RPI ID:
2019-001-401 / 2019-001-601
Innovation Summary:
A machine‑learning‑enhanced reconstruction pipeline improves both contrast sensitivity and spatial resolution in grating‑based X‑ray and neutron interferometry by fusing complementary image channels through a trained convolutional neural network (CNN). Paired interferometer images—capturing sensitivity and high‑resolution information—are size‑matched and combined with numerically generated phantoms to train a configurable CNN featuring convolution layers, ReLU activation, and iterative forward/backward propagation to minimize a task‑specific loss function. Once trained, the model outputs higher‑quality phase and dark‑field images without requiring additional exposures, serving as a drop‑in software upgrade for existing systems. The framework is hardware‑agnostic and adaptable through tunable parameters such as filter count, padding, and stride to match different system geometries, beam energies, and grating configurations.
Challenges / Opportunities:
Generalizing across instruments requires strategies such as transfer learning, calibration management, and domain‑adapted phantom generation to account for variations in noise, grating period, and acquisition drift. Robust ground‑truth datasets remain essential, particularly for safety‑critical nondestructive evaluation (NDE) pipelines where explainability and validation are required. Integration with existing reconstruction workflows and industrial formats may entail engineering refinement and periodic model retraining. Opportunities include bundling with imaging software suites, expanding to neutron and multimodal imaging, adding multi‑contrast fusion, and optimizing for real‑time inspection via accelerated inference or on‑detector hardware.
Key Benefits / Advantages:
✔ Enhances both contrast sensitivity and spatial resolution simultaneously
✔ Reduces dose/time by improving image quality without extra exposures
✔ Software‑based upgrade compatible with legacy interferometer hardware
✔ Extensible CNN architecture configurable to varied imaging setups
✔ Supports X‑ray and neutron interferometry modalities
✔ Enables improved micro‑defect detection and metrology
Applications:
• Nondestructive testing of composites, batteries, additive‑manufactured parts
• Phase‑contrast and dark‑field imaging for micro‑defect detection
• Materials metrology, stress/texture analysis, and research instrumentation
• Inline or near‑real‑time inspection for industrial and security systems
• Preclinical imaging where dose and acquisition time are constrained
Keywords:
grating interferometer, CNN reconstruction, phase contrast imaging, dark‑field imaging, nondestructive testing, machine‑learning fusion
Intellectual Property:
Issued US patent application 12,266,162; Published US patent application no. US2025/0225782 A1
