I-Corps: Coordinates and Volumetrics in MRI Imaging

I-Corps项目的更广泛的影响/商业潜力是，通过开发临床就绪的软件套件，通过磁共振成像（MRI）计算医生批准的大脑关键三维（3D）结构的体积，填补神经放射学、神经肿瘤学和放射治疗的关键需求。目前，医生通过视觉检查和测量单个二维轴向图像的最大垂直直径来判断肿瘤的状态。放射肿瘤学家使用人工分割来划定肿瘤区域的边界；这些轮廓用于治疗计划。目前的实践存在显著的局限性，阻碍了患者护理的质量。通过生成肿瘤和大脑其他关键区域的三维体积结构，该软件套件将有可能改变现状。医生可以很容易地应用这些信息来更好地了解疾病并优化护理，从而实现早期治疗，降低发病率，延长生存时间。这个I-Corps项目生产了一套实用的、临床就绪的磁共振（MR）分析和显示工具，以解决可靠使用mri指导脑癌放射治疗的首要障碍：准确及时的大脑关键结构的多模态3D分割。MR套件的主要引擎是一个数学优化框架，它结合了变化演算和深度学习技术；因此可以几乎实时地进行像素级精确的3D分割。所提出的技术是从一种多模态脑分割算法发展而来的，该算法包括：(a)一种自动化、准确和鲁棒的3D图像分割算法，以及(b)需要医生批准的半自动和交互式多模态标记。

The broader impact/commercial potential of this I-Corps project is to fill a critical need in neuro-radiology, neuro-oncology, and radiation therapy, by developing a clinic-ready software suite that computes physician-approved volumes of key three-dimensional (3D) structures in the brain from Magnetic Resonance Imaging (MRI). Currently, physicians make a decision on tumor status by visual inspection and by measuring the largest perpendicular diameters from a single two-dimensional (2D) axial image. Radiation oncologists use manual segmentation to delineate the boundaries of the tumor regions; these contours are used for treatment planning. Current practice is associated with significant limitations that impede the quality of patient care. This software suite will potentially change the status-quo by generating 3D volumetric structures of tumors and other key regions in the brain. The physician can readily apply this information to better understand the disease and optimize care leading to early treatment, less morbidity, and longer survival times.This I-Corps project produces a practical, clinic-ready suite of Magnetic Resonance (MR) analyses and display tools to solve the number one impediment to reliable use of MRIs for guiding radiation treatment of brain cancer: accurate and timely multimodal 3D segmentation of key structures in the brain. The main engine of the MR suite is a mathematical optimization framework that combines calculus of variation with deep learning techniques; thus amenable to pixel-level-accurate 3D segmentation in almost real-time. The proposed technology was developed from an algorithm for multimodal brain segmentation, which consists of (a) an automated, accurate and robust algorithm for 3D image segmentation, combined with (b) semi-automated and interactive multimodal labeling that requires physician approval.