VISTA Noise Reduction

Abstract

An optical coherence tomography angiography (OCT-A) method, the method including: generating at least two OCT-A images based on different interscan times; generating a short interscan time (SIT) representative image and a long interscan time (LIT) representative image based on the at least two denoised OCT-A images; estimating relative blood flow velocities based on the SIT representative image and the LIT representative image; and generating a blood flow image based on the relative blood flow velocities.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 63 / 269,307, entitled "VISTA Noise Removal," filed March 14, 2022, which is incorporated by reference in its entirety. [Background technology]

[0002] Optical coherence tomography (OCT) is a non-invasive imaging technique that is often used in ophthalmology. OCT utilizes the principles of interferometry to image and collect information about an object (such as a subject's eye). Specifically, light from a light source is split into a sample arm, where it is reflected by the object being imaged, and a reference arm, where it is reflected by a reference object, such as a mirror. The reflected light is then combined into a detection arm, producing an interference pattern that is detected, for example, by a spectrometer or photodiode. The detected interference signal is processed to reconstruct the object and generate an OCT image.

[0003] More specifically, structural OCT images and volumes are generated by combining multiple depth profiles (A-lines, e.g., along the Z-depth direction at XY locations) into a single cross-sectional image (B-scans, e.g., in the XZ or YZ planes) and by combining multiple B-scans into one volume. These depth profiles are generated by scanning along the X and Y directions. En face images in the XY plane can be generated by flattening the volume in all or part of the Z-depth direction, and C-scan images may be generated by extracting slices of the volume at a given depth. Angiographic (OCT-A) images may be generated by comparing information from structural images and / or volumes at different times (e.g., from repeated scans). Assuming that the structures of interest remain the same for a relatively short time (on the order of milliseconds to seconds) between scans, the changes are due to blood flow, which may identify the vasculature.

[0004] As with many imaging techniques, OCT and OCT-A imaging can generally suffer from unwanted signal noise. For example, OCT and OCT-A images are prone to noise and artifacts caused by variations in flow velocity, signal quality, and patient motion. The presence of such noise makes it difficult to detect vascular structures and blood flow velocities, especially in small capillaries. Several approaches exist for removing / mitigating noise, but these approaches often require more scans, thereby increasing total scan time and processing requirements. Summary of the Invention [Means for solving the problem]

[0005] According to one example of the present disclosure, a method includes generating at least three structural optical coherence tomography (OCT) images of the same location on a subject; generating at least two OCT angiography (OCT-A) images based on the structural OCT images, where the at least two OCT-A images have different interscan times between the structural OCT images used to generate the OCT-A images; removing noise from the at least two OCT-A images; generating a representative image with a short interscan time (SIT) and a representative image with a long interscan time (LIT) based on the at least two OCT-A images; and estimating relative blood flow velocity based on the SIT representative image and the LIT representative image.

[0006] In various embodiments of the above examples, the at least two OCT-A images are cross-sectional B-scans; the method further comprises generating at least two OCT-A images for a plurality of positions of the object, thereby forming a plurality of OCT-A volumes, and denoising an en face image of each of the plurality of OCT-A volumes after denoising the at least two OCT-A images, the SIT representative image and the LIT representative image being based on the denoised en face images; generating a blood flow image based on an estimated relative blood flow velocity; the blood flow image being a colour mapping image, the pixel colours of which correspond to the estimated relative blood flow velocity; the denoising being based on at least one training the generating of the SIT representative image comprises statistically combining denoised OCT-A images having interscan times shorter than a predetermined threshold, and the generating of the LIT representative image comprises statistically combining denoised OCT-A images having interscan times longer than a predetermined threshold; the estimated relative blood flow velocity at the given location is a ratio of the SIT representative image at the given location to the LIT representative image at the given location; estimating the relative blood flow velocity is a pixel-wise determination of the ratio of the SIT representative image to the LIT representative image; the ratio is raised to a power of 1.5 or greater; and / or the subject is a retina.

[0007] According to another example of the present disclosure, a method includes generating a plurality of optical coherence tomography angiography (OCT-A) volumes, where the plurality of OCT-A volumes have different interscan times between structural OCT images used to generate each OCT-A volume; denoising the plurality of OCT-A volumes, where the denoising comprises: denoising B-scan images of the plurality of OCT-A volumes; and denoising en face images of the plurality of OCT-A volumes after denoising the B-scan images; generating a representative image with a short interscan time (SIT) by statistically combining the denoised en face images of the OCT-A volumes having an interscan time shorter than a predetermined threshold; and generating a representative image with a long interscan time (LIT) by statistically combining the denoised en face images of the OCT-A volumes having an interscan time longer than a predetermined threshold.

[0008] In various embodiments of the above examples, the method further comprises estimating a relative blood flow velocity based on the SIT representative image and the LIT representative image, and generating a blood flow image based on the estimated relative blood flow velocity; the noise removal is performed by at least one trained machine learning system; the method further comprises estimating the relative blood flow velocity as a pixel-wise determination of the ratio of the SIT representative image to the LIT representative image at a given location raised to a power of 1.5 or greater, and generating a blood flow image based on the estimated relative blood flow velocity; and / or the subject is a retina. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 shows a schematic diagram of an example optical coherence tomography system of the present disclosure. [Diagram 2] FIG. 2 illustrates an example method of the present disclosure. [Diagram 3] FIG. 3 illustrates an example noise reduction method of the present disclosure. [Figure 4] FIG. 4 illustrates an example noise reduction method of the present disclosure. [Diagram 5] FIG. 5 shows an example image showing blood flow velocity. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] In view of the above, the present disclosure relates to techniques for reducing OCT-A noise while reducing total scan time. More specifically, the present disclosure utilizes machine learning systems to denoise OCT-A images and estimate relative blood flow velocities associated with:

[0011] Using the methods described below, OCT-A images can be more reliably suppressed without the use of filters. Due to the better noise suppression, such OCT-A images show the vasculature in greater detail and can be used to identify specific locations of slow and fast blood flow. Using the described ratios to generate color mapping images, a wider dynamic range can be obtained and estimated relative blood flow velocities can be determined.

[0012] The present disclosure may utilize an OCT system 101 as shown in FIG. 1. As described above, the system 101 includes a light source 100. Light generated by the light source 100 is split, for example, by a beam splitter (as part of the interferometer optics 108) and sent to a reference arm 104 and a sample arm 106. The light in the sample arm 106 is backscattered or otherwise reflected from an object, such as the retina of an eye 112. The light in the reference arm 104 is backscattered or otherwise reflected by a mirror 110 or similar object. The light from the sample arm 106 and the reference arm 104 are recombined in the optics 108, and a corresponding interference signal is detected by a detector 102. The detector 102 may be a spectrometer, a photodetector, or any other light detection device. The detector 102 outputs an electrical signal corresponding to the interference signal to a processor 114, where it may be stored and processed into OCT signal data and / or OCT-A data. The processor 114 may then further generate a corresponding image or otherwise perform analysis of the data. The processor 114 may also be associated with an input / output interface (not shown), including a display for outputting the processed images or information related to the analysis of those images. The input / output interface may include hardware such as buttons, keys, or other controls for receiving user input to the system. In some embodiments, the processor 114 may also be used to control the light source and the imaging process.

[0013] 2, an exemplary method of the present disclosure first acquires two or more repeated B-scans at the same location of the object 201. These B-scans may be acquired, for example, using the OCT system 101 shown in FIG. 1 by capturing multiple A-lines and generating structural OCT images and volumes therefrom using the processor 114. Specifically, these structural OCT images (e.g., B-scans) are generated by combining multiple depth profiles (A-lines) into a cross-sectional image (B-scan), and the volume is obtained from the multiple combined B-scans. Similarly, the processor 114 may generate en face images by flattening the volume in the depth direction, or C-scans by extracting slices of the volume at a given depth.

[0014] The repeated B-scans may be acquired according to any scanning protocol. For example, the entire OCT volume may be acquired before acquiring repeated data from any location within the volume. In other embodiments, individual A-lines or B-scans may be repeated before proceeding to the next A-line or B-scan. In this manner, multiple B-scans and / or volumes are effectively acquired simultaneously.

[0015] The processor 114 uses at least two repeated B-scans for each location to generate an OCT-A image 202 for that location. While this method is possible with two repeated B-scans, the OCT system 101 can generate many more B-scans. An OCT-A image is generated for each pair of images at a given location, regardless of the number of B-scans. These comparisons may be based on any or all possible combinations of B-scans. By way of example, the Inter-Scan Time-Variable Analysis (VISTA) method may be used to generate the OCT-A images. By way of example, these OCT-A images may be generated as described in U.S. Pat. No. 10,839,515, entitled "System and Method for Generating and Displaying OCT Angiography Data Using Inter-Scan Time-Variable Analysis," which is incorporated herein by reference in its entirety. As with the structural OCT images described above, multiple OCT-A images from multiple cross-sectional locations may be combined to form an OCT-A volume.

[0016] More specifically, VISTA involves generating OCT-A images corresponding to different interscan times and then interpreting the differences in these images / data as being related to blood flow velocity, speed, or related quantities. The speed of the OCT-A system can also affect the acquisition of blood flow; for example, if the OCT-A system has a fast A-scan rate (e.g., 400 kHz), it may be difficult to capture slow blood flow.

[0017] For example, considering four repeated B1-B4 scans at times t1-t4, OCT-A images may be generated for pairs B1-B2, B1-B3, B1-B4, B2-B3, B2-B4, and B3-B4. Thus, for four repeated B-scans, six OCT-A images may be generated. The time between repeated OCT B-scans is referred to herein as the "inter-scan time." In the above example, assuming a constant time Δt between OCT B-scans, the OCT-A images may have Δt (e.g., t2-t1) for the OCT-A images based on OCT B-scan pairs B1-B2, B2-B3, and B3-B4, 2Δt (e.g., t3-t1) for the OCT-A images based on OCT B-scan pairs B1-B3 and B2-B4, and 3Δt (e.g., t4-t1) for the OCT-A images based on OCT B-scan pairs B1-B3 and B2-B4, and 3Δt (e.g., t4-t1) for the OCT-A images based on OCT B-scan pairs B1-B4.

[0018] The interscan time determines the sensitivity and saturation of the OCT-A signal (and image) to blood flow velocity. In other words, a longer interscan time is more sensitive to slow flow velocities, but produces a saturated OCT-A signal at higher velocities. A shorter interscan time can detect these faster flows, but usually reduces the OCT-A signal and may not detect slower blood flow velocities. The relationship between OCT-A signal and blood flow velocity is approximately linear. For example, doubling the interscan time results in approximately the same change in OCT-A signal as doubling the blood flow velocity. This relationship can be used to estimate blood flow velocity and / or related quantities.

[0019] Following OCT-A image generation 202, the processor 114 may denoise the OCT-A image 203. Denoising may be achieved by various machine learning techniques, such as spatial filtering, time accumulation or deep learning reconstruction. The process of denoising may comprise one or more levels of denoising. In some embodiments, noise reduction is achieved by applying deep learning based noise reduction techniques such as those described in U.S. Pat. No. 11,257,190, entitled "Image Quality Improvement Method for Optical Coherence Tomography," which is incorporated herein by reference in its entirety. Additionally, shadow and projection artifacts may be reduced by applying image processing and / or deep learning techniques such as those described in U.S. Pat. No. 11,361,481, entitled "3D Shadow Reduction Signal Processing Method for Optical Coherence Tomography (OCT) Images," which is incorporated herein by reference in its entirety.

[0020] For example, as shown in Figure 3, the OCT-A volume may be denoised at the B-scan level. In other words, an OCT-A cross-sectional B-scan 303 is input into a B-scan noise reduction machine learning system 302, which is trained to output a denoised B-scan 304. The B-scan noise reduction machine learning system 302 may be trained using a variety of machine learning training techniques, such as supervised, semi-supervised, unsupervised or enhanced. The training data 301 for the machine learning system may include pairs of B-scan OCT-A images taken at the same position, where one image contains noise and the other image is denoised.

[0021] In another embodiment, paired cross-sectional OCT-A B-scans from a common location are both not denoised and are input as training data 301. In this way, the machine learning system learns to recognise random noise between pairs of OCT-A training images. This recognised random noise can be removed from the other input OCT-A B-scan to output a denoised OCT-A B-scan. In other words, the machine learning system 302 can be trained to recognise noise in OCT-A images by providing the machine learning system 302 with training data 301 comprising pairs of OCT-A images representing the same location. Differences between the OCT-A images can simply be considered as noise, since any structural differences between the OCT-A images are already taken into account by the OCT-A processing.

[0022] Similarly, as shown in FIG. 4, noise reduction can occur at the frontal level. Frontal noise reduction can be achieved in a similar manner as described above for B-scans. Depending on the embodiment, frontal images from multiple depths (or depth ranges) can be denoised for a single OCT-A volume. For example, frontal images 403 from an OCT-A volume can be input to a frontal noise reduction machine learning system 402, which outputs a denoised frontal image 404. The machine learning system 402 can be trained to recognise and remove noise from the input frontal image 403. The frontal AI noise reduction system can be trained using various machine learning training techniques, such as supervised, semi-supervised, unsupervised or reinforcement. Similar to the B-scan noise reduction system, the training data 401 can include pairs of frontal images taken at the same position, where one image contains noise and the other image is denoised. In other embodiments, the training data 401 can include pairs of frontal images with random noise, where the difference between each image is random noise.

[0023] The noise reduction process can be implemented in a variety of ways. For example, in one embodiment, the noise reduction process 203 can first denoise the B-scans of the OCT-A volume (e.g., using the B-scan noise reduction machine learning system 302) and then perform frontal level noise reduction (e.g., using the frontal noise reduction machine learning system 402). In other words, the denoised B-scans can be combined with the frontal image to perform frontal level denoising. In other embodiments, the noise reduction process 203 first denoises the OCT-A volume at the frontal level before denoising at the B-scan level. In yet other embodiments, the OCT-A volume may be denoised separately at the B-scan level and the frontal level. In these cases, the resulting B-scan level denoised volume and the frontal level denoised volume can be recombined in any statistical manner to generate a fully denoised OCT-A volume. In still other embodiments, only one of the B-scan level and frontal level denoising can be performed on the OCT-A volume to generate a denoised OCT-A image and / or volume. Using the above process and the example of four scans repeated per position, the result of denoising will be six denoised OCT-A en face images, B-scans and / or volumes.

[0024] Returning to FIG. 2, the denoised OCT-A volumes (and images therefrom) described above are averaged or statistically combined 204. In some embodiments, the processor 114 can determine an average (or similar statistical combination) for OCT-A images or volumes having short interscan times (SIT) (e.g., Δt) and long interscan times (LIT) (e.g., longer than Δt) 204. In these cases, Δt represents a predefined threshold that separates "short" interscan times from "long" interscan times. For example, in the above example with six denoised OCT-A en face images, three denoised OCT-A en face images (from OCT B-scan pairs B1-B2, B2-B3, and B3-B4) have SIT Δt, and three denoised OCT-A en face images (from OCT B-scan pairs B1-B3, B2-B4, and B1-B4) have LITs longer than Δt. Thus, the processor 114 can determine an average of the three denoised OCT-A en face images of the SIT to generate a single en face image that is representative of the SIT. The processor 114 can also determine an average of the three denoised OCT-A en face images of the LIT to generate a single en face image that is representative of the LIT.

[0025] The relationship between the OCT-A image with SIT and the OCT-A image with LIT can be utilized to determine the relative blood flow velocity. In one embodiment, the processor 114 can determine the relative blood flow velocity 205 based on the ratio between the SIT and LIT images. For example, a single en face blood flow image can be generated by taking a pixel-wise ratio of the SIT representative en face image to the LIT representative en face image. This resulting en face blood flow image can be analyzed and processed by the processor 114 to determine blood flow velocity and similar related quantities. For example, the individual pixel values ​​(ratio values) of the en face blood flow image can correspond to the relative blood flow velocity. These en face blood flow images correspond to the depth at which the denoised OCT-A en face images were taken (and thus represented by the SIT and LIT representative images). The en face blood flow images can be at one or more depths, since the OCT-A en face images can be denoised at another depth. For example, en face blood flow images can be generated at superficial depths, deeper depths (e.g., in the choroid), and choriocapillaries.

[0026] A longer interscan time is more sensitive to slow flow velocities and a shorter interscan time is more sensitive to faster flow velocities, so a smaller ratio value (smaller SIT numerator but larger LIT denominator) indicates a slower estimated blood flow and a larger ratio value (larger SIT numerator but smaller LIT denominator) indicates a faster blood flow. Of course, the inverse of the ratio (LIT denominator, SIT numerator) can also be utilized. In yet another embodiment, the dynamic range of the estimated relative blood flow velocity determined from the en face blood flow image can be improved by using a ratio of a power greater than 1. For example, the ratio can be a power of 1.5. Increasing the dynamic range of the estimated relative blood flow velocity allows a wider range of values ​​and more detailed estimation.

[0027] As discussed above, each pixel in the en face perfusion image can correspond to a relative blood flow velocity, which may be a ratio of the SIT representative image to the LIT representative image, or other statistical combinations of the SIT representative image and the LIT representative image (e.g., ratios taken to the power of 1.5). The processor 114 can further generate other types of images from the en face perfusion image, e.g., B-scans, volumes, etc.

[0028] These generated en face blood flow images can be color mapped, for example, depicting relative blood flow velocities with different colors (e.g., blue for slower blood flow and red for faster blood flow). In other words, the relative blood flow velocities (e.g., ratio values) are mapped to the hues of pixels in the en face blood flow images. Such images showing blood flow velocities can be useful in alerting clinicians to vessel types and in identifying diseases such as vessel bulges, narrowings, and even leaks in the vasculature.

[0029] In some embodiments, for example where the en face blood flow image is grayscale, the relative blood flow values ​​(eg, ratio values) can be expressed as pixel intensities.

[0030] FIG. 5 shows an example of an en face blood flow image 500 in grayscale. Grayscale images can represent relative blood flow as light intensity ranging from black for weakest intensity (and minimum flow velocity) to white for strongest (and maximum flow velocity). For example, location 502 in en face blood flow image 500 is the foveal avascular zone and is therefore black and associated with a lack of relative blood flow. In contrast, location 504 includes vasculature depicted by higher intensity information associated with faster relative blood flow, while location 506 in image 500 shows vasculature with moderate intensity associated with slower relative blood flow. As previously mentioned, these images are useful for identifying vasculature diseases and problems. For example, location 508 in image 500 shows vasculature leakage. The color-mapped images can be displayed on a display or stored in a computer-readable medium such as a random access memory (RAM) or hard drive.

[0031] While there are various features described above, it should be understood that the features may be used alone or in any combination thereof. It should also be understood that variations and modifications may occur to those skilled in the art to which the claimed examples pertain.

Claims

1. generating at least three structural optical coherence tomography images (structural OCT images) of the same location on the subject; generating at least two OCT angiography images (OCT-A images) based on the structural OCT images, the at least two OCT-A images having different inter-scan times between corresponding structural OCT images used to generate the OCT-A images; denoising the at least two OCT-A images; generating a representative image with a short interscan time (SIT representative image) and a representative image with a long interscan time (LIT representative image) based on the at least two OCT-A images; estimating a relative blood flow velocity based on the SIT representative image and the LIT representative image; A method comprising:

2. The method of claim 1 , wherein the at least two OCT-A images are cross-sectional B-scans.

3. The method comprises: generating the at least two OCT-A images for a plurality of locations of the subject, thereby forming a plurality of OCT-A volumes; and and denoising an en face image of each of the plurality of OCT-A volumes after denoising the at least two OCT-A images; The method of claim 2 , wherein the SIT representative image and the LIT representative image are based on the denoised frontal image.

4. The method of claim 1 , further comprising generating a blood flow image based on the estimated relative blood flow velocity.

5. The method of claim 4 , wherein the blood flow image is a color mapping image in which pixel colors correspond to the estimated relative blood flow velocities.

6. The method of claim 1 , wherein the noise removal is performed by at least one trained machine learning system.

7. generating the SIT representative image includes statistically combining denoised OCT-A images having inter-scan times shorter than a predetermined threshold; The method of claim 1 , wherein generating the LIT representative image comprises statistically combining denoised OCT-A images having inter-scan times longer than the predetermined threshold.

8. The method of claim 1 , wherein the estimated relative blood flow velocity at a given location is a ratio of a SIT representative image at the given location to a LIT representative image at the given location.

9. The method of claim 8 , wherein estimating the relative blood velocity is a pixel-wise determination of the ratio of the SIT representative image to the LIT representative image.

10. The method of claim 8 , wherein the ratio is raised to a power of 1.5 or greater.

11. The method of claim 1 , wherein the object is a retina.

12. generating a plurality of optical coherence tomography angiography (OCT-A) volumes, wherein the inter-scan times between the structural OCT images used to generate each OCT-A volume of the plurality of OCT-A volumes are different; denoising the plurality of OCT-A volumes by denoising B-scan images of the plurality of OCT-A volumes, and after denoising the B-scan images, denoising frontal images of the plurality of OCT-A volumes; generating a representative image with a short interscan time (SIT representative image) by statistically combining denoised en face images of the OCT-A volume having an interscan time shorter than a predetermined threshold; and generating a long interscan time representative image (LIT representative image) by statistically combining denoised en face images of OCT-A volumes having interscan times longer than the predetermined threshold; A method comprising:

13. The method comprises: Estimating relative blood flow velocity based on the SIT representative image and the LIT representative image; and generating a blood flow image based on the estimated relative blood flow velocity; The method of claim 12 further comprising:

14. The method of claim 12 , wherein the noise removal is performed by at least one trained machine learning system.

15. The method comprises: estimating relative blood flow velocity as a pixel-wise determination of the ratio of the SIT representative image at a given location to the LIT representative image at the given location, raised to a power of 1.5 or greater; and generating a blood flow image based on the estimated relative blood flow velocity; The method of claim 12 further comprising:

16. The method described in claim 12, wherein the subject of the image is the retina.

Images