Image Science for the New X-ray: Taking NEQ to Task

描述（由申请人提供）：平板成像仪（fpi）正在一系列先进的x射线成像模式中进行研究，这些模式提供了更好的组织可视化和深度识别，包括双能成像（DE）、断层合成和锥束CT （CBCT）。每一种都代表了诊断和图像引导程序的一种有前途的技术。在DE成像中，在不同的x射线能量下获得的两幅图像被处理成骨骼或软组织的图像。在断层合成和CBCT中，x射线源和FPI在患者周围移动，从多个角度重构三维图像。实现这些模式的全部潜力需要成像性能的定量评估和揭示限制图像质量的物理因素的方法。传统的基于傅立叶的度量，如噪声等效量子（NEQ），为空间分辨率和信噪比提供了实用的、普遍的价值数字。然而，对于这些先进的成像模式，目前还没有严格的性能计量（基于傅立叶或其他）；此外，NEQ传统上只描述检测器的性能，而不考虑图像中感兴趣的结构（即任务）和观察者的响应。本建议采用一种基本的图像科学方法，在这些先进的模式中进行绩效评估，在许多方面都有新的突破：NEQ和相关的傅立叶度量被扩展到每个先进的模式，以提供成像性能评估的实验和理论方法；2)。考虑到与两种最致命的癌症（肺癌和肝转移）相关的各种简单和高阶任务（检测、定位和大小估计），根据成像要求对图像中感兴趣的结构进行量化；和3)。在DE成像、断层合成和CBCT中，将成像任务和NEQ的定量公式严格结合起来，产生基于任务的成像性能指标，这些基于任务的指标在多大程度上为图像质量提供了有意义的替代指标，并通过与人类观察者的反应相关联来验证。因此，本提案开始通过严格的成像任务量化，弥合流行的基于傅立叶的方法和传统的基于观察者的方法（例如ROC分析）之间的差距。例如，使用DE成像和断层合成的检测和大小估计任务来评估肺结节早期检测的成像性能，而根据CBCT的定位任务来评估肝转移的射频消融或放射治疗的指导。该项目的成功完成将为这些和其他新型成像模式的设计、评估和优化提供实用的、任务驱动的方法。

DESCRIPTION (provided by applicant): Flat-panel imagers (FPIs) are being investigated in a host of advanced x-ray imaging modalities that offer improved tissue visualization and depth discrimination, including dual-energy (DE) imaging, tomosynthesis, and cone-beam CT (CBCT). Each represents a promising technology for diagnostic and image-guided procedures. In DE imaging, two images acquired at different x-ray energies are processed to yield images of bone or soft-tissue. In tomosynthesis and CBCT, the x-ray source and FPI move about the patient, and 3D images are reconstructed from multiple perspectives. Realizing the full potential of these modalities requires quantitative evaluation of imaging performance and a methodology for revealing the physical factors that limit image quality. Conventional Fourier-based metrics, such as Noise-Equivalent Quanta (NEQ), provide practical, prevalent figures of merit for spatial resolution and signal-to-noise ratio. For these advanced imaging modalities, however, there is currently no rigorous performance metrology (Fourier-based or otherwise); furthermore, the NEQ conventionally describes only the performance of the detector, and considers neither the structures of interest in the image (i.e., the task) nor the response of observers. This proposal pursues a fundamental image science approach to performance evaluation in these advanced modalities, breaking new ground in numerous respects: 1.) The NEQ and associated Fourier metrics are extended to each of the advanced modalities to provide an experimental and theoretical methodology for imaging performance evaluation; 2.) Structures of interest in the image are quantified in terms of the imaging t ask, considering a variety of simple and higher-order tasks (detection, localization, and size estimation) in relation to two of the most lethal cancers (lung carcinoma and liver metastases); and 3.) Quantitative formulations of imaging task and NEQ are rigorously combined to yield task-based metrics for imaging performance in DE imaging, tomosynthesis, and CBCT, and the extent to which such task-based metrics provide a meaningful surrogate for image quality is validated by correlation with the response of human observers. Hence, this proposal begins to bridge the gap between the prevalent Fourier-based approach and traditional observer-based approaches (e.g., ROC analysis) through rigorous quantitation of imaging task. For example, imaging performance in early detection of lung nodules is evaluated using Detection and Size Estimation tasks for DE imaging and tomosynthesis, while guidance of RF ablation or radiation therapy of liver metastases is evaluated in terms of a Localization task for CBCT. Successful completion of this program will offer a practical, task-driven approach to the design, evaluation, and optimization of these and other novel imaging modalities.