Search

CN-122023598-A - Cone beam CT scanning image rapid reconstruction system and method based on deep learning

CN122023598ACN 122023598 ACN122023598 ACN 122023598ACN-122023598-A

Abstract

The invention relates to the technical field of CT image reconstruction, and discloses a cone beam CT scanning image rapid reconstruction system and method based on deep learning, wherein the system comprises a double-ray generation module, a scanning module, a deep learning reconstruction module and a result output module; the dual-ray generation module comprises two ray generators, wherein a beam outlet of one ray generator is provided with 0.1mm low-atomic-number material aluminum for filtering an X-ray extremely low energy part, and the other beam outlet is provided with 0.1mm high-atomic-number material copper for attenuating the X-ray extremely low energy part. According to the invention, dual-energy spectrum projection data are acquired through a dual-ray generator, PMMA and aluminum are used as base materials for fitting and solving tissue linear attenuation coefficients, background interference is corrected by combining air reference data, the error of a reconstructed CT value in a soft tissue area is less than or equal to +/-10 HU, and the error of a cortical bone area is less than or equal to +/-30 HU, and compared with the traditional CBCT, the method is more accurate, and a reliable foundation is laid for preventing the calculation of the aspect ratio.

Inventors

  • GU LONG
  • Weng xing
  • Su Xingkang
  • WANG GUAN
  • HUANG SHENGCONG
  • ZHANG LU

Assignees

  • 福建睿斯科医疗技术有限公司

Dates

Publication Date
20260512
Application Date
20260413

Claims (10)

  1. 1. The CT cone beam scanning image rapid reconstruction system based on the deep learning is characterized by comprising a double-ray generation module, a scanning module, a deep learning reconstruction module and a result output module; The dual-ray generation module comprises two ray generators, wherein a beam outlet of one ray generator is provided with 0.1mm low-atomic-number material aluminum for filtering an X-ray extremely low energy part, and the other beam outlet is provided with 0.1mm high-atomic-number material copper for attenuating the X-ray extremely low energy part; the scanning module is used for carrying out single rotation scanning on the patient after the positioning is completed, and synchronously acquiring CBCT original projection data of two groups of different energy spectrums; The deep learning reconstruction module takes PMMA and aluminum as base materials, inputs two groups of original projection data into a pre-trained deep learning model, and solves the weight of the base materials through model fitting And Calculating the linear attenuation coefficient of the human tissue by combining the linear attenuation coefficient of the base material, and substituting the linear attenuation coefficient into a CT value formula to generate an accurate CT image; The result output module is used for outputting reconstructed CT images and corresponding substance blocking power ratio and particle beam path equivalent water depth data.
  2. 2. The rapid reconstruction system of CT cone beam scanning image based on deep learning of claim 1, wherein the thickness of the base material PMMA is in the range of 4-28mm, the thickness of the aluminum is in the range of 3-15mm, and the samples for model pre-training comprise PMMA, aluminum and two-by-two combined samples with different thicknesses.
  3. 3. The rapid reconstruction system of CT cone beam scanning image based on deep learning of claim 1, wherein when said deep learning model is pre-trained, air is scanned to obtain reference data Wherein For air scan reference projection data when energy is E, When the energy is E, the incident ray intensity of the air scanning, When the energy is E, the transmitted ray intensity of the object scanning is used for correcting the background interference of the projection data.
  4. 4. The fast reconstruction system for CT cone-beam scanning images based on deep learning of claim 1, wherein said basis weight And Is fitted with the formula of 、 Wherein Fitting the sample scan data to a fixed coefficient, wherein The pixel values of the set of projection data are filtered for aluminum, Pixel values for the copper attenuation group projection data.
  5. 5. The fast reconstruction system for CT cone-beam scanning images based on deep learning as set forth in claim 1, wherein said linear attenuation coefficient of human tissue is calculated by the formula Calculated, wherein 、 The linear attenuation coefficients of PMMA and aluminum at energy E, respectively.
  6. 6. The rapid reconstruction system for CT cone-beam scanning images based on deep learning of claim 1, wherein the CT value formula is Wherein Is the linear attenuation coefficient of water.
  7. 7. The rapid reconstruction system of CT cone beam scanning images based on deep learning of claim 1, wherein the scanning module has a single rotation scanning angle of 360 degrees, and the scanning time is less than or equal to 60s, so as to meet the requirements of rapid positioning verification before particle treatment.
  8. 8. The rapid reconstruction system of CT cone beam scanning images based on deep learning of claim 1, wherein the substance blocking power ratio output by the result output module is calculated based on a preset corresponding relation between a reconstructed CT value and human tissue blocking power, and the equivalent water depth error is less than or equal to +/-1 mm.
  9. 9. A fast reconstruction method of a CT cone beam scanning image based on deep learning, which is applied to the system as claimed in any one of claims 1 to 8, and is characterized in that: s1, synchronous scanning and acquisition, namely carrying out single rotation scanning on a patient positioned before treatment by using a double-ray generator, wherein one ray generator outputs beams through a 0.1mm aluminum filter sheet, the other beam is outputted through a 0.1mm copper attenuation sheet, and the two ray generators synchronously work and synchronously acquire CBCT original projection data of two groups of different energy spectrums; S2, model pre-training, namely selecting PMMA as a human soft tissue equivalent base material and aluminum as a human cortical bone equivalent base material, constructing a base material linear attenuation coefficient database, and pre-training a deep learning model by utilizing the scanning data of cuboid PMMA, aluminum and combined samples of the PMMA and the aluminum with different thicknesses and combining air scanning reference data; S3, weight fitting and solving, namely inputting two groups of original projection data into a pre-trained deep learning model, and solving the weight of a base material through model fitting 、 Wherein 、 The pixel values of the two sets of projection data respectively, Fitting coefficients for the model; s4, reconstructing CT image by using the formula Calculating linear attenuation coefficient of human tissue, substituting into CT value formula Obtaining an accurate CT value and completing image reconstruction; and S5, outputting key parameters, namely establishing a corresponding relation between a CT value and a substance blocking power ratio based on the reconstructed CT image, calculating equivalent water depth on a particle beam path and outputting the equivalent water depth.
  10. 10. The rapid reconstruction method of CT cone beam scanning image based on deep learning as set forth in claim 9, wherein the radiation transmission windows of the dual radiation generator correspond to the radiation receiving plate A and the radiation receiving plate B, respectively, and the particle beam outlet is located between the radiation receiving plates A and B and is adapted to the structure of the external irradiation particle therapy rotating system.

Description

Cone beam CT scanning image rapid reconstruction system and method based on deep learning Technical Field The invention relates to the technical field of CT image reconstruction, in particular to a CT cone beam scanning image rapid reconstruction system and method based on deep learning. Background CBCT is a computed tomography technique using cone-beam X-rays, which is different from the fan-beam of conventional CT, and is an important imaging means in the field of modern medicine and industrial detection, and its principle is that an X-ray generator rotates and scans around an illuminant with a cone-beam X-ray, and a flat panel detector receives two-dimensional projection data and reconstructs a three-dimensional image through a computer algorithm. At present, most CBCT only depends on monoenergetic X-ray scanning, energy spectrum response differences of different human tissues cannot be distinguished, and only the linear attenuation coefficient of the tissues can be estimated through an empirical formula, so that attenuation characteristics between tumor tissues and peripheral tissues or adjacent different tissues are mixed, further, the calculation error of the material blocking power ratio is more than or equal to 8%, the accuracy of prejudgment of the range of therapeutic particles such as proton heavy ions is directly influenced, and therefore, a CT cone beam scanning image rapid reconstruction system and method based on deep learning are provided. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a rapid reconstruction system and a rapid reconstruction method of CT cone beam scanning images based on deep learning, which solve the problems that most of CBCTs in the prior art only depend on single-energy X-ray scanning, energy spectrum response differences of different human tissues cannot be distinguished, linear attenuation coefficients of the tissues can be estimated only through an empirical formula, attenuation characteristics between tumor tissues and peripheral tissues or adjacent different tissues are mixed, and further, the calculation error of a substance blocking ratio is more than or equal to 8% and the particle range pre-judging precision is directly influenced. The CT cone beam scanning image rapid reconstruction system based on deep learning comprises a double-ray generation module, a scanning module, a deep learning reconstruction module and a result output module; The dual-ray generation module comprises two ray generators, wherein a beam outlet of one ray generator is provided with 0.1mm low-atomic-number material aluminum for filtering an X-ray extremely low energy part, and the other beam outlet is provided with 0.1mm high-atomic-number material copper for attenuating the X-ray extremely low energy part; the scanning module is used for carrying out single rotation scanning on the patient after the positioning is completed, and synchronously acquiring CBCT original projection data of two groups of different energy spectrums; The deep learning reconstruction module takes PMMA and aluminum as base materials, inputs two groups of original projection data into a pre-trained deep learning model, and solves the weight of the base materials through model fitting AndCalculating the linear attenuation coefficient of the human tissue by combining the linear attenuation coefficient of the base material, and substituting the linear attenuation coefficient into a CT value formula to generate an accurate CT image; The result output module is used for outputting reconstructed CT images and corresponding substance blocking power ratio and particle beam path equivalent water depth data. Further, the thickness range of the base material PMMA is 4-28mm, the thickness range of aluminum is 3-15mm, the samples used for model pre-training comprise PMMA, aluminum and two-to-two combined samples with different thicknesses, the model is pre-trained by the base material samples with different thicknesses, the weight of the base material can be quickly fitted and solved, the reconstruction time is less than or equal to 10s, the speed is increased by more than 40% compared with the traditional dual-energy CBCT reconstruction method, the aging requirement of quick positioning verification before particle therapy is met, and the continuity of diagnosis and treatment processes is improved. Furthermore, when the deep learning model is pre-trained, air is required to be scanned to obtain reference dataWhereinFor air scan reference projection data when energy is E,When the energy is E, the incident ray intensity of the air scanning,When the energy is E, the transmitted ray intensity of object scanning is used for correcting the background interference of projection data, double-energy spectrum projection data are obtained through a double-ray generator, PMMA and aluminum are used as base materials to fit and solve the linear attenuation coefficient of the tissue, the background in