Method of generating a trained model that transforms forward-projected X-ray CT image data into raw projection data
A medical information processing method according to an embodiment includes: acquiring an X-ray CT image (I 1 ) and spectral information on imaging of the X-ray CT image (I 1 ); acquiring sets of distribution data (D 11 , D 12 , and D 13 ) on substances in the X-ray CT image by performing segmentation of the X-ray CT image (I 1 ) according to substance; acquiring plural sets of forward projection data (P 11 , P 12 , and P 13 ) on the respective substances by performing forward projection processes for the sets of distribution data (D 11 , D 12 , and D 13 ) on the basis of the spectral information and attenuation coefficients for the respective substances; and generating a trained model (M 1 ) by machine learning based on the plural sets of forward projection data (P 11 , P 12 , and P 13 ) and raw data (R 1 ) used in generation of the X-ray CT image (I 1 ).
1 . A medical information processing method, including:
acquiring an X-ray CT image and spectral information on imaging of the X-ray CT image, performing segmentation of the X-ray CT image according to a substance and acquiring distribution data on substances in the X-ray CT image, performing a forward projection process for the distribution data based on the spectral information and an attenuation coefficient for each substance, and acquiring plural sets of forward projection data respectively for the substances; and
generating a trained model by machine learning based on the plural sets of forward projection data and raw data used in generation of the X-ray CT image.
2 . The medical information processing method according to claim 1 , wherein
input data that are the plural sets of forward projection data and output data that are the raw data are input to a neural network, and
the trained model is generated by causing the neural network to learn to minimize an error between a sum of the plural sets of forward projection data and the raw data.
3 . The medical information processing method according to claim 2 , further including:
acquiring the trained model, another X-ray CT image different from the X-ray CT image, and spectral information, performing segmentation of the another X-ray CT image according to substance and acquiring distribution data on substances in the another X-ray CT image, performing a forward projection process for the distribution data based on the spectral information and the attenuation coefficient for each substance, and acquiring plural sets of forward projection data respectively for the substances; and
acquiring raw data corresponding to the spectral information by inputting the plural sets of forward projection data based on the another X-ray CT image into the trained model.
4 . The medical information processing method according to claim 3 , wherein the spectral information is set based on an X-ray energy value input by a user.
5 . A medical information processing apparatus, comprising:
processing circuitry configured to
acquire an X-ray CT image and spectral information on imaging of the X-ray CT image, perform segmentation of the X-ray CT image according to substance and acquires distribution data on substances in the X-ray CT image, perform a forward projection process for the distribution data based on the spectral information and an attenuation coefficient for each substance, and acquire plural sets of forward projection data respectively for the substances; and
generate a trained model by machine learning based on the plural sets of forward projection data and raw data used in generation of the X-ray CT image.
6 . A medical information processing method, wherein
acquiring a first CT image of a subject, the first CT image being acquired by a CT scan corresponding to first spectral information, and acquiring sets of distribution data on plural substances by application of a computer segmentation process to the first CT image,
acquiring a second CT image corresponding to second spectral information is by a conversion process based on the sets of distribution data on the plural substances,
outputting the second CT image to be displayed or analyzed, and
executing any one of the computer segmentation process or the conversion process based on a trained model acquired by machine learning.
7 . The medical information processing method according to claim 6 , wherein the step of acquiring the sets of distribution data on the plural substances comprises acquiring the sets of distribution data from the first CT image based on the trained model in the computer segmentation process.
8 . The medical information processing method according to claim 6 , wherein
acquiring plural sets of forward projection data by a forward projection process based on the second spectral information for each of the sets of distribution data on the plural substances,
acquiring processed plural sets of forward projection data by application of the trained model to the plural sets of forward projection data,
acquiring combined forward projection data by a combination of the processed plural sets of forward projection data, and
reconstructing the second CT image based on the combined forward projection data.
9 . The medical information processing method according to claim 6 , further comprising:
acquiring processed sets of distribution data on the plural substances by application of the trained model to each of the sets of distribution data on the plural substances in the conversion process, the processed sets of distribution data corresponding to the second spectral information, and
acquiring the second CT image corresponding to the second spectral information by combination of the processed sets of distribution data on the plural substances.
10 . The medical information processing method according to claim 6 , further comprising:
in the conversion process, acquiring plural sets of forward projection data by a forward projection process based on the second spectral information for each of the sets of distribution data on the plural substances,
acquiring raw data corresponding to the second spectral information by application of the trained model to the plural sets of forward projection data, and
reconstructing the second CT image based on the raw data.