System and method for direct learning from raw tomographic data

1. A method for computing classification of raw tomographic data, the method comprising:

supplying the raw tomographic data to a sinogram-convolutional neural network, the sinogram-convolutional neural network comprising a plurality of blocks connected to compute a plurality of features, at least one of the blocks being configured to perform a convolution of the raw tomographic data in Radon space with a convolutional kernel in Radon space by:

slicing the raw tomographic data into a plurality of one-dimensional tomographic data slices along an angle dimension of the raw tomographic data;

slicing the convolutional kernel into a plurality of one-dimensional kernel slices along the angle dimension of the convolutional kernel;

for each angle of the angle dimension, computing a one-dimensional convolution between:

a corresponding one of the one-dimensional tomographic data slices at the angle; and

a corresponding one of the one-dimensional kernel slices at the angle; and

collecting the one-dimensional convolutions at each of the angles to compute the convolution of the raw tomographic data in Radon space with the convolutional kernel in Radon space;

extracting a plurality of features from the raw tomographic data in Radon space using the sinogram-convolutional network; and

computing the classifications of the raw tomographic data based on the features.

2. The method of claim 1 , wherein at least one of the blocks is configured to compute an average-pooling, wherein the convolutional kernel is a pooling kernel.

3. The method of claim 2 , wherein the at least one of the blocks is configured to pool data along a range dimension of the convolution.

4. The method of claim 1 , wherein a connection between two of the blocks comprises a non-linear activation function.

5. The method of claim 4 , wherein the non-linear activation function is a rectified linear unit.

6. The method of claim 5 , wherein the non-linear activation function is implemented as a Taylor expansion around zero.

7. The method of claim 1 , wherein the sinogram-convolutional neural network is configured to compute the features without performing image reconstruction of the raw tomographic data.

8. The method of claim 1 , wherein the sinogram-convolutional neural network is trained based on training data comprising synthetic aperture radar data.

9. The method of claim 1 , wherein the sinogram-convolutional neural network is trained based on training data comprising biological computed tomography data.

10. The method of claim 1 , wherein the sinogram-convolutional neural network is trained based on training data comprising tomography data captured of a three-dimensional printed object.

11. A system for computing classifications of raw tomographic data, comprising:

a sinogram-convolutional neural network comprising a plurality of blocks connected to compute a plurality of features, at least one of the blocks being configured to perform a convolution of the raw tomographic data in Radon space with a convolutional kernel in Radon space by:

slicing the raw tomographic data into a plurality of one-dimensional tomographic data slices along an angle dimension of the raw tomographic data;

slicing the convolutional kernel into a plurality of one-dimensional kernel slices along the angle dimension of the convolutional kernel;

for each angle of the angle dimension, computing a one-dimensional convolution between:

a corresponding one of the one-dimensional tomographic data slices at the angle; and

a corresponding one of the one-dimensional kernel slices at the angle; and

collecting the one-dimensional convolutions at each of the angles to compute the convolution of the raw tomographic data in Radon space and the convolutional kernel in Radon space,

the system being configured to:

extract a plurality of features from the raw tomographic data in Radon space using the sinogram-convolutional neural network; and

compute the classifications of the raw tomographic data based on the features.

12. The system of claim 11 , wherein at least one of the blocks is configured to compute an average-pooling, wherein the convolutional kernel is a pooling kernel.

13. The system of claim 12 , wherein the at least one of the blocks is configured to pool data along a range dimension of the convolution.

14. The system of claim 11 , wherein a connection between two of the blocks comprises a non-linear activation function.

15. The system of claim 14 , wherein the non-linear activation function is a rectified linear unit.

16. The system of claim 15 , wherein the non-linear activation function is implemented as a Taylor expansion around zero.

17. The system of claim 11 , wherein the sinogram-convolutional neural network is configured to compute the features without performing image reconstruction of the raw tomographic data.

18. The system of claim 11 , wherein the sinogram-convolutional neural network is trained based on training data comprising synthetic aperture radar data.

19. The system of claim 11 , wherein the sinogram-convolutional neural network is trained based on training data comprising biological computed tomography data.

20. The system of claim 11 , wherein the sinogram-convolutional neural network is trained based on training data comprising tomography data captured of a three-dimensional printed object.