PCCT energy calibration from X-ray tube spectra using a neural network

Patent metadata

Field	Value
Publication number	US-12306362-B2
Application number	US-202318152856-A
Country	US
Kind code	B2
Filing date	Jan 11, 2023
Priority date	Jan 11, 2023
Publication date	May 20, 2025
Grant date	May 20, 2025

Abstract

An apparatus for calibrating a detector, including acquiring an energy spectrum obtained from a scan using an X-ray tube as a source of radiation, estimating calibration parameters, such as a gain and an offset, for each of several channels of the detector by applying the acquired first energy spectrum to inputs of a trained neural network that outputs the calibration parameters, and calibrating each of the plurality of channels using the estimation parameters. The neural network is trained to produce target output calibration parameters, using two or more measurements selected from isotope peak positions, K-edge absorption features, or K-edge emission peaks.

First claim

The invention claimed is:

1. A method for calibrating a detector, the method comprising:
acquiring a first energy spectrum obtained from a scan using an X-ray tube as a source of radiation;
estimating calibration parameters, for each of a plurality of channels of the detector, by applying the acquired first energy spectrum to inputs of a trained neural network that outputs the calibration parameters; and
calibrating each of the plurality of channels using the estimated calibration parameters.

2. The method of claim 1 , wherein the acquiring step further comprises acquiring at least two full-resolution energy spectrum scans at different voltages of the X-ray tube, and
the estimating step further comprises applying the at least two full-resolution energy spectrum scans to the inputs of the trained neural network.

3. The method of claim 1 , wherein in the estimating step, the acquired first energy spectrum applied to the inputs of the trained neural network includes six or fewer energy bins.

4. The method in claim 1 , further comprising training the neural network based on particular mechanical or spectral features of the detector.

5. The method of claim 4 , further comprising training the neural network based on spatial variation by additionally inputting a position of a channel into the neural network.

6. The method of claim 4 , further comprising training the neural network based on bow-tie effects due to a bow-tie filter, wherein the first energy spectrum variation is accounted for by additionally inputting a thickness and a material of the bow-tie filter into the neural network.

7. The method in claim 4 , further comprising training the neural network based on a type of bow-tie filter used when acquiring the first energy spectrum.

8. The method in claim 7 , wherein the neural network has an input indicating a type of a bow-tie filter used when acquiring the first energy spectrum.

9. The method in claim 7 , wherein the neural network has inputs indicating a thickness and a material of a bow-tie filter used when acquiring the first energy spectrum.

10. The method in claim 4 , further comprising training the neural network based on a type of spectral filter used when acquiring the first energy spectrum.

11. The method of claim 1 , further comprising:
repeating the acquiring and estimating steps in order to detect changes in the calibration parameters, and
recalibrating the detector based on the changes in the calibration parameters and established tolerances.

12. The method of claim 1 , further comprising:
augmenting the first energy spectrum to increase an amount of training data; and
training the neural network using the augmented training data.

13. The method of claim 1 , wherein, for each of the plurality of channels, the calibration parameters are a gain and an offset defining a linear function relating measured energy to calibrated energy.

14. The method of claim 1 , further comprising:
training the neural network to output the calibration parameters using training data determined using two or more measurements selected from isotope peak positions, K-edge absorption features, and K-edge emission peaks.

15. A method for calibrating a detector, the method comprising:
acquiring a first energy spectrum obtained from a scan using an X-ray tube as a source of radiation;
training a plurality of different neural networks for corresponding different spatial regions of the detector;
estimating calibration parameters, for each of a plurality of channels of the detector, by applying the acquired first energy spectrum to corresponding inputs of the plurality of trained neural networks, which output the calibration parameters for the corresponding different spatial regions; and
calibrating each of the plurality of channels using the estimated calibration parameters.

16. An apparatus for calibrating a detector, the apparatus comprising:
processing circuitry configured to
acquire a first energy spectrum obtained from a scan using an X-ray tube as a source of radiation;
estimate calibration parameters, for each of a plurality of channels of the detector, by applying the acquired first energy spectrum to inputs of a trained neural network device that outputs the calibration parameters; and
calibrate each of the plurality of channels using the estimated calibration parameters.

17. The apparatus of claim 16 , wherein the processing circuitry is further configured to acquire at least two full-resolution energy spectrum scans at different voltages of the X-ray tube.

18. The apparatus of claim 16 , wherein the processing circuitry is further configured to acquire the first energy spectrum for six or fewer energy bins.

19. The apparatus of claim 16 , wherein the processing circuitry is further configured to train the neural network based on particular mechanical or spectral features of the detector.

20. The apparatus of claim 19 , wherein the processing circuitry is further configured to train the neural network based on spatial variation by additionally inputting a position of a channel into the neural network.

21. The apparatus of claim 19 , wherein the processing circuitry is further configured to train the neural network based on bow-tie effects due to a bow-tie filter, wherein the first energy spectrum variation is accounted for by additionally inputting a thickness and a material of the bow-tie filter into the neural network.

22. The apparatus of claim 16 , wherein the processing circuitry is further configured to recalibrate the detector based on changes in calibration parameters.

23. The apparatus of claim 16 , wherein the processing circuitry is further configured to
augment the first energy spectrum to increase an amount of training data; and
train the neural network using the augmented training data.

24. The apparatus of claim 16 , wherein the processing circuitry is further configured to estimate, for each of the plurality of channels, a gain and an offset as the calibration parameters defining a linear function relating measured energy to calibrated energy.

Assignees

• Canon Medical Systems Corp

Inventors

• Burr Kent C
• Markov Nikolay
• Qiang Yi

Classifications

• G06N3/08 Primary

  Learning methods · CPC title

• A61B6/4035

  the source being combined with a filter or grating · CPC title

• A61B6/585

  Calibration of detector units · CPC title

• A61B6/4241

  using energy resolving detectors, e.g. photon counting · CPC title

• G01T7/005 Primary

  calibration techniques (stabilization of spectrometer G01T1/40) · CPC title