者将被分类为息肉或非息肉。我们将比较3D MTANN与二维MTANN的性能，效率和属性。为了获得可靠的评估结果，我们将收集一个大型的CTC病例数据库，包括有无息肉。通过与金标准光学结肠镜对同一患者的诊断报告进行比较，我们将确定放射科医生阅读CTC图像时的“漏诊”病例，即假阴性病例。我们将开发一个原型CAD工作站的基础上，先进的CAD系统，包括多三维MTANN，并评估工作站的性能与数据库的自由响应接收器工作特性（FROC）分析。我们计划进行一项观察者性能研究，通过使用多读者多情况接收器操作特征分析来评估CAD方案的潜在有用性。CAD系统结合多三维MTANN将为放射科医生提供高度可疑病变的位置，它有可能提高结直肠癌早期检测的诊断准确性，这可能会导致改善患者的预后。

DESCRIPTION (provided by applicant): The goal of the proposed research is to develop a 12three-dimensional massive-training artificial neural network (3D MTANN) for a computer-aided diagnostic (CAD) scheme for detection of colorectal polyps in computed tomographic colonography (CTC). The CAD output will be used as a "second opinion" to assist radiologists in detecting polyps for early detection of colorectal cancer. We will develop a CAD scheme incorporating a 3D MTANN for distinction between polyps and non-polyps (false positives) to reduce the number of false positives as much as possible, while maintaining a high sensitivity level. The 3D MTANN is a 3D volume-processing technique based on an artificial neural network which is capable of operating on image data directly. With input CTC volumes and the corresponding teaching volumes, the 3D MTANN can be trained for enhancement of polyps and suppression of non-polyps. We plan to develop a multiple 3D MTANN scheme (multi-3D MTANN) consisting of several expert 3D MTANNs for reduction of various types of false positives including folds, stool, the ileocecal valve, and rectal tubes. By applying a scoring method on the output volumes of the 3D MTANNs, polyp candidates will be classified as polyps or non-polyps. We will compare 3D MTANNs with two-dimensional MTANNs in terms of performance, efficiency, and properties. To obtain reliable evaluation results, we will collect a large database of CTC cases with and without polyps. By comparing with the diagnostic report of the gold standard optical colonoscopy on the same patients, we will determine "missed" cases which are false-negative cases when radiologists read CTC images. We will develop a prototype CAD workstation based on an advanced CAD system incorporating the multi-3D MTANN, and evaluate the performance of the workstation with the database by free-response receiver operating characteristic (FROC) analysis. We plan to carry out an observer performance study to evaluate the potential usefulness of the CAD scheme by use of multi-reader multi-case receiver operating characteristic analysis. The CAD system incorporating with the multi-3D MTANN will provide radiologists with the location of highly suspected lesions, and it has the potential to improve diagnostic accuracy in the early detection of colorectal cancer, which may lead to improved prognosis of patients.