Image enhancement with variable number of excitation (NEX) acquisitions accelerated using compressed sensing

1. A magnetic resonance (MR) imaging device comprising:

a body coil adapted to transmit radio-frequency (RF) signals to a subject anatomy and acquire first MR signals from the subject anatomy in a first number of acquisition (NEX) and second MR signals from the subject anatomy in a second NEX;

a controller adapted to control the body coil to acquire the first and second MR signals such that first k-space points corresponding to the first MR signals are distinct from second k-space points corresponding to the second MR signals in the k-space, and wherein the controller is further configured for compressed sensing (CS) reconstruction of the k-space images; and

an image enhancement module adapted to receive the compressed sensing (CS) reconstructed images from the controller and output images of enhanced quality;

wherein a view sharing reconstruction scheme comprises copying unacquired views in the first NEX into the second NEX and vice-versa to reconstruct images having a relatively high signal-to-noise ratio (SNR).

2. The magnetic resonance (MR) imaging device as claimed in claim 1 , wherein the body coil is a single channel body coil.

3. The magnetic resonance (MR) imaging device as claimed in claim 1 , wherein the controller includes a first neural network and a second neural network, wherein the first neural network is connected to the second neural network through a processor.

4. The magnetic resonance (MR) imaging device as claimed in claim 3 , wherein the first neural network may be a shallow learning network, or a deep learning network and the second neural network may be a deep learning network.

5. The magnetic resonance (MR) imaging device as claimed in claim 3 , the image enhancement module is located on the first neural network.

6. The magnetic resonance (MR) imaging device as claimed in claim 3 , wherein the first neural network is adapted to receive MR signals acquired by the body coil and configured to generate an image having a relatively high quality based on the MR signals acquired by the body coil, wherein the first neural network is trained with the MR signals acquired by the body coil as inputs and co-registered higher channel count images as outputs.

7. The magnetic resonance (MR) imaging device as claimed in claim 3 , wherein the high-quality images obtained from the image enhancement module are stored in an image database and a second neural network which is a trained deep learning network is connected to the image database through a processor to assess the quality of the images and provide feedback to the first neural network and the processor.

8. The magnetic resonance (MR) imaging device as claimed in claim 3 , wherein the second neural network periodically updates the first neural network.

9. The magnetic resonance (MR) imaging device as claimed in claim 3 , wherein the first neural network is fine-tuned based on quality of images associated with the imaging devices.

10. The magnetic resonance (MR) imaging device as claimed in claim 1 , wherein image enhancement module comprises an adder-module that averages the images.

11. The magnetic resonance (MR) imaging device as claimed in claim 1 , wherein the image enhancement module is a neural network architecture which is trained with input body coil images against their co-registered higher channel count images as labels.

12. A method of magnetic resonance (MR) imaging comprising:

transmitting electromagnetic waves using a body coil to a subject anatomy;

acquiring an electromagnetic signal from the subject anatomy using the body coil;

acquiring from the body coil distinct k-space images at a first number of excitations (NEX) and at a second number of excitations (NEX) that are different than the first number of excitations (NEX) by a controller;

carrying out compressed sensing (CS) reconstruction of k-space images obtained at first number of excitations (NEX) and at second number of excitations (NEX) to generate compressed sensing (CS) reconstructed images; and

obtaining enhanced quality images by processing the CS reconstructed images using an image enhancement module;

wherein the method further comprises using a reduced number of channel count as compared to any higher channel count surface coil.

13. The method of magnetic resonance (MR) imaging as claimed in claim 12 , wherein obtaining enhanced quality images by processing the compressed sensing (CS) reconstructed images comprises applying a neural network to get enhanced reconstructed images.

14. The method of magnetic resonance (MR) imaging as claimed in claim 13 , wherein applying the neural network to get enhanced reconstructed images comprises receiving body coil images using a first neural network having an image enhancement module to generate high quality images and storing the high-quality images in an image database.

15. The method of magnetic resonance (MR) imaging as claimed in claim 13 , wherein applying a neural network to get enhanced reconstructed images comprises training a second neural network with body coil images, and assessing the high-quality images stored in the image database to the images stored on the trained second neural network and providing feedback to the first neural network.

16. The method of magnetic resonance (MR) imaging as claimed in claim 13 , further comprising fine tuning the first neural network based on the quality of images associated with the imaging device.

17. The method of magnetic resonance (MR) imaging as claimed in claim 12 , wherein any parameter combination of repetition time (TR), echo time (TE), and echo tail length (ETL) is used in generating a variation in contrast.

18. The method of magnetic resonance (MR) imaging as claimed in claim 12 , wherein compressed sensing (CS) reconstruction of k-space images includes variable acceleration factors for each number of excitations (NEX).

19. A magnetic resonance (MR) imaging device comprising:

a body coil adapted to transmit radio-frequency (RF) signals to a subject anatomy and acquire first MR signals from the subject anatomy in a first number of acquisition (NEX) and second MR signals from the subject anatomy in a second NEX;

a controller adapted to control the body coil to acquire the first and second MR signals such that first k-space points corresponding to the first MR signals are distinct from second k-space points corresponding to the second MR signals in the k-space, and wherein the controller is further configured for compressed sensing (CS) reconstruction of the k-space images;

an image enhancement module adapted to receive the compressed sensing (CS) reconstructed images from the controller and output images of enhanced quality;

wherein the controller includes a first neural network and a second neural network, wherein the first neural network is connected to the second neural network through a processor; and

wherein the first neural network is adapted to receive MR signals acquired by the body coil and configured to generate an image having a relatively high quality based on the MR signals acquired by the body coil, wherein the first neural network is trained with the MR signals acquired by the body coil as inputs and co-registered higher channel count images as outputs.

20. A magnetic resonance (MR) imaging device comprising:

a body coil adapted to transmit radio-frequency (RF) signals to a subject anatomy and acquire first MR signals from the subject anatomy in a first number of acquisition (NEX) and second MR signals from the subject anatomy in a second NEX;

a controller adapted to control the body coil to acquire the first and second MR signals such that first k-space points corresponding to the first MR signals are distinct from second k-space points corresponding to the second MR signals in the k-space, and wherein the controller is further configured for compressed sensing (CS) reconstruction of the k-space images;

an image enhancement module adapted to receive the compressed sensing (CS) reconstructed images from the controller and output images of enhanced quality; and

wherein the image enhancement module is a neural network architecture which is trained with input body coil images against their co-registered higher channel count images as labels.