CAD in Breast MRI based on Biological Neural Network

描述（由申请人提供）：乳腺MRI CAD中使用的标准技术是基于监督人工神经网络的，其判别结果不理想，应用能力有限。与这些技术相关的主要缺点是：(1)需要固定的磁共振成像方案；(2)难以诊断直径仅为几毫米的小乳房肿块；(3)无法捕获病变结构；(4)由于病变数据池不均匀而造成的训练限制。为了克服上述问题，本研究计划的主题成为采用严格关注观察到的完整MRI信号时间序列的生物神经网络，实现对动态增强乳房MRI时间序列的自组织数据驱动分割，即信号幅度和动态的细粒度差异，如不确定乳腺病变患者的局灶增强。本项目的目标是在一个跨学科的框架内提高乳腺MRI的诊断质量。具体而言，本项目的目标是：(1)开发、评估和测试用于动态增强乳房MRI数据的功能和结构分割、可视化和分类的新型神经网络技术，并因此(2)通过改进对传统x射线乳房x线摄影检测到的可疑病变的进一步评估，为乳腺癌诊断做出实质性贡献。PI是一位具有模式识别背景的电子和计算机工程师，他一直在开发新的分类方法，这些方法来源于最新的生物学发现，旨在模仿生物系统中的决策和感觉处理。该指导定量研究职业发展奖将允许PI获得癌症研究技术和计算机辅助放射学方面的培训，并使用这些技能将这些新的理论工具扩展并有效地应用于生物医学应用。因此，PI的长期职业目标是在模式识别的生物医学应用方面成为一名有效的研究人员，特别强调计算机辅助诊断。该研究的结果有望对医疗政策产生重大影响，有助于通过非侵入性成像诊断不确定的乳房病变。

DESCRIPTION (provided by applicant): Standard techniques used in CAD for breast MRI are based on supervised artificial neural networks and have shown unsatisfactory discriminative results and limited application capabilities. The major disadvantages associated with these techniques are: (1) requirement of a fixed MR imaging protocol, (2) difficulties in diagnosing small breast masses with a diameter of only a few mm, (3) incapacity of capturing the lesion structure, and (4) training limitations due to an inhomogeneous lesions data pool. To overcome the above mentioned problems, the theme of this research plan becomes to employ biological neural networks which focus strictly on the observed complete MRI signal time-series, and enable a self-organized data-driven segmentation of dynamic contrast-enhanced breast MRI time-series w.r.t. fine-grained differences of signal amplitude, and dynamics, such as focal enhancement in patients with indeterminate breast lesions. The goal of the present project is to improve in an interdisciplinary framework the diagnostic quality in breast MRI. Specifically, the objectives of this proposed project are to: (1) develop, evaluate and test novel neural network techniques for functional and structural segmentation, visualization, and classification of dynamic contrast-enhanced breast MRI data, and thus, (2) substantially contribute to breast cancer diagnosis by improved further evaluation of suspicious lesions detected by conventional X-ray mammography. The PI is an electrical and computer engineer with a background in pattern recognition who has been developing new classification methods derived from the newest biological discoveries aiming to imitate decision-making, and sensory processing in biological systems. This Mentored Quantitative Research Career Development Award will permit the PI to acquire training in cancer research techniques and in computer assisted radiology, and to use these skills to extend and productively apply these new theoretical tools to biomedical applications. Accordingly, the long-term career goal of the PI is to become an effective researcher in the biomedical applications of pattern recognition, with specific emphasis in computer-aided diagnosis. The outcome of the proposed research is expected to have substantial implications in healthcare politics by contributing to the diagnosis of indeterminate breast lesions by non-invasive imaging.