System and methods for medical image quality assessment using deep neural networks

1. A method, comprising:

acquiring a medical image of an anatomical region;

mapping the medical image to a positional attribute of an anatomical feature using a trained deep neural network;

determining an image quality metric based on the positional attribute of the anatomical feature;

determining if the image quality metric satisfies an image quality criterion; and

responding to the image quality metric not satisfying the image quality criterion by:

displaying the medical image, the image quality metric, and an indication of the unsatisfied image quality criterion, via a display device,

wherein mapping the medical image to the positional attribute of the anatomical feature using the trained deep neural network comprises:

mapping the medical image to a segmentation mask of the anatomical feature; and

determining one or more of a length, a width, a center point, an area, and an orientation of the anatomical feature, based on the segmentation mask.

2. The method of claim 1 , the method further comprising:

responding to the image quality metric satisfying the image quality criterion by:

displaying the medical image, the image quality metric, and an indication of the satisfied image quality criterion, via the display device.

3. The method of claim 1 , wherein the positional attribute is one or more of a location of the anatomical feature with respect to the medical image, an orientation of the anatomical feature with respect to the medical image, an area of the anatomical feature with respect to the medical image, and a classification score indicating a probability of inclusion of the anatomical feature within a field of view of the medical image.

4. The method of claim 1 , wherein the image quality metric includes one or more of a costophrenic angle, a cardiothoracic ratio, a mediastinal width, an angle of rotation of an imaging subject with respect to a projection plane of the medical image, and an extent of inspiration of lungs of the imaging subject.

5. The method of claim 1 , wherein the trained deep neural network, the image quality metric, and the image quality criterion, are indexed in non-transitory memory based on an imaging protocol used to acquire the medical image.

6. The method of claim 1 , wherein the image quality metric is an angle of rotation of an imaging subject with respect to a projection plane of the medical image, and wherein determining the image quality metric based on the positional attribute of the anatomical feature includes:

determining a first distance between a first location of a first anatomical feature and a second location of a second anatomical feature;

determining a second distance between the second location of the second anatomical feature and a third location of a third anatomical feature;

determining a ratio between the first distance and the second distance; and

determining the angle of rotation of the imaging subject with respect to the projection plane based on the ratio.

7. The method of claim 6 , wherein determining if the image quality metric satisfies the image quality criterion comprises:

accessing a pre-determined rotation range;

comparing the angle of rotation with the pre-determined rotation range; and

responding to the angle of rotation not falling within the pre-determined rotation range by updating a status of the image quality criterion to indicate the angle of rotation does not match the pre-determined rotation range.

8. A method, comprising:

acquiring a medical image of an anatomical region;

mapping the medical image to a positional attribute of an anatomical feature using a trained deep neural network;

determining an image quality metric based on the positional attribute of the anatomical feature;

determining if the image quality metric satisfies an image quality criterion; and

responding to the image quality metric not satisfying the image quality criterion by:

displaying the medical image, the image quality metric, and an indication of the unsatisfied image quality criterion, via a display device,

wherein the image quality metric is an extent of clipping of the anatomical feature, and wherein determining if the image quality metric satisfies the image quality criterion comprises:

determining if the extent of clipping of the anatomical feature is below a threshold extent of clipping; and

responding to the extent of clipping of the anatomical feature being below the threshold extent of clipping by:

setting a status of the image quality criterion to a value indicating the criterion is satisfied.

9. The method of claim 8 , the method further comprising:

responding to the image quality metric satisfying the image quality criterion by:

displaying the medical image, the image quality metric, and an indication of the satisfied image quality criterion, via the display device.

10. The method of claim 8 , wherein the positional attribute is one or more of a location of the anatomical feature with respect to the medical image, an orientation of the anatomical feature with respect to the medical image, an area of the anatomical feature with respect to the medical image, and a classification score indicating a probability of inclusion of the anatomical feature within a field of view of the medical image.

11. The method of claim 8 , wherein the trained deep neural network, the image quality metric, and the image quality criterion, are indexed in non-transitory memory based on an imaging protocol used to acquire the medical image.

12. A method, comprising:

receiving an imaging protocol selection;

acquiring a medical image based on the imaging protocol selection;

mapping the medical image to a first positional attribute of a first anatomical feature using a trained deep neural network;

mapping the medical image to a second positional attribute of a second anatomical feature using a second trained deep neural network;

determining an image quality metric based on the first positional attribute of the first anatomical feature and the second positional attribute of the second anatomical feature;

determining if the image quality metric satisfies an image quality criterion; and

responding to the image quality metric satisfying the image quality criterion by:

displaying the medical image, the image quality metric, and an indication of the satisfied image quality criterion, via a display device,

wherein the first positional attribute is a first segmentation mask of the first anatomical feature, and the second positional attribute is a second segmentation mask of the second anatomical feature, and

wherein the image quality metric is an extent of overlap between the first anatomical feature and the second anatomical feature, wherein the extent of overlap between the first anatomical feature and the second anatomical feature is determined based on an extent of intersection between the first segmentation mask and the second segmentation mask.

13. The method of claim 12 , wherein the first trained deep neural network and the second trained deep neural network share one or more feature encoding layers.

14. The method of claim 12 , wherein determining if the image quality metric satisfies the image quality criterion comprises determining if the extent of overlap between the first anatomical feature and the second anatomical feature is less than a threshold extent of overlap.