Computer aided diagnosis of cancer metastases in the brain

这个项目的首要目标是提高诊断脑内癌症转移的准确性。 通过开发一种新的计算机辅助诊断(CAD)技术。在今天的癌症治疗中，它 通常不是原发癌症，而是导致死亡的转移性癌症。许多癌症，包括肺癌， 肾癌、卵巢癌、乳腺癌和黑色素瘤有转移到大脑和 仅在美国，脑转移瘤的数量就高达每年17万例。因此，准确的诊断 脑转移瘤在挽救生命和改善患者健康方面起着至关重要的作用。磁性 磁共振成像(MRI)是最广泛使用的扫描大脑潜在转移的方式，但 诊断转移是一项非常具有挑战性的任务，具有相当大的假阴性率。第一 诊断转移的困难在于，在早期，转移是无症状的。第二个困难 转移瘤在mri上表现为弱信号改变，通常表现为高信号。 类似于正常的大脑结构，如小血管，这意味着一个人必须在他/她的脑海中想象 观察到的对象是转移瘤还是血管。错过转移会带来严重的后果 因为患者将不会被要求进一步治疗。准确诊断转移的好处，在 另一方面，像立体定向放射外科(SRS)这样的治疗可以对患者有显著的好处 许多情况下彻底消除转移的肿瘤，使患者的生命延长三到四年 大多数情况下是几年。 通过识别异常，CAD在提高脑转移瘤诊断准确性方面发挥关键作用 信号强度的变化，并标记它们以供放射科医生检查。在这一过程中，CAD将起到辅助作用 补充人类在解释大脑核磁共振方面的专业知识的工具。然而，尽管找到 而治疗脑转移瘤，目前还缺乏一种CAD方法来解决这个问题。许多现有的 脑MRI的计算技术是根据在研究环境中获得的MRI数据量身定做的，研究环境通常 涉及许多其他MRI技术，如DWI、DTI和功能MRI。但在临床上，只有解剖核磁共振 就像使用T1和T2加权MRI来扫描患者一样，因此，CAD方法必须量身定制 临床环境，以协助放射科医生阅读脑部核磁共振。在这个项目中，我们提出了一个CAD设计，它是 基于新的计算技术，并与常规的临床MRI采集相结合。计算机辅助设计 具有最少的用户干预和参数选择、高度的稳健性和用户友好性。我们会 还可以利用图形处理器(GPU)在实施中的可用性来加速 计算。我们期望所提出的CAD方法将提高对大脑的诊断准确率 转移，进而挽救生命，有益于患者的福祉。

The overarching goal of this project is to improve the accuracy in diagnosing cancer metastases in the brain through the development of a novel computer-aided diagnosis (CAD) technique. In today’s cancer treatment, it is often not the primary cancer but the metastasized cancer that causes fatality. Many cancer, including lung, kidney, ovarian, and breast cancer, and melanoma, have a tendency metastasizing to the brain and the number of brain metastases is as high as 170,000 a year in the US alone. Therefore, accurate diagnosis of brain metastases is of utmost importance in saving lives and improving patient’s well-being. Magnetic resonance imaging (MRI) is the most widely used modality to scan brain for potential metastases but diagnosing metastases is a very challenging task that has a considerable rate of false-negatives. The first difficulty in diagnosing metastases is that, at early stage, metastases are asymptomatic. The second difficulty is that metastases manifest as weak signal intensity changes on MRI and their appearance is often highly similar to normal brain structures, such as small blood vessels, meaning that one must visualize in his/her mind whether an observed object is a metastasis or a blood vessel. Missing a metastasis has a severe consequence as the patient will not be called for further treatment. The benefit of accurate diagnosis of metastases, on the other hand, can have a significant benefit to the patient as treatment like stereotactic radiosurgery (SRS) can completely eliminate the metastasized tumor in many cases and extend patient’s life span by three to four years in most cases. CAD can play a key role in improving the accuracy in diagnosing brain metastases by identifying abnormal signal intensity changes and mark them for radiologists to examine. In this process, CAD will function as an aid tool to complement human’s expertise in interpreting brain MRI. However, despite the importance of finding and treating brain metastases, there currently is lacking a CAD approach to this problem. Many existing computational techniques on brain MRI were tailored to MRI data acquired in a research setting that often involves many other MRI techniques such as DWI, DTI, and functional MRI. But in clinics only anatomic MRI like T1- and T2-weighted MRI are used to scan a patient, therefore, a CAD approach must be tailored to the clinical setting to assist radiologists in reading the brain MRI. In this project we propose a CAD design that is based on novel computational techniques and integrated with routine clinical MRI acquisition. The CAD design features minimum user intervention and parameter selection, high robustness, and user-friendliness. We will also take advantage of the availability of graphics processing unit (GPU) in implementation to speed up the computations. We expect the proposed CAD approach will improve the accuracy of diagnosing brain metastases, and in turn, save lives and benefit patients’ well-being.