基于医学超声多模态影像信息挖掘的乳腺癌辅助诊断技术研究

Clinical medical ultrasound is one of the most important imaging tools for diagnosing breast cancer. Nowadays, there have been a number of imaging modalities based on the ultrasound, and they can be applied for the diagnosis of breast cancer. Nevertheless, employment of a single ultrasonic imaging modality cannot lead to the diagnosis with high accuracy. In this proposal, we firstly propose the computer-aided diagnostic (CAD) technologies based on multi-modality medical ultrasound imaging information for the breast cancer. This novel CAD technology combines a number of different ultrasonic imaging methods to improve the clinical diagnostic accuracy of the breast cancer. We will deeply investigate the characteristics of several currently used medical ultrasound imaging modalities and their effectiveness on the diagnoses of breast cancer. The methods for digitizing clinician's evaluations based on different imaging modalities will be studied. The data warehouse for storing multi-modality ultrasound image data will be constructed. We firstly propose a novel perspective called supervised biclustering which makes use of supervised learning to discover meaningful local coherent patterns. The theory and algorithms for supervised biclustering will be studied and then applied to the multi-modality medical ultrasound image data in order to achieve significant local patterns for diagnosis of breast cancer in this project. Finally, a CAD platform will be established, and a training system for inexperienced clinicians will be embedded into the CAD platform. It is the first research to combine the image features and parameters from multiple ultrasound imaging modalities using data mining techniques for CAD of breast cancer. Supervised learning and mining the diagnostic patterns with significant clinical merit and automatic generation of individual ultrasonic examination scheme(IUES) would be of great importance to clinically improve the diagnostic accuracy of breast cancer and the practical skill of clinicians in the department of ultrasonic imaging in hospitals.

临床医学超声是诊断乳腺癌的最重要影像工具之一。目前，已有多种基于超声的不同影像模态可应用于乳腺癌诊断，但单独使用一种影像模态的准确率还不够高。本项目首先提出基于多模态超声影像技术的乳腺癌计算机辅助诊断技术，综合多种超声影像技术，提高乳腺癌临床诊断的准确率。我们将深入研究常用的各类超声影像模态特点以及其在乳腺癌诊断过程中的作用；研究不同模态下影像诊断特征的数字化方法；构造乳腺癌超声影像数据仓库，提出有监督双聚类数据挖掘理论和算法，将其应用于多模态乳腺癌超声数据集，提取显著的乳腺病灶良恶性诊断模式；最后建立计算机辅助诊断分析平台和医师培训系统。本研究首次提出综合多模态超声影像数据特征，有监督学习、挖掘具有临床价值的诊断模式，并自动生成个性化超声诊断方案，对提高乳腺癌临床诊断准确率和提高医师的操作水平有重要作用。

临床医学超声是诊断乳腺癌的最重要影像工具之一。目前，已有多种基于超声 的不同影像模态可应用于乳腺癌诊断，但单独使用一种影像模态的准确率还不够高。本项目首先提出基于多模态超声影像技术的乳腺癌计算机辅助诊断技术，综合多种超声影像技术，提高乳腺癌临床诊断的准确率。.我们深入研究了常用的各类超声影像模态特点以及其在乳腺癌诊断过程中的作用；研究了不同模态下影像诊断特征的数字化方法；构造了乳腺癌超声影像数据仓库，并提出了有监督双聚类数据挖掘理论和算法，将其应用于多模态乳腺癌超声数据集，提取显著的乳腺病灶良恶性诊断模式；最后成功建立了计算机辅助诊断分析平台和医师培训系统。该系统现已在中山大学肿瘤医院超声科进行临床测试，有着广阔的应用前景。.项目开展过程中，项目组发表了15篇SCI论文，申请12项发明专利，培养了9名毕业硕士研究生。