Methods and systems for resting state fMRI brain mapping with reduced imaging time
A method for mapping brain function of a subject includes receiving a dataset of resting state fMRI (RS-fMRI) three dimensional (3D) image frames of the subject's brain, and inputting the 3D image frames to a deep learning artificial neural network. For each voxel of each 3D image frame and for each resting state network of a plurality of resting state networks, the deep learning artificial neural network calculates a probability that the voxel belongs to the resting state network. The deep learning artificial neural network is trained beforehand using a plurality of 3D image frames including previously defined resting state networks obtained from a plurality of calibration subjects. The method includes generating one or more functional map of the plurality of resting state networks of the subject's brain using the probabilities calculated by the artificial neural network.
1 . A method for mapping brain function of a subject, the method comprising:
receiving a dataset of time-series resting state fMRI (RS-fMRI) comprising a series of individual three dimensional (3D) image frames of the subject's brain, wherein each frame represents a snapshot in time over the course of the RS-fMRI scan and the dataset comprises less than about 200 individual 3D image frames;
inputting the 3D image frames to a 3D convolutional neural network designed to account for the spatial relationship between voxels by performing three dimensional convolutions, wherein for each voxel of the subject's brain and for each defined resting state network of a plurality of defined resting state networks the 3D convolutional neural network calculates a probability that the voxel belongs to each of the resting state networks, each of the plurality of defined resting state networks is characterized by its location being in a same general region across a population of subjects but where there is significant variability in network boundaries at an individual level, and the 3D convolutional neural network is trained beforehand using a plurality of 3D image frames including previously defined resting state networks obtained from a plurality of calibration subjects; and
generating one or more functional map of the plurality of resting state networks of the subject's brain using the probabilities calculated by the 3D convolutional neural network.
2 . The method of claim 1 , wherein the 3D image frames of the subject's brain are input to the 3D convolutional neural network as tensors.
3 . The method of claim 1 , wherein the dataset comprises less than about 150 3D image frames.
4 . The method of claim 1 , further comprising generating 3D correlation maps correlating each voxel in each 3D image frame to other time correlated voxels in the 3D image frame.
5 . The method of claim 4 , wherein inputting the 3D image frames to the artificial neural network comprises inputting the generated 3D correlation maps to the 3D convolutional neural network.
6 . A method for mapping brain function of a subject, the method comprising:
receiving a dataset of time-series resting state fMRI (RS-fMRI) comprising a series of individual three dimensional (3D) image frames of the subject's brain, wherein each frame represents a snapshot in time over the course of the RS-fMRI scan;
inputting the 3D image frames to a a 3D convolutional neural network designed to account for the spatial relationship between voxels by performing three dimensional convolutions, wherein for the subject's brain and for each defined resting state network of a plurality of defined resting state networks the 3D convolutional neural network calculates a probability that the voxel belongs to each of the resting state networks, each of the plurality of defined resting state networks is characterized by its location being in a same general region across a population of subjects but where there is significant variability in network boundaries at an individual level, and the 3D convolutional neural network is trained beforehand using a plurality of 3D image frames including previously defined resting state networks obtained from a plurality of calibration subjects; and
generating one or more functional map of the plurality of resting state networks of the subject's brain using the probabilities calculated by the 3D convolutional neural network.
7 . The method of claim 6 , wherein the 3D image frames of the subject's brain are input to the 3D convolutional neural network as tensors.
8 . The method of claim 6 , further comprising generating 3D correlation maps correlating each voxel in each 3D image frame to other time correlated voxels in the 3D image frame.
9 . The method of claim 8 , wherein inputting the 3D image frames to the 3D convolutional neural network comprises inputting the generated 3D correlation maps to the 3D convolutional neural network.
10 . A method for mapping of brain function of a subject for a purpose for which a first number of resting state fMRI (RS-fMRI) individual three dimensional (3D) image frames of the subject's brain is typically acquired, the method comprising:
receiving a dataset of time-series RS-fMRI comprising a series of individual 3D image frames of the subject's brain, wherein each frame represents a snapshot in time over the course of the RS-fMRI scan and the dataset comprising less than half of the first number of individual 3D image frames;
inputting the 3D image frames to a 3D convolutional neural network account for the spatial relationship between voxels by performing three dimensional convolutions, wherein for each voxel of the subject's brain and for each defined resting state network of a plurality of defined resting state networks the 3D convolutional neural network calculates a probability that the voxel belongs to each of the resting state networks, each of the plurality of defined resting state networks is characterized by its location being in a same general region across a population of subjects but where there is significant variability in network boundaries at an individual level, and the 3D convolutional neural network is trained beforehand using a plurality of 3D image frames including previously defined resting state networks obtained from a plurality of calibration subjects; and
generating one or more functional map of the plurality of resting state networks of the subject's brain using the probabilities calculated by the 3D convolutional neural network.
11 . The method of claim 10 , wherein the 3D image frames of the subject's brain are input to the 3D convolutional neural network as tensors.
12 . The method of claim 10 , wherein the purpose comprises pre-surgical planning in a patient with brain tumors and epilepsy, and the first number of RS-fMRI 3D image frames is more than 1000.
13 . The method of claim 12 , wherein the dataset comprising less than half of the first number of 3D image frames comprises less than 200 3D image frames.