Automatic computerized joint segmentation and inflammation quantification in MRI

A computer-readable storage medium storing program instructions that, when executed by a processor of a computer system, cause the computer system to perform a method comprising:

• obtaining training data including MRI data generated by imaging a targeted joint in each a plurality of training subjects;
• training an automated categorizer to assign individual image slices of the training data to image-slice categories, wherein the image-slice category for each image slice is selected from a group consisting of:
• a first image-slice category in which the image slice is a superficial slice that does not include either bone or tendon tissue;
• a second image-slice category in which the image slice includes both bone and tendon tissue;
• a third image-slice category that includes all image slices not assigned to the first image-slice category or the second image-slice category;
• training a first automated classifier to identify bone tissue areas in an image slice, wherein training of the first automated classifier uses image slices from only the second image-slice category;
• training a second automated classifier to identify bone tissue areas in an image slice, wherein training of the second automated classifier uses image slices from only the third image-slice category;
• obtaining testing data including MRI data generated by imaging the targeted joint in a test subject;
• applying the trained automated categorizer and the trained first and second automated classifiers to identify bone tissue segments in the testing data, wherein applying the trained automated categorizer and the trained first and second automated classifiers includes:
• applying the trained automated categorizer to assign each of a plurality of individual image slices of the testing data to one of the image-slice categories;
• applying the trained first automated classifier to segment the individual image slices assigned by the trained automated categorizer to the second image-slice category;
• applying the trained second automated classifier to segment the individual image slices assigned by the trained automated categorizer to the third image-slice category;
• and
• performing a post-processing operation that combines segmentation results from the trained first automated classifier and the trained second automated classifier across a plurality of image slices;
• and
• quantifying inflammation in bone tissue of the test subject, wherein quantifying inflammation in bone tissue is based on the bone tissue segments identified in the testing data.

A computer system comprising:

• a memory storing program code;
• and
• a processing system coupled to the memory and configured by the program code to:
• obtain training data including MRI data generated by imaging a targeted joint in each a plurality of training subjects;
• train an automated categorizer to assign individual image slices of the training data to image-slice categories, wherein the image-slice category for each image slice is selected from a group consisting of:
• a first image-slice category in which the image slice is a superficial slice that does not include either bone or tendon tissue;
• a second image-slice category in which the image slice includes both bone and tendon tissue;
• a third image-slice category that includes all image slices not assigned to the first image-slice category or the second image-slice category;
• train a first automated classifier to identify bone tissue areas in an image slice, wherein training of the first automated classifier uses image slices from only the second image-slice category;
• train a second automated classifier to identify bone tissue areas in an image slice, wherein training of the second automated classifier uses image slices from only the third image-slice category;
• obtain testing data including MRI data generated by imaging the targeted joint in a test subject;
• apply the trained automated categorizer and the trained first and second automated classifiers to identify bone tissue segments in the testing data, wherein applying the trained automated categorizer and the trained first and second automated classifiers includes:
• applying the trained automated categorizer to assign each of a plurality of individual image slices of the testing data to one of the image-slice categories;
• applying the trained first automated classifier to segment the individual image slices assigned by the trained automated categorizer to the second image-slice category;
• applying the trained second automated classifier to segment the individual image slices assigned by the trained automated categorizer to the third image-slice category;
• and
• performing a post-processing operation that combines segmentation results from the trained first automated classifier and the trained second automated classifier across a plurality of image slices;
• and
• quantify inflammation in bone tissue of the test subject, wherein quantifying inflammation in bone tissue is based on the bone tissue segments identified in the testing data.