System and method for assessing image quality

1. A method of using a machine learning image classification algorithm, comprising:

receiving a trained autoencoder model trained on a training data set, wherein the training data set includes training images classified as being of a first class, wherein the trained autoencoder model comprises one or more convolution layers, each convolution layer comprising a plurality of filters, wherein subsets of convolution layers are organized as comprising convolution passes, wherein each convolution pass is followed by one of an up-sampling pass or a down-sampling pass;

deriving an encoder model from the trained autoencoder model;

using the encoder model to classify input images, wherein each respective input image is classified as being either of the first class or one or more other classes; and

outputting a classification for one or more of the input images.

2. The method of claim 1 , wherein the training images classified as being of the first class correspond to previously acquired images determined to be of acceptable diagnostic quality.

3. The method of claim 1 , wherein the one or more other classes comprise classes determined to be of unacceptable diagnostic quality.

4. The method of claim 1 , wherein the encoder model comprises half of the trained autoencoder model.

5. The method of claim 1 , wherein each respective input image is classified based on a degree of distance from the training images of the training data set as determined by true pixels counts.

6. The method of claim 1 , wherein one or both of the training images and reference images are one of two-dimensional images, three-dimensional volume images, or four-dimensional time-varying volume images.

7. The method of claim 1 , wherein one or both of the training images and input images comprise images acquired using a magnetic resonance imaging (MRI) system.

8. The method of claim 1 , further comprising:

using the encoder model to encode training images of the training data set;

determining a cluster statistic for the training data set based on the encoded training images;

determining true-pixels counts (TPC) for the training images of the training data set using the cluster statistic;

determining one or more TPC thresholds based on the determined TPC for the training images of the training data set; and

comparing a TPC of a respective input image to the one or more TPC thresholds to classify the respective input image as being either of the first class or the one or more other classes.

9. The method of claim 8 , further comprising:

using one or more additional classes of images to determine one or both of the cluster statistics or the TPC threshold.

10. The method of claim 1 , wherein an output of the encoder model is a convolution layer output.

11. The method of claim 10 , wherein the convolution layer output is from a last convolution pass from a sequence of convolutions followed by down-sampling in the trained autoencoder model.

12. A method for classifying images, comprising:

encoding training images of a training data set using an encoder model, wherein the encoder model comprises a subset of a trained autoencoder model;

determining a cluster statistic for the training data set based on the encoded training images;

determining true-pixels counts (TPC) for the training images of the training data set using the cluster statistic;

determining one or more TPC thresholds based on the determined TPC for the training images of the training data set;

comparing a TPC of a respective input image to the one or more TPC thresholds;

based on the comparison to the TPC of the respective input image to the one or more TPC thresholds, classifying the respective input image, wherein each respective input image is classified as being either of a first class used to train the autoencoder model or one or more other classes; and

outputting a classification of the respective input image.

13. The method of claim 12 , further comprising:

using two or more classes of images to determine one or both of the cluster statistics or the TPC threshold.

14. The method of claim 12 , wherein the autoencoder model comprises:

one or more convolution layers, each convolution layer comprising a plurality of filters, wherein subsets of convolution layers are organized as comprising convolution passes, wherein each convolution pass is followed by one of an up-sampling pass or a down-sampling pass.

15. The method of claim 12 , wherein the encoder model comprises half of the autoencoder model.

16. The method of claim 12 , wherein one or both of the training images and input images are one of two-dimensional images, three-dimensional volume images, or four-dimensional time-varying volume images.

17. The method of claim 12 , wherein act of classifying the respective input image takes into account an acceptance parameter defined for the clinical site where the respective input image is acquired.

18. The method of claim 12 , wherein the respective input image is a tumor image, wherein the first class comprises images of healthy tissue, and wherein the classification of the tumor image corresponds to one or both of a tumor sub-type or tumor severity.

19. An image classification system, comprising:

processing circuitry configured to execute one or more stored routines, wherein the routines, when executed, cause the processing circuitry to:

encode training images of a training data set using an encoder model, wherein the encoder model comprises a subset of a trained autoencoder model;

determine a cluster statistic for the training data set based on the encoded training images;

determine true-pixels counts (TPC) for the training images of the training data set using the cluster statistic;

determine one or more TPC thresholds based on the determined TPC for the training images of the training data set;

compare a TPC of a respective input image to the one or more TPC thresholds;

based on the comparison to the TPC of the respective input image to the one or more TPC thresholds, classify the respective input image, wherein each respective input image is classified as being either of a first class used to train the autoencoder model or one or more other classes; and

output a classification of the respective input image.

20. The image classification system of claim 19 , wherein the routines, when executed, further cause the processing circuitry to:

use two or more classes of images to determine one or both of the cluster statistics or the TPC threshold.

21. The image classification system of claim 19 , wherein the autoencoder model comprises:

one or more convolution layers, each convolution layer comprising a plurality of filters, wherein subsets of convolution layers are organized as comprising convolution passes, wherein each convolution pass is followed by one of an up-sampling pass or a down-sampling pass.

22. The image classification system of claim 19 , wherein the encoder model comprises half of the autoencoder model.