Neural network system for non-destructive optical coherence tomography

1. A system for non-destructive optical coherence tomography (OCT), the system comprising:

an input interface for receiving OCT data of an object;

a processing unit executable to detect one or more features on a surface or subsurface of the object via a data science module, the data science module configured to:

using a trained A-scan neural network, analyze each of a plurality of A-scans from the OCT data to detect the presence of an A-scan feature associated with the object;

segment each of a plurality of B-scans from the OCT data to determine thresholds associated with the object;

using a trained B-scan neural network, analyze each segmented B-scan to detect the presence of a B-scan feature associated with the object;

convert a C-scan from the OCT data to one or more two-dimensional representations; and

using a trained C-scan neural network, detect the presence of a C-scan feature associated with the object; and

an output interface for outputting the analyses and detections of the data science module.

2. The system of claim 1 , wherein the OCT data comprises at least one C-scan, and wherein the data science module is further configured to separate the at least one C-scan into the plurality of A-scans.

3. The system of claim 2 , wherein the data science module is further configured to separate the at least one C-scan into the plurality of B-scans.

4. The system of claim 1 , wherein each of the plurality of A-scans are analyzed successively and each of the plurality of B-scans are analyzed successively.

5. The system of claim 1 , wherein the training data for the A-scan neural network, the B-scan neural network, and the C-scan neural network comprises historical data of previous OCT scans and associated features.

6. The system of claim 1 , wherein the training data for the A-scan neural network, the B-scan neural network, and the C-scan neural network is received by initially operating in a training mode where features associated with the OCT data are determined with manual input via the input interface.

7. The system of claim 1 , wherein the A-scan neural network uses a trained Long-Term Short Memory (LSTM) machine learning model, the B-scan neural network uses another trained Long-Term Short Memory (LSTM) machine learning model, and the C-scan neural network uses a Convolutional Neural Network (CNN).

8. The system of claim 7 , wherein the A-scan neural network outputs a score representing a probability for the detection of the presence of the A-scan feature.

9. The system of claim 1 , wherein the segmenting of each B-scan comprises determining thresholds using at least one of Canny edge detection, Otsu clustering-based image thresholding, and Integral Image Thresholding.

10. The system of claim 1 , wherein the converting of the C-scan from the OCT data to one or more two-dimensional representations comprises extracting a surface map from the C-scan and determining a projection of the C-scan.

11. The system of claim 10 , wherein the extracting of the surface map comprises determining regions-of-interest (ROI) from the segmented B-scans, using image segmentation techniques to segment layers of the object into B-Scans, and using the layers to create surface maps by mapping each B-scan to a line that represents a distance from a surface of the B-scan.

12. The system of claim 10 , wherein the determining of the projection comprises mapping each A-scan onto a point that is an average or sum of the intensities along an A-scan axis and combining each of the mapped points in the C-scan to construct a two-dimensional projection map.

13. The system of claim 10 , wherein detecting the presence of the C-scan feature comprises using a Gabor filter with the projection and surface map.

14. The system of claim 1 , the processing unit further executable to determine coordinates, size, or both, of at least one of the A-scan feature, B-scan feature, and C-scan feature via an interpretation module.

15. A computer-implemented method for non-destructive optical coherence tomography (OCT), the method comprising:

receiving OCT data of an object;

detecting one or more features on a surface or subsurface of the object by:

using a trained A-scan neural network, analyzing each of a plurality of A-scans from the OCT data to detect the presence of an A-scan feature associated with the object;

segmenting each of a plurality of B-scans from the OCT data to determine thresholds associated with the object;

using a trained B-scan neural network, analyzing each segmented B-scan to detect the presence of a B-scan feature associated with the object;

converting a C-scan from the OCT data to one or more two-dimensional representations; and

using a trained C-scan neural network, detecting the presence of a C-scan feature associated with the object; and

outputting the analyses and detections.

16. The method of claim 15 , wherein the OCT data comprises at least one C-scan, the method further comprising separating each C-scan into the plurality of B-scans and separating the B-scans into the plurality of A-scans.

17. The method of claim 15 , wherein converting of the C-scan from the OCT data to one or more two-dimensional representations comprises extracting a surface map from the C-scan and determining a projection of the C-scan.

18. The method of claim 17 , wherein extracting of the surface map comprises determining regions-of-interest (ROI) from the segmented B-scans, using image segmentation techniques to segment layers of the object into B-Scans, and using the layers to create surface maps by mapping each B-scan to a line that represents a distance from a surface of the B-scan.

19. The method of claim 17 , wherein determining the projection comprises mapping each A-scan onto a point that is an average or sum of the intensities along an A-scan axis and combining each of the mapped points in the C-scan to construct a two-dimensional projection map.

20. The method of claim 17 , wherein detecting the presence of the C-scan feature comprises using a Gabor filter with the projection and surface map.