Ultrasound diagnostic apparatus, ultrasound diagnostic method, non-transitory computer-readable recording medium storing program therein, and model training apparatus
Provided is an ultrasound diagnostic apparatus including a transceiver that transmits and receives first ultrasound of a low acoustic output and second ultrasound of a high acoustic output, and a first hardware processor that generates the ultrasound image based on a first reception signal acquired by transmitting and receiving the first ultrasound. The first hardware processor includes a learning model formed of a neural network, and generates the ultrasound image based on an output result of the learning model.
1. An ultrasound diagnostic apparatus that generates and displays an ultrasound image based on a reception signal corresponding to reflection ultrasound reflected in a subject, the ultrasound diagnostic apparatus comprising:
a transceiver that transmits and receives first ultrasound of a low acoustic output and second ultrasound of a high acoustic output; and
a first hardware processor that generates the ultrasound image based on a first reception signal acquired by transmitting and receiving the first ultrasound;
a second hardware processor that generates the ultrasound image based on a second reception signal acquired by transmitting and receiving the second ultrasound;
a first power supply formed of a battery;
a second power supply that receives power from an external power supply; and
a third hardware processor, wherein the transceiver transmits and receives the first ultrasound and the third hardware processor selects the first hardware processor in a case where the driving power is supplied from the first power supply, and the transceiver transmits and receives the second ultrasound and the third hardware processor selects the second hardware processor in a case where the driving power is supplied from the second power supply, wherein
the first hardware processor includes a learning model formed of a neural network utilizing a machine learning algorithm,
the learning model is acquired by machine learning using teaching data formed of a pair of low acoustic output data acquired by transmitting and receiving the first ultrasound and high acoustic output data acquired by transmitting and receiving the second ultrasound, and
the teaching data is generated based on reception signals acquired by alternately and continuously transmitting the first ultrasound and the second ultrasound with respect to the same object.
2. The ultrasound diagnostic apparatus according to claim 1 , wherein the third hardware processor selects the first hardware processor in a case where a power mode is a low power mode.
3. The ultrasound diagnostic apparatus according to claim 1 , wherein the teaching data is sound ray data before image conversion, the sound ray data spreading in an axial direction and in an azimuth direction.
4. The ultrasound diagnostic apparatus according to claim 3 , wherein the teaching data further spreads in an elevation direction.
5. The ultrasound diagnostic apparatus according to claim 1 , wherein the teaching data is image data after image conversion of the reception signal.
6. The ultrasound diagnostic apparatus according to claim 1 , further comprising:
a fourth hardware processor that trains the learning model; and
a first memory that stores a neural network similar to the learning model, wherein
the fourth hardware processor
acquires the teaching data,
optimizes an adjustment value of the neural network by using the teaching data, and
outputs the adjustment value of the neural network, wherein
the adjustment value is reflected in the learning model.
7. The ultrasound diagnostic apparatus according to claim 6 , wherein
the fourth hardware processor includes:
a generator that is formed of the neural network and generates output data from the low acoustic output data, and
a discriminator that compares the output data with the high acoustic output data and determines authenticity of the output data, wherein
the adjustment value is optimized while a determination result of the discriminator is back-propagated to the generator and the discriminator and the generator and the discriminator are alternately caused to contest with each other.
8. The ultrasound diagnostic apparatus according to claim 6 , wherein the fourth hardware processor generates the teaching data based on the reception signals, which are continuously acquired in units of frames.
9. The ultrasound diagnostic apparatus according to claim 6 , wherein the fourth hardware processor generates the teaching data based on the reception signals, which are continuously acquired in units of sound rays.
10. The ultrasound diagnostic apparatus according to claim 6 , wherein the teaching data generated by the fourth hardware processor is transferred to a cloud storage on a network.
11. The ultrasound diagnostic apparatus according to claim 6 , wherein the fourth hardware processor acquires the teaching data from a cloud storage on a network.
12. A model training apparatus that trains the learning model to be implemented in the ultrasound diagnostic apparatus according to claim 1 , the model training apparatus comprising:
a fifth hardware processor; and
a second memory that stores a neural network similar to the learning model, wherein
the fifth hardware processor
acquires the teaching data,
optimizes an adjustment value of the neural network by using the teaching data, and
outputs the adjustment value of the neural network.
13. The ultrasound diagnostic apparatus according to claim 1 , wherein the first hardware processor performs image processing using the learning model on the ultrasound image based on the first reception signal to increase an image quality of the ultrasound image based on the first reception signal.
14. The ultrasound diagnostic apparatus according to claim 1 , wherein
the first hardware processor generates the ultrasound image based on the first reception signal using an image generator;
the second hardware processor generates the ultrasound image based on the second reception signal using the image generator or another image generator, wherein the ultrasound image based on the first reception signal has a lower image quality than the ultrasound image based on the second reception signal; and
the first hardware processor performs image processing using the learning model to increase an image quality of the ultrasound image based on the first reception signal.
15. The ultrasound diagnostic apparatus according to claim 1 , wherein
the transmission voltage of the first ultrasound is smaller than that of the second ultrasound.
16. An ultrasound diagnostic method in which a diagnosis is performed with an ultrasound image based on a reception signal corresponding to reflection ultrasound reflected in a subject, the ultrasound diagnostic method comprising:
transmitting and receiving first ultrasound of a low acoustic output or second ultrasound of a high acoustic output; and
generating, using a first hardware processor, the ultrasound image based on a first reception signal acquired by transmitting and receiving the first ultrasound, wherein the ultrasound image is generated by utilizing a learning model and based on an output result of the learning model when the first reception signal is input, the learning model being formed of a neural network;
generating, using a second hardware processor, the ultrasound image based on a second reception signal acquired by transmitting and receiving the second ultrasound; and
selecting the first hardware processor to generate the ultrasound image based on the first reception signal or the second hardware processor to generate the ultrasound image based on the second reception signal,
in a case where driving power is supplied from a first power supply formed of a battery, the step of transmitting and receiving includes transmitting and receiving the first ultrasound and the step of selecting includes selecting the first hardware processor, and
in a case where the driving power is supplied from a second power supply that receives power from an external power supply, the step of transmitting and receiving includes transmitting and receiving the second ultrasound and the step of selecting includes selecting the second hardware processor,
wherein the learning model is acquired by machine learning using teaching data formed of a pair of low acoustic output data acquired by transmitting and receiving the first ultrasound and high acoustic output data acquired by transmitting and receiving the second ultrasound, and
the teaching data is generated based on reception signals acquired by alternately and continuously transmitting the first ultrasound and the second ultrasound with respect to the same object.
17. A non-transitory computer-readable recording medium storing therein a program that causes a computer of an ultrasound diagnostic apparatus to execute predetermined processing, the ultrasound diagnostic apparatus generating and displaying an ultrasound image based on a reception signal corresponding to reflection ultrasound reflected in a subject, the predetermined processing comprising:
transmitting and receiving first ultrasound of a low acoustic output or second ultrasound of a high acoustic output;
generating, using a first hardware processor, the ultrasound image based on a first reception signal acquired by transmitting and receiving the first ultrasound, wherein the ultrasound image is generated by utilizing a learning model and based on an output result of the learning model when the first reception signal is input, the learning model being formed of a neural network utilizing a machine learning algorithm;
generating, using a second hardware processor, the ultrasound image based on a second reception signal acquired by transmitting and receiving the second ultrasound; and
selecting the first hardware processor to generate the ultrasound image based on the first reception signal or the second hardware processor to generate the ultrasound image based on the second reception signal,
in a case where driving power is supplied from a first power supply formed of a battery, the step of transmitting and receiving includes transmitting and receiving the first ultrasound and the step of selecting includes selecting the first hardware processor, and
in a case where the driving power is supplied from a second power supply that receives power from an external power supply, the step of transmitting and receiving includes transmitting and receiving the second ultrasound and the step of selecting includes selecting the second hardware processor,
wherein the learning model is acquired by machine learning using teaching data formed of a pair of low acoustic output data acquired by transmitting and receiving the first ultrasound and high acoustic output data acquired by transmitting and receiving the second ultrasound, and
the teaching data is generated based on reception signals acquired by alternately and continuously transmitting the first ultrasound and the second ultrasound with respect to the same object.