Neural Network Material Decomposition for Multi-Spectral X-Ray Imaging

The patent extends conventional single-energy x-ray imaging to multi-spectral imaging by adding the energy dimension. The system acquires x-ray projections at multiple different energy spectra, transforming the problem from 2D spatial imaging to 3D imaging plus energy spectrum analysis. This dimensional expansion preserves energy-dependent information that would be lost in conventional imaging, enabling accurate material composition determination through spectral decomposition.

The patent performs preliminary calibration by acquiring measurements from known basis materials (e.g., water and bone) and pre-computing the system response matrices A and B. These calibration data and derived parameters are stored for later use. During actual material decomposition, the pre-computed calibration information is reused, eliminating the need to repeatedly solve complex nonlinear equations and reducing both processing power and memory requirements while maintaining accuracy.

The patent transforms the material decomposition problem from solving complex nonlinear equations to a linear system solution by changing the parameter representation. By expressing the attenuation coefficients as linear combinations of basis material coefficients and using pre-computed system response matrices, the method converts a difficult nonlinear optimization problem into a straightforward linear algebra problem that can be solved efficiently with standard computational methods while maintaining physical accuracy.

The patent introduces basis materials as intermediaries to bridge the gap between raw multi-spectral detector measurements and material composition information. By measuring the system response to known basis materials and using these as reference standards, the method creates an intermediate calibration layer that accounts for detector nonidealities and system variations, simplifying the overall measurement process while improving accuracy.