Whole body image analysis for diagnosing patients with monoclonal plasma cell disorders

全身计算机断层扫描(CT)和磁共振成像(MRI)越来越多地用于单克隆性浆细胞疾病的分期和疗效评估。PET成像为诊断多发性骨髓瘤患者提供了额外的手段，也可以与CT或MRI扫描一起获得。除了对全球肿瘤负担的估计外，所有这些成像方式都提供了有关疾病状态的局部信息，人们可能会认为，在局部水平评估病变可能比根据血液样本诊断的全球全身反应更敏感。尽管如此，使用全身MRI、CT和PET来诊断和分期疾病的一个普遍困难是要处理的图像信息量。即使对于训练有素的放射科医生来说，数据集也不容易读取，而且到目前为止只提取了基本信息。此外，由于时间限制，当仅分析源图像的子集时，彻底读取完整的3D体积是一项非常乏味和耗时的任务，容易出现疏忽和系统错误。这个问题在临床研究中表现得尤为突出，因为必须评估数百种多模式全身扫描，并从数据中提取疾病进展的一般模式。为此，我们将开发计算工具，使临床医生能够量化多发性骨髓瘤的状态，并在全身图像数据集中交互跟踪随时间的进展。为了获取大型多模全身图像序列的全部信息，我们将解决不同模式和时间的空间对应问题，用于识别和重新识别全身数据中的位置，并绘制病变地图。此外，我们将开发概率模型和生物数学模型，其参数可以作为新的成像生物标志物，并可以与临床、分子和遗传证据相关联，以确认和提高现有图像信息的诊断价值。最后，我们将模拟疾病随时间和一大批患者的进展，系统地评估临床诊断提供的大型数据库，这将有助于更好地了解观察到的疾病进展模式，并验证本项目将开发的工具和模型。

Whole-body computed tomography (CT) and magnetic resonance imaging (MRI) are increasingly used in staging of monoclonal plasma cell disorders and for evaluating treatment response. PET imaging provides additional means for diagnosing multiple myeloma patients, and is also acquired jointly with CT or MRI scans. All these imaging modalities provide localized information about the state of the disease in addition to an estimate of the global tumor burden, and one might expect that evaluating lesions at the local level may be more sensitive to changes than global systemic reactions diagnosed from blood samples. Still, a general difficulty in using whole-body MRI, CT, and PET for diagnosing and staging the disease is the amount of image information to be processed. Data sets are not easily readable even for trained radiologist and to date only basic information is extracted. Moreover, thoroughly reading full 3D volumes is a very tedious and time consuming task that is is prone to oversights and systematic errors when only a subset of the source images are analyzed due to time constraints. This problem exponentiates in clinical studies when hundreds of multimodal whole-body scans have to be evaluated, and general patterns of disease progression have to be abstracted from the data.To this end we will develop computational tools that will enable clinicians to quantify multiple myeloma status and to inter-actively track progression over time in whole-body image data sets. To access the full information of large multimodal whole-body image series, we will solve the problem of spatial correspondence in different modalities and time, to be used for identifying and re-identifying locations in whole-body data, and map lesions. Moreover, we will develop probabilistic and bio-mathematical models whose parameters can serve as new imaging biomarkers, and can be correlated with clinical, molecular, and genetic evidence to confirm and improve the diagnostic value of the available image information. Finally, we will model disease progression over time and across a large cohort of patients, systematically evaluating large data bases available from clinical diagnostics, which will help to better understand the observed disease progression patterns and to validate the tools and models to be developed in this project.