Improving the Specificity of Dynamic MRI in Breast Cancer Diagnosis

描述(申请人提供)：本项目旨在显著提高动态MRI在检测和诊断乳腺癌方面的特异性。虽然目前的动态乳腺MRI对乳腺癌强化的检测具有较高的敏感性，但在肿瘤诊断中的特异性有限。这部分是由于目前用于动态乳腺MRI增强分割的配准和分割方法的局限性。值得注意的是，图像配准和分割的准确性不仅影响到对比度增强的分割，而且最终影响对比度增强的良恶性分类，正如文献所证明的那样。更好的图像配准和分割可以更准确地分割增强的焦点，从理论上减少假阴性。另一方面，当前动态乳腺MRI的低特异度也与目前仅使用动态动力学参数或形态特征或其简单组合用于癌症诊断的增强分类方法的局限性有关。更重要的是，据我们所知，几乎所有的癌症诊断方法都是为只有肿块强化而不是非肿块强化而设计的。这个项目的目标是通过开发两个分别用于增强分割和分类的新模块来克服这些限制。在目标1中，将开发一个高级增强分割模块，用于以高灵敏度和高特异度分割潜在的可疑增强。特别是，将开发一种新的时空配准方法来在图像采集过程中一致地估计患者的运动，并将开发一种基于图割的分割技术来自适应地分割增强区域。在目标2中，将开发一种新的增强分类模块，通过考虑其各自的增强形态来区分肿块和非肿块增强病例的良、恶性增强。特别是，肿瘤结构的演变模式将完全从动态MRI序列中捕获并用于增强分类，而不是像文献中流行的那样仅使用简单的区域动态特征。这两个模块的疗效将通过一个现有的数据库进行评估，该数据库包含700多名患者的核磁共振扫描。公共卫生相关性：该项目旨在显著提高动态MRI在检测和诊断乳腺癌方面的特异性。特别是，将开发一个高级增强分割模块，用于以高灵敏度和高特异度分割潜在的可疑增强。并且，将开发一种新的增强分类模块，通过考虑其各自的增强形态来区分肿块和非肿块增强病例的良恶性增强。

DESCRIPTION (provided by applicant): This project aims at significant improvement of specificity of dynamic MRI in detecting and diagnosing breast cancer. Although current dynamic breast MRI has high sensitivity in detecting enhancement of breast cancer, it has limited specificity in cancer diagnosis. This is partially due to the limitations of the current registration and segmentation methods used for enhancement segmentation in the dynamic breast MRI. It is worth noting that the accuracy of image registration and segmentation affects not only the segmentation of contrast enhancement, but also eventually the classification of contrast enhancement as benign or malignant, as having been proved in the literature. Better image registration and segmentation allow more accurate segmentation of foci of enhancement and in theory reduce false negatives. On the other hand, the low specificity of current dynamic breast MRI is also related to the limitation of current enhancement classification methods which use only the dynamic kinetic parameters or morphologic features, or their simple combinations for cancer diagnosis. More importantly, according to our knowledge, almost all cancer diagnosis methods were designed for classification of only mass enhancement, not the non-mass enhancement. The goal of this project is to overcome these limitations by developing two novel modules for enhancement segmentation and classification, respectively. In Aim 1, an advanced enhancement segmentation module will be developed for segmenting potentially suspicious enhancement with high sensitivity and specificity. In particular, a novel spatiotemporal registration method will be developed for consistent estimation of patient motion during the image acquisition, and also a graph-cut based segmentation technique will be developed for adaptive segmentation of enhancement regions. In Aim 2, a novel enhancement classification module will be developed to differentiate benign enhancement from malignant enhancement for both mass and non-mass enhancement cases by considering their respective enhancement morphologies. In particular, the evolving pattern of tumor architecture will be completely captured from dynamic MRI sequence and used for enhancement classification, instead of using only simple region-wise dynamic features as popularly used in the literature. The efficacy of these two modules will be evaluated by an existing database with MRI scans from over 700 patients. PUBLIC HEALTH RELEVANCE: This project aims at significant improvement of specificity of dynamic MRI in detecting and diagnosing breast cancer. In particular, an advanced enhancement segmentation module will be developed for segmenting potentially suspicious enhancement with high sensitivity and specificity. And, a novel enhancement classification module will be developed to differentiate benign from malignant enhancement for both mass and non- mass enhancement cases by considering their respective enhancement morphologies.