Monitoring dynamics of patient brain state during neurosurgical procedures

1. A computer-implemented method of monitoring dynamics of patient brain state during neurosurgical procedures, the method comprising:

receiving a plurality of brain response images acquired using a functional MRI (fMRI) scan of a patient over a plurality of time points;

selecting a plurality of image features from the brain response images;

using one or more recurrent neural network (RNN) models to directly estimate one or more brain state measurements at each time point based on the image features; and

presenting the one or more brain state measurements on a display.

2. The method of claim 1 , wherein the plurality of image features are selected by applying one or more 3D convolutional networks to the plurality of brain response images.

3. The method of claim 1 , wherein the plurality of image features are selected from the brain response images using an informativeness-based voxel selection method.

4. The method of claim 3 , wherein the informativeness-based voxel selection method is applied to a subset of the voxels in each brain response image remaining after applying a cortical mask to each brain response image.

5. The method of claim 4 , wherein the subset of the voxels in each brain response image comprises approximately 10% voxels in each brain response image.

6. The method of claim 4 , wherein the informativeness-based voxel selection method determines an informativeness score for each voxel in the subset of the voxels by applying one or more ridge regression models to predict one or more dimensions of the brain state measurements from each voxel and its adjacent neighbors in 3D.

7. The method of claim 1 , wherein the one or more RNN models comprise (a) a first layer long short-term memory (LSTM) model receiving the plurality of image features and generating internal output data and (b) a second layer LSTM model receiving the internal output data and estimating the one or more brain state measurements.

8. The method of claim 7 , wherein the brain state for a given time point is used as an input to the second layer LSTM model for subsequent time points.

9. A computer-implemented method of monitoring dynamics of patient brain state during neurosurgical procedures, the method comprising:

receiving a plurality of brain response images collected using a functional MRI scan of a patient over a plurality of time points;

selecting a plurality of image features from the brain response images;

estimating one or more brain state measurements at each time point by (a) applying a decoding model that convolves the image features with a canonical hemodynamic response function (HRF) to yield combined features and (b) estimating the brain state measurements as a linear combination of the combined features; and

presenting the one or more brain state measurements on a display.

10. The method of claim 9 , wherein the plurality of image features are selected by applying one or more 3D convolutional networks to the plurality of brain response images.

11. The method of claim 9 , wherein the plurality of image features are selected from the brain response images using an informativeness-based voxel selection method.

12. The method of claim 11 , wherein the informativeness-based voxel selection method is applied to a subset of the voxels in each brain response image remaining after applying a cortical mask to each brain response image.

13. The method of claim 12 , wherein the subset of the voxels in each brain response image comprises approximately 10% voxels in each brain response image.

14. The method of claim 12 , wherein the informativeness-based voxel selection method determines an informativeness score for each voxel in the subset of the voxels by applying one or more ridge regression models to predict one or more dimensions of the brain state measurements from each voxel and its adjacent neighbors in 3D.

15. The method of claim 9 , wherein the brain state measurements are estimated using one or more RNN models comprising (a) a first layer LSTM model receiving the plurality of image features and generating output data and (b) a second layer LSTM model receiving the output data and estimating the one or more brain state measurements.

16. The method of claim 15 , wherein the brain state for a given time point is used as an input to the second layer LSTM model for subsequent time points.

17. A system for monitoring dynamics of patient brain state during neurosurgical procedures, the system comprising:

a functional MRI scanner configured to acquire a plurality of brain response images over a plurality of time points;

one or more processors configured to (a) select a plurality of image features from the brain response images and (b) apply one or more machine learning models to estimate one or more brain state measurements at each time point based on the image features;

a display configured to present the one or more brain state measurements.

18. The system of claim 17 , wherein the one or more brain state measurements are estimated by:

using one or more long short-term memory (LSTM) models to directly estimate one or more brain state measurements at each time point based on the image features.

19. The system of claim 17 , wherein the one or more brain state measurements are estimated by:

estimating one or more brain state measurements at each time point by (a) applying a decoding model that convolves the image features with a canonical hemodynamic response function (HRF) to yield combined features and (b) estimating the brain state measurements as a linear combination of the combined features.