Machine learning techniques for MRI processing using regional scoring of non-parametric voxel integrity rankings
Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing predictive recommendations using an MRI acquisition set associated with a common target object. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform predictive recommendations based at least in part on an MRI set and utilizing one or more of techniques using image preprocessing models, techniques using image segmentation models, techniques using voxel integrity score generation machine learning models, and techniques using integrity score normalization models.
1. A computer-implemented method for generating one or more predictive recommendations based at least in part on a magnetic resonance imaging (MRI) set characterized by one or more MRI images that are associated with a common target object having a common target object type, the computer-implemented method comprising:
generating, using one or more processors and an image preprocessing model for the common target object type, and based at least in part on the MRI set, a standardized MRI set;
generating, using the one or more processors and an image segmentation model, and based at least in part on the standardized MRI set: (i) a plurality of disjoint image regions of the standardized MRI set, and (ii) for each disjoint image region, a subset of a group of image voxels of the standardized MRI set;
for each image voxel, using the one or more processors:
generating, using a voxel integrity score generation machine learning model, and based at least in part on a voxel input representation for the image voxel, a voxel integrity score for the image voxel, and
generating, using an integrity score normalization model characterized by a normalization space that is defined by one or more normalization variables associated with the common target object, and based at least in part on the voxel integrity score for the image voxel, a non-parametric integrity ranking for the image voxel;
generating, using the one or more processors, a group of region scores based at least in part on each non-parametric integrity ranking, wherein:
each region score is associated with a respective disjoint image region and is generated based at least in part on each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and
the group of region scores comprise, for each disjoint image region: (i) a normality-distribution region score that is generated based at least in part on an average measure of each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and (ii) a normality-ratio region score that is generated based at least in part on a ratio of those image voxels that are in the disjoint image region and that are associated with non-threshold-satisfying non-parametric integrity rankings and those image voxels that are in the disjoint image region and that are associated with threshold-satisfying non-parametric integrity rankings;
generating, using the one or more processors and a predictive recommendation model and based at least in part on the group of region scores, the one or more predictive recommendations, wherein:
the predictive recommendation model is configured to map the group of region scores to a selected subset of a plurality of candidate predictive recommendations,
the predictive recommendation model is characterized by a plurality of predictive recommendation mapping rules comprising one or more normality-distribution predictive recommendation mapping rules and one or more normality-ratio predictive recommendation mapping rules,
each normality-distribution predictive recommendation mapping rule is characterized by a normality-distribution region score threshold, and
each normality-ratio predictive recommendation mapping rule is characterized by a normality-ratio region score threshold; and
performing, using the one or more processors, one or more prediction-based actions based at least in part on the selected subset.
2. The computer-implemented method of claim 1 , wherein the common target object is a brain region, and the plurality of disjoint image regions comprise a plurality of defined brain regions.
3. The computer-implemented method of claim 1 , wherein the one or more MRI images comprise a T1-weighted MRI image, a T2-weighted MRI image, a T2*-weighted MRI image, and a Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
4. The computer-implemented method of claim 3 , wherein each voxel input representation for a particular image voxel is a four-dimensional vector comprising a first value for the image voxel as determined based at least in part on the T1-weighted MRI image, a second value for the image voxel as determined based at least in part on the T2-weighted MRI image, a third value for the image voxel as determined based at least in part on the T2*-weighted MRI image, and a fourth value for the image voxel as determined based at least in part on the Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
5. The computer-implemented method of claim 1 , wherein the voxel integrity score generation machine learning model is a Gaussian mixture model cluster analysis machine learning model.
6. The computer-implemented method of claim 1 , wherein the one or more normalization variables comprise an age variable.
7. The computer-implemented method of claim 1 , wherein the normalization space comprises, for each historical MRI set of one or more historical MRI sets that are associated with the one or more normalization variables, a group of historical voxel integrity scores for the historical MRI set that are generated based at least in part on output of processing the historical MRI set using one or more of the image preprocessing model, the image segmentation model, and the voxel integrity score generation machine learning model.
8. An apparatus for generating one or more predictive recommendations based at least in part on a magnetic resonance imaging (MRI) set characterized by one or more MRI images that are associated with a common target object having a common target object type, the apparatus comprising at least one processor and at least one memory including program code, the at least one memory and the program code configured to, with the processor, cause the apparatus to at least:
generate, using an image preprocessing model for the common target object type, and based at least in part on the MRI set, a standardized MRI set;
generate, using an image segmentation model, and based at least in part on the standardized MRI set: (i) a plurality of disjoint image regions of the standardized MRI set, and (ii) for each disjoint image region, a subset of a group of image voxels of the standardized MRI set;
for each image voxel:
generate, using a voxel integrity score generation machine learning model, and based at least in part on a voxel input representation for the image voxel, a voxel integrity score for the image voxel, and
generate, using an integrity score normalization model characterized by a normalization space that is defined by one or more normalization variables associated with the common target object, and based at least in part on the voxel integrity score for the image voxel, a non-parametric integrity ranking for the image voxel;
generate a group of region scores based at least in part on each non-parametric integrity ranking, wherein:
each region score is associated with a respective disjoint image region and is generated based at least in part on each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and
the group of region scores comprise, for each disjoint image region: (i) a normality-distribution region score that is generated based at least in part on an average measure of each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and (ii) a normality-ratio region score that is generated based at least in part on a ratio of those image voxels that are in the disjoint image region and that are associated with non-threshold-satisfying non-parametric integrity rankings and those image voxels that are in the disjoint image region and that are associated with threshold-satisfying non-parametric integrity rankings;
generate, using a predictive recommendation model and based at least in part on the group of region scores, the one or more predictive recommendations, wherein:
the predictive recommendation model is configured to map the group of region scores to a selected subset of a plurality of candidate predictive recommendations,
the predictive recommendation model is characterized by a plurality of predictive recommendation mapping rules comprising one or more normality-distribution predictive recommendation mapping rules and one or more normality-ratio predictive recommendation mapping rules,
each normality-distribution predictive recommendation mapping rule is characterized by a normality-distribution region score threshold, and
each normality-ratio predictive recommendation mapping rule is characterized by a normality-ratio region score threshold; and
perform one or more prediction-based actions based at least in part on the selected subset.
9. The apparatus of claim 8 , wherein the common target object is a brain region, and the plurality of disjoint image regions comprise a plurality of defined brain regions.
10. The apparatus of claim 8 , wherein the one or more MRI images comprise a T1-weighted MRI image, a T2-weighted MRI image, a T2*-weighted MRI image, and a Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
11. The apparatus of claim 10 , wherein each voxel input representation for a particular image voxel is a four-dimensional vector comprising a first value for the image voxel as determined based at least in part on the T1-weighted MRI image, a second value for the image voxel as determined based at least in part on the T2-weighted MRI image, a third value for the image voxel as determined based at least in part on the T2*-weighted MRI image, and a fourth value for the image voxel as determined based at least in part on the Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
12. The apparatus of claim 8 , wherein the voxel integrity score generation machine learning model is a Gaussian mixture model cluster analysis machine learning model.
13. The apparatus of claim 8 , wherein the one or more normalization variables comprise an age variable.
14. The apparatus of claim 8 , wherein the normalization space comprises, for each historical MRI set of one or more historical MRI sets that are associated with the one or more normalization variables, a group of historical voxel integrity scores for the historical MRI set that are generated based at least in part on output of processing the historical MRI set using one or more of the image preprocessing model, the image segmentation model, and the voxel integrity score generation machine learning model.
15. A computer program product for generating one or more predictive recommendations based at least in part on a magnetic resonance imaging (MRI) acquisition protocol characterized by one or more MRI images that are associated with a common target object having a common target object type, the computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions configured to:
generate, using an image preprocessing model for the common target object type, and based at least in part on the MRI set, a standardized MRI set;
generate, using an image segmentation model, and based at least in part on the standardized MRI set: (i) a plurality of disjoint image regions of the standardized MRI set, and (ii) for each disjoint image region, a subset of a group of image voxels of the standardized MRI set;
for each image voxel:
generate, using a voxel integrity score generation machine learning model, and based at least in part on a voxel input representation for the image voxel, a voxel integrity score for the image voxel, and
generate, using an integrity score normalization model characterized by a normalization space that is defined by one or more normalization variables associated with the common target object, and based at least in part on the voxel integrity score for the image voxel, a non-parametric integrity ranking for the image voxel;
generate a group of region scores based at least in part on each non-parametric integrity ranking, wherein:
each region score is associated with a respective disjoint image region and is generated based at least in part on each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and
the group of region scores comprise, for each disjoint image region: (i) a normality-distribution region score that is generated based at least in part on an average measure of each non-parametric integrity ranking for those image voxels that are in the respective disjoint image region, and (ii) a normality-ratio region score that is generated based at least in part on a ratio of those image voxels that are in the disjoint image region and that are associated with non-threshold-satisfying non-parametric integrity rankings and those image voxels that are in the disjoint image region and that are associated with threshold-satisfying non-parametric integrity rankings;
generate, using a predictive recommendation model and based at least in part on the group of region scores, the one or more predictive recommendations, wherein:
the predictive recommendation model is configured to map the group of region scores to a selected subset of a plurality of candidate predictive recommendations,
the predictive recommendation model is characterized by a plurality of predictive recommendation mapping rules comprising one or more normality-distribution predictive recommendation mapping rules and one or more normality-ratio predictive recommendation mapping rules,
each normality-distribution predictive recommendation mapping rule is characterized by a normality-distribution region score threshold, and
each normality-ratio predictive recommendation mapping rule is characterized by a normality-ratio region score threshold; and
perform one or more prediction-based actions based at least in part on the selected subset.
16. The computer program product of claim 15 , wherein the common target object is a brain region, and the plurality of disjoint image regions comprise a plurality of defined brain regions.
17. The computer program product of claim 15 , wherein the one or more MRI images comprise a T1-weighted MRI image, a T2-weighted MRI image, a T2*-weighted MRI image, and a Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
18. The computer program product of claim 17 , wherein each voxel input representation for a particular image voxel is a four-dimensional vector comprising a first value for the image voxel as determined based at least in part on the T1-weighted MRI image, a second value for the image voxel as determined based at least in part on the T2-weighted MRI image, a third value for the image voxel as determined based at least in part on the T2*-weighted MRI image, and a fourth value for the image voxel as determined based at least in part on the Fluid Attenuated Inversion Recovery (FLAIR) MRI image.
19. The computer program product of claim 15 , wherein the voxel integrity score generation machine learning model is a Gaussian mixture model cluster analysis machine learning model.
20. The computer program product of claim 15 , wherein the one or more normalization variables comprise an age variable.