SBIR Phase I: Single-shot X-ray Phase-contrast Imaging Using Deep Learning Approaches

这项小企业创新研究（SBIR）第一阶段项目的广泛影响涉及下一代x射线成像系统的好处。提出的单镜头平台克服了目前差分相衬（DPC） x射线成像广泛商业化的障碍。如果成功，单次发射技术将使下一代x射线成像仪的开发和应用成为可能，用于探测液体、小型爆炸物以及航空和商业应用中的其他安全威胁。降低机场误报率将提高客户满意度，提高安全性，并降低成本。DPC成像还可以大大提高对食物害虫的检测，从而减少食物浪费，节省数十亿美元。在另一个市场，无损检测可以显著改善增材制造产品的检验，通过更少的迭代减少制造时间，创造高质量的产品。医疗DPC成像仪将提供类似MRI（磁共振成像）的分辨率和诊断，其成本比目前的MRI低一个数量级。这个小企业创新研究第一阶段项目旨在开发一种深度学习方法来实现“单次”x射线相衬成像。为了使该技术商业化，深度学习算法需要有效而准确地识别更复杂的现实世界物体。深度学习方法需要成千上万的训练样本来建立一个可靠的模型。然而，目前还没有这种独特技术的成像库。该研发计划最初采用x射线相衬成像的标准慢扫描方法，以获得DPC三特征计算机断层扫描（CT）图像。三特征CT图像为精确的材料表征（如吸收系数、折射率和散射特性）提供了基础。一旦材料被表征，它们就构成了创造数百万个现实世界物体的数字表示的基础。这些对象随后形成了有效和高效地训练深度学习模型的核心，而无需进一步的实验。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project relates to the benefits of next-generation X-ray imaging systems. The proposed single-shot platform overcomes the current barriers to widespread commercialization of Differential Phase Contrast (DPC) X-ray imaging. If successful, the single-shot technology will enable the development and application of next generation X-ray imagers for detecting liquids, small explosives, and other security threats for aviation and business applications. Reducing false alarm rates at airports will increase customer satisfaction, improve security, and reduce cost. DPC imaging could also substantially increase the detection of food pests, thereby reducing food waste and saving billions of dollars. In another market, non-destructive testing could significantly improve the inspection of additive manufacturing products, reducing manufacturing time through fewer iterations and creating high-quality products. Medical DPC imagers would provide MRI (Magnetic Resonance Imaging)-like resolution and diagnostics at an order of magnitude lower cost than current MRI.This Small Business Innovation Research Phase I project aims to develop a deep-learning approach to realize “single-shot” X-ray phase-contrast imaging. To commercialize the technology, the deep-learning algorithm needs to identify more complicated real-world objects effectively and accurately. Deep-learning methods require thousands to millions of training samples to make a reliable model. However, no imaging library for this unique technology currently exists. The research and development plan initially incorporates the standard slow scanning method of X-ray phase-contrast imaging to obtain DPC tri-signature computed tomography (CT) images. The tri-signature CT images provide the basis for precise material characterization (e.g., absorption coefficients, indices of refraction, and scatter characteristics). Once the materials have been characterized, they form the basis for creating millions of numerical representations of real-world objects. These objects subsequently form the core for effectively and efficiently training deep-learning models without further experimentation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.