Data processing device, magnetic resonance imaging device, and data processing method
A data processing device according to an embodiment generates output data by inputting, to a learned model, as input data, first input data of a complex value and second input data that is data related to a complex phase, the learned model having been trained using, as input data, first input data for learning of a complex value obtained based on collected data collected by a medical image diagnostic device and second input data for learning that is data related to a complex phase of the collected data, and using, as output data, output data for learning.
1 . A data processing device, comprising:
processing circuitry configured to
generate output data by inputting, to a learned model, as input data, both first input data of a complex value and second input data that is data related to a complex phase,
the learned model having been trained using, as training input data, first input data for learning of a complex value obtained based on collected data collected by a medical image diagnostic device and second input data for learning that is data related to a complex phase of the collected data, and using, as training output data, output data for learning, wherein the medical image diagnostic device is a magnetic resonance imaging device or an ultrasonic diagnostic apparatus.
2 . The data processing device according to claim 1 , wherein
the medical image diagnostic device is a magnetic resonance imaging device, and
the second input data for learning is data obtained by multiplying a shift in resonance frequency at each location by a relative time of an echo center of a collection pulse sequence performed by the magnetic resonance imaging device, with respect to a zero phase.
3 . The data processing device according to claim 1 , wherein
the medical image diagnostic device is a magnetic resonance imaging device, and
the second input data for learning is data representing a shift in resonance frequency at each location.
4 . The data processing device according to claim 1 , wherein
the medical image diagnostic device is a magnetic resonance imaging device, and
the second input data for learning is data representing a shift in resonance frequency at each location, and data obtained by multiplying the shift in the resonance frequency at each location by a relative time of an echo center of a collection pulse sequence performed by the magnetic resonance imaging device, with respect to a zero phase.
5 . The data processing device according to claim 2 , wherein the shift in the resonance frequency is obtained based on scanning for estimating a shimming map performed by the magnetic resonance imaging device.
6 . The data processing device according to claim 1 , wherein the output data for learning is data from which noise is reduced compared to the first input data for learning.
7 . The data processing device according to claim 1 , wherein the output data for learning is data representing segment information at each location of the first input data for learning.
8 . The data processing device according to claim 1 , wherein the learned model is a learned model that outputs the output data by taking data consistency into consideration.
9 . The data processing device according to claim 1 , wherein the learned model is a learned model to which an automated transform by manifold approximation (AUTOMAP) method is applied.
10 . A magnetic resonance imaging device that collects data, the magnetic resonance imaging device comprising:
sequence control circuitry configured to execute a pulse sequence, and
a processing circuit configured to generate output data by inputting both first input data of a complex value and second input data that is data related to a complex phase obtained by the pulse sequence as input data, to a learned model that has been trained using first input data for learning of a complex value obtained based on the collected data and second input data for learning that is data related to a complex phase of the collected data as training input data, and using output data for learning as training output data.
11 . A data processing method, comprising:
generating output data by inputting, to a learned model, as input data, both first input data of a complex value and second input data that is data related to a complex phase,
the learned model having been trained using, as training input data, first input data for learning of a complex value obtained based on collected data collected by a medical image diagnostic device and second input data for learning that is data related to a complex phase of the collected data, and using, as training output data, output data for learning, wherein the medical image diagnostic device is a magnetic resonance imaging device or an ultrasonic diagnostic apparatus.