# US 2021/0239618 A1 · High-speed delay scanning and deep learning techniques for spectroscopic srs imaging

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0239618 A1
- Document ID: US20210239618A1
- Status: Published application
- Kind: A1
- Application number: 17/162,455
- Filed: 2021-01-29
- Published: 2021-08-05
- Earliest claimed: 2020-01-30
- Assignees: Trustees of Boston University
- Applicants: Trustees of Boston University
- Inventors: Ji-Xin CHENG; Haonan LIN
- Web page: https://patentgrep.com/patent/US20210239618A1

## CPC

- G01N 21/65: Raman scattering
- A61B 5/0062: {Arrangements for scanning}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0495: Quantised networks; Sparse networks; Compressed networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06T 5/00: Image enhancement or restoration
- G06T 5/60: using machine learning, e.g. neural networks
- G06T 5/70: Denoising; Smoothing
- A61B 5/0075: {by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy (A61B5/0071 takes precedence)}
- G01N 2021/655: {Stimulated Raman}
- G06T 2207/10056: Microscopic image
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30016: Brain
- G01N: INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES (measuring or testing apparatus or processes other than immunoassay, involving enzymes or microorganisms C12M, C12Q)
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

Systems and methods implement of high-speed delay scanning for spectroscopic SRS imaging characterized by scanning a first pulsed beam across a stepwise reflective surface (such as a stepwise mirror or a reflective blazed grating) in a Littrow configuration to generate near continuous temporal delays relative to a second pulsed beam. Systems and methods also implement deep learning techniques for image restoration of spectroscopic SRS images using a trained encoder-decoder convolution neural network (CNN) which in some embodiments may be designed as a spatial-spectral residual net (SS-ResNet) characterized by two parallel filters including a first convolution filter on the spatial domain and a second convolution filter on the spectral domain.

## Claims

### Claim 1

A high-speed delay scanning assembly for imaging modalities that utilize varying temporal delays between pulsed interrogation beams, the assembly comprising a fast linear scanner and a stepwise reflective surface, wherein the scanner is configured to repeatedly scan a first pulsed interrogation beam along a scan line across the stepwise reflective surface in a Littrow configuration thereby changing a path distance of the first beam and introducing a sequence of varying temporal delays relative to a second pulsed interrogation beam.

### Claim 2 (depends on 1)

The assembly of claim 1, wherein the scanner is a polygon scanner.

### Claim 3 (depends on 1)

The assembly of claim 1, wherein the second pulsed interrogation beam has a constant beam path.

### Claim 4 (depends on 1)

The assembly of claim 1, wherein the assembly is for spectroscopic SRS imaging wherein the first pulsed interrogation beam is one of (i) a Stokes beam or (ii) a pump beam.

### Claim 5 (depends on 1)

The assembly of claim 1, wherein the assembly is for transient absorption spectroscopy wherein the first pulsed interrogation beam is one of (i) a pump beam or (ii) a probe beam.

### Claim 6 (depends on 1)

The assembly of claim 1, wherein the assembly is for Impulsive Stimulated Raman Scattering (ISRS) wherein the first pulsed interrogation beam is one of (i) a Stokes beam or (ii) a pump beam.

### Claim 7 (depends on 1)

The assembly of claim 1, wherein the stepwise reflective surface retroflects the first pulsed interrogation beam back along its original path, whereby the first interrogation beam is combined with the second pulsed interrogation beam and linearly chirped by a high dispersion medium to temporally separate different frequency components prior to sample interrogation.

### Claim 8 (depends on 7)

The assembly of claim 7, wherein the dispersion medium includes high dispersion glass rods.

### Claim 9 (depends on 1)

The assembly of claim 1, wherein the first and second pulsed interrogation beams are generated by (i) femtosecond lasers, (ii) supercontinuum lasers, or (iii) broadband lasers.

### Claim 10 (depends on 1)

The assembly of claim 1, wherein a maximum delay or delay range is tunable by rotating the scan line relative to the stepwise reflective surface thereby changing an angle between the scan line and a contour line of the stepwise reflective surface while still maintaining the Littrow configuration.

### Claim 11 (depends on 1)

The assembly of claim 1, wherein the maximum delay or delay range is selected to match pulse chirping of the first pulsed interrogation beam.

### Claim 12 (depends on 1)

The assembly of claim 1, wherein the assembly is configured to provide for high spectral linearity.

### Claim 13

A method for image restoration of spectroscopic images with spectral and spatial domains the method comprising:
Providing an encoder-decoder convolution neural network (CNN) characterized by an architecture that includes convolutions involving both spectral and spatial domains;
training the CNN with a pairs of raw and ground truth spectroscopic images which include both spectral and spatial information, wherein training determines parameters for the CNN that minimize a loss function between a CNN predictions and corresponding ground truth; and
applying the trained CNN to improve signal to noise in raw images.

### Claim 14 (depends on 13)

The method of claim 13, wherein the CNN is a 3D U-net network that includes a 3×3×3 convolution filter.

### Claim 15 (depends on 13)

The method of claim 13, wherein the CNN is a spatial-spectral residual net (SS-ResNet) that includes two parallel filters including a first convolution filter on the spatial domain and a second convolution filter on the spectral domain.

### Claim 16 (depends on 15)

The method of claim 15, wherein the CNN includes a parallel convolution sublayer which includes a first convolution filter (with a (1,3,3) kernel size) for the spatial domain and a second convolution filter (with a (3,1,1,) kernel size) for the spectral domain.

### Claim 17 (depends on 13)

The method of claim 13, wherein image restoration further includes applying pixel-wise LASSO unmixing to suppress the crosstalk between different chemical maps, wherein unmixing is based upon a principle that at each spatial location, only a few chemical components have dominant contributions.

### Claim 18 (depends on 13)

The method of claim 13, wherein images are spectroscopic SRS images.

### Claim 19 (depends on 18)

The method of claim 18, wherein the images are in the fingerprint domain.

### Claim 20

A fast spectroscopic SRS imaging system, the system comprising:
a high-speed delay scanning assembly including a fast linear scanner and a reflective blazed grating, wherein the scanner is configured to repeatedly scan a first pulsed interrogation beam along a scan line across the blazed grating in a Littrow configuration thereby continuously changing a path distance of the first beam and introducing a substantially continuous sequence of varying temporal delays relative to a second pulsed interrogation beam; and
an image processor configured to restore raw images based on application of a trained encoder-decoder convolution neural network (CNN) characterized by an architecture that includes convolutions involving both spectral and spatial domains.
