Method and system for diagnosis of attention deficit hyperactivity disorder from magnetic resonance images

Patent metadata

Field	Value
Publication number	US-9510756-B2
Application number	US-201313785050-A
Country	US
Kind code	B2
Filing date	Mar 5, 2013
Priority date	Mar 5, 2012
Publication date	Dec 6, 2016
Grant date	Dec 6, 2016

Abstract

A method and system for automated diagnosis of attention deficit hyperactivity disorder (ADHD) from magnetic resonance images is disclosed. Anatomical features are extracted from a structural magnetic resonance image (MRI) of a patient. Functional features are extracted from a resting-state functional MRI (rsFMRI) series of the patient. An ADHD diagnosis for the patient is determined based on the anatomical features, the functional features, and phenotypic features of the patient using a trained classifier. An ADHD subtype may then be determined for patients diagnosed as ADHD positive using a second trained classifier.

First claim

The invention claimed is:

1. A method for automated diagnosis of attention deficit hyperactivity disorder (ADHD), comprising:
extracting anatomical features from a structural magnetic resonance image (MRI) of a patient;
extracting functional features from a resting-state functional MRI (rsFMRI) series of the patient, comprising:
extracting an rsFMRI time series for each of a plurality of brain regions by mapping voxels in each of a plurality of image volumes in the rsFMRI series to a plurality of brain regions and extracting an rsFMRI time series for each brain region based on the voxels mapped to that brain region in the plurality of image volumes in the rsFMRI series by calculating, for each of M brain regions, an average of voxels mapped to that brain region in each of N image volumes in the rsFMRI series, resulting in an M ×N matrix including the rsFMRI time series for each of the brain regions, and
extracting the functional features based on the rsFMRI time series for each of the plurality of brain regions; and

determining an ADHD diagnosis for the patient based on the anatomical features, the functional features, and phenotypic features of the patient using a trained machine learning classifier.

2. The method of claim 1 , wherein the structural MRI is a T-1weighted MRI.

3. The method of claim 1 , wherein the functional MRI series is a T2-weighted MRI series.

4. The method of claim 1 , where the phenotypic features of the patient include age and gender of the patient.

5. The method of claim 4 , wherein the phenotypic features of the patient further include handedness, verbal IQ, and performance IQ of the patient.

6. The method of claim 1 , wherein extracting anatomical features from a structural magnetic resonance image (MRI) of a patient comprises:
segmenting cortical hemisphere surfaces in the structural MRI; and
extracting at least one anatomical feature at each of a plurality of uniformly spaced vertices on each of the cortical hemisphere surfaces.

7. The method of claim 6 , wherein extracting at least one anatomical feature at each of a plurality of uniformly spaced vertices on each of the cortical hemisphere surfaces comprises:
extracting at least one of cortical thickness or mean curvature at each of the plurality of uniformly spaced vertices on each of the cortical hemisphere surfaces.

8. The method of claim 6 , wherein extracting anatomical features from a structural magnetic resonance image (MRI) of a patient further comprises:
grouping the plurality of uniformly spaced vertices on each of the cortical hemisphere surfaces into a plurality of cortical parcels; and
calculating a surface area of each of the plurality of cortical parcels.

9. The method of claim 6 , wherein extracting anatomical features from a structural magnetic resonance image (MRI) of a patient further comprises:
segmenting subcortical brain structures in the structural MRI;
calculating a volume of each subcortical brain structure; and
normalizing the calculated volume of each subcortical brain structure by an intracranial volume of the patient.

10. The method of claim 9 , wherein extracting anatomical features from a structural magnetic resonance image (MRI) of a patient further comprises:
calculating a volume of a subcortical area with hypointensities in gray matter and a volume of a subcortical cortical area with hypointensities in white matter based on the segmented subcortical brain structures; and
normalizing the volume of the subcortical area with hypointensities in gray matter and the volume of the subcortical cortical area with hypointensities in white matter by the intracranial volume of the patient.

11. The method of claim 1 , wherein extracting the functional features based on the rsFMRI time series for each of the plurality of brain regions comprises:
generating a network graph representing connectivity between the plurality of brain regions based on the rsFMRI time series for each of the plurality of brain regions; and
extracting network features from the network graph.

12. The method of claim 11 , wherein extracting an rsFMRI time series for each of a plurality of brain regions comprises:
calculating a first transformation to warp the structural MRI into a template space of a brain atlas image defining the plurality of brain regions;
aligning each of the plurality of image volumes in the rsFMRI series with a first one of the plurality of image volumes in the rsFMRI series;
calculating a second transformation to co-register the aligned plurality of image volumes in the rsFMRI series with the structural MRI;
transforming the plurality of image volumes in the rsFMRI series to the template space using the first transformation;
mapping the voxels in each of the plurality of image volumes in the rsFMRI series to the plurality of brain regions defined by the brain atlas image; and
extracting the rsFMRI time series for each brain region based on the voxels mapped to that brain region in the plurality of image volumes in the rsFMRI series.

13. The method of claim 12 , wherein extracting an rsFMRI time series for each of a plurality of brain regions further comprises:
performing temporal interpolation on each of the plurality of image volumes in the rsFMRI series prior to aligning each of the plurality of image volumes in the rsFMRI series.

14. The method of claim 12 , wherein extracting an rsFMRI time series for each of a plurality of brain regions further comprises:
discarding non-brain matter voxels from the plurality of image volumes in the rsFMRI series prior to mapping the voxels in each of the plurality of image volumes in the rsFMRI series to the plurality of brain regions.

15. The method of claim 12 , wherein extracting an rsFMRI time series for each of a plurality of brain regions further comprises:
performing linear regression over the rsFMRI series for each voxel to remove effects correlated with a mean time course of a measured signal calculated in voxels corresponding to white matter and cerebrospinal fluid prior to mapping the voxels in each of the plurality of image volumes in the rsFMRI series to the plurality of brain regions.

16. The method of claim 12 , wherein extracting an rsFMRI time series for each of a plurality of brain regions further comprises:
isolating portions of the rsFMRI series that are within a predetermined frequency range using a bandpass filter; and
smoothing the rsFMRI series resulting from the bandpass filter with a 3D Gaussian smoothing kernel prior to mapping the voxels in each of the plurality of image volumes in the rsFMRI series to the plurality of brain regions.

17. The method of claim 12 , wherein extracting an rsFMRI times series for each brain region based on the voxels mapped to that brain region in the plurality of image volumes in the rsFMRI series comprises:
removing voxels in the plurality of brain regions determined to be located in white matter;
subdividing each brain region into sub-regions by detecting groups of voxels in the brain region with similar time series and assigning the groups of voxels to the same sub-region; and
extracting an rsFMRI time series for each sub-region of each brain region based on the voxels assigned to each sub-region in the plurality of the image volumes in the rsFMRI series.

18. The method of claim 11 , wherein generating a network graph representing connectivity between the plurality of brain regions based on the rsFMRI time series for each of the plurality of brain regions comprises:
calculating an affinity matrix representing connectiv

Assignees

• Grady Leo
• Saperstein Sara
• Bohland Jason
• Siemens Healthcare Gmbh
• Univ Boston

Inventors

• Grady Leo
• Saperstein Sara
• Bohland Jason

Classifications

• G01R33/5602

  by filtering or weighting based on different relaxation times within the sample, e.g. T1 weighting using an inversion pulse · CPC title

• G06T7/0012

  Biomedical image inspection · CPC title

• A61B2576/026

  for the brain · CPC title

• A61B5/4076

  Diagnosing or monitoring particular conditions of the nervous system · CPC title

• A61B5/168

  Evaluating attention deficit, hyperactivity · CPC title