Real-time photoacoustic microscopy for rapid diagnosis of pathology

1 . A photoacoustic microscopy system for intraoperative pathology imaging, the system comprising:

a pulsed laser source that generates an ultraviolet laser beam;

a polygon scanner having a plurality of reflective facets, wherein the polygon scanner deflects the ultraviolet laser beam along a fast axis;

a telephoto lens assembly that receives the deflected ultraviolet laser beam from the polygon scanner, wherein the telephoto lens assembly comprising a positive lens group and a negative lens group that expand the deflected ultraviolet laser beam;

an objective lens configured to focus the expanded ultraviolet laser beam onto a tissue sample;

a motorized stage having at least two axes of motion, including at least one axis of an imaging plane and an axis orthogonal to the imaging plane, wherein the motorized stage is configured to support the tissue sample and configured to translate the tissue sample along the at least one axis of the imaging plane and vertically along the orthogonal axis;

an ultrasound transducer that is configured to detect photoacoustic signals generated from the tissue sample in response to illumination by the ultraviolet laser beam;

a data acquisition device that digitizes the photoacoustic signals; and

a controller that:

controls the motorized stage and is configured to perform a rapid pre-scan of the tissue sample;

extracts surface topology data from the photoacoustic signals acquired during the pre-scan to generate a contour map representing variations in surface height of the tissue sample;

synchronizes operation of the pulsed laser source, the polygon scanner, the motorized stage, and the data acquisition device during imaging; and

controls the motorized stage and is configured to dynamically adjust a position of the tissue sample vertically along the orthogonal axis during a high-resolution imaging scan based on the contour map.

2 . The system of claim 1 , comprising a nonlinear crystal positioned in an optical path of a laser beam at a first wavelength emitted by the pulsed laser source, wherein the nonlinear crystal converts the laser beam of the first wavelength to generate the ultraviolet laser beam through harmonic generation.

3 . The system of claim 1 , comprising:

an optical fiber having a stripped tip positioned at a beginning of a scanning trajectory of the ultraviolet laser beam; and

a photodiode optically coupled to the optical fiber, wherein the photodiode generates a start-of-scan timing signal when the deflected ultraviolet laser beam strikes the stripped tip of the optical fiber.

4 . The system of claim 1 , wherein:

the ultrasound transducer is to detect the photoacoustic signals in a transmission mode of signal detection;

in a first configuration, the ultrasound transducer comprises a cylindrically focused transducer providing a one-dimensional (1D) line focus for signal detection, wherein the ultraviolet laser beam is confocal with the 1D line focus and is scanned along the 1D line focus; and

in a second configuration, the ultrasound transducer is arranged in a linear ultrasound transducer array such that transducer elements detect signals from an illuminated spot on the tissue sample.

5 . The system of claim 1 , wherein the ultrasound transducer is to detect photoacoustic signals in a reflection mode of signal detection, and wherein the ultrasound transducer comprises a cylindrically focused transducer element having a central rectangular slit aperture that permits coaxial transmission of the ultraviolet laser beam therethrough.

6 . The system of claim 1 , comprising:

a right-angle prism and a rhomboid prism separated by an optical coupling medium, wherein the ultrasound transducer is to detect photoacoustic signals in a reflection mode of signal detection,

wherein the rhomboid prism includes an acoustic cylindrical lens ground into a surface thereof,

wherein the right-angle prism includes an optical correction lens ground on a top surface thereof,

wherein the ultrasound transducer is mounted on the rhomboid prism and acoustically coupled thereto, and

wherein the right-angle prism and the rhomboid prism guide the ultraviolet laser beam and acoustic waves to a common focal region.

7 . The system of claim 1 , comprising a processor that executes a conditional diffusion model comprising a neural network, wherein the processor transforms photoacoustic microscopy images into virtual histologically-stained images.

8 . The system of claim 7 , wherein the processor executes a semantic segmentation network to identify tumor regions and non-tumor regions in the virtual histologically-stained images.

9 . The system of claim 1 , comprising a processor that:

analyzes photoacoustic microscopy images acquired from the tissue sample using ultraviolet laser excitation;

applies a conditional diffusion model to the photoacoustic microscopy images to generate virtual histologically-stained images, wherein the conditional diffusion model comprises a denoising neural network trained to reverse a forward diffusion process while being conditioned on the photoacoustic microscopy images; and

outputs the virtual histologically-stained images for pathology assessment.

10 . The system of claim 9 , wherein the processor preprocess the photoacoustic microscopy images prior to applying the conditional diffusion model by performing at least one of background flattening, global intensity normalization, filter-based denoising, or multi-channel enhancement.

11 . The system of claim 9 , wherein the conditional diffusion model generates the virtual histologically-stained images at a higher resolution than the photoacoustic microscopy images through learned upsampling.

12 . The system of claim 9 ,

wherein the processor applies a semantic segmentation network to the virtual histologically-stained images to generate a segmentation map identifying tumor regions and non-tumor regions.

13 . The system of claim 12 , wherein the semantic segmentation network receives as input both the virtual histologically-stained images and the photoacoustic microscopy images.

14 . The system of claim 12 , wherein the semantic segmentation network is trained using a loss function comprising at least one of Dice loss, cross-entropy loss, or boundary-aware regularization terms.

15 . A method for rapid intraoperative pathology imaging, the method comprising:

generating an ultraviolet laser beam;

deflecting the ultraviolet laser beam using a rotating polygon scanner along a fast axis;

directing the deflected ultraviolet laser beam through a telephoto lens assembly comprising a positive lens group and a negative lens group to expand the deflected ultraviolet laser beam;

focusing the expanded ultraviolet laser beam onto a surface of a tissue sample using an objective lens;

performing a pre-scan of the tissue sample by:

raster scanning the ultraviolet laser beam across the tissue sample with a reduced resolution relative to a subsequent high-resolution imaging scan,

detecting photoacoustic signals generated during the pre-scan, and

extracting surface topology information from time-of-flight data contained in the photoacoustic signals to generate a contour map;

performing a high-resolution imaging scan by:

raster scanning the ultraviolet laser beam across the tissue sample,

dynamically adjusting a position of the tissue sample along an orthogonal axis in real-time based on the contour map to maintain the surface within an optical depth of focus, and

detecting photoacoustic signals to generate pathology images.

16 . The method of claim 15 , wherein generating the ultraviolet laser beam comprises:

emitting a pulsed laser beam at a first wavelength; and

converting the first wavelength to an ultraviolet wavelength through second harmonic generation using a nonlinear crystal.

17 . The method of claim 15 , comprising:

detecting a start-of-scan signal using an optical fiber positioned at a beginning of a scanning trajectory; and

synchronizing laser pulsing, data acquisition, and stage movement based on the start-of-scan signal.

18 . The method of claim 15 , wherein detecting photoacoustic signals comprises detecting the photoacoustic signals in transmission mode using a cylindrically focused ultrasound transducer positioned opposite to an illumination side of the tissue sample.

19 . The method of claim 15 , wherein detecting photoacoustic signals comprises detecting the photoacoustic signals in reflection mode through a cylindrically focused ultrasound transducer with a slit aperture, wherein the ultraviolet laser beam passes coaxially through the slit aperture.

20 . The method of claim 15 , comprising applying a conditional diffusion model to the pathology images to generate virtual histologically-stained images that simulate conventional histological staining without physical staining of the tissue sample.

21 . The method of claim 20 , comprising applying a semantic segmentation network to the virtual histologically-stained images to automatically detect tumor margins within the tissue sample.

22 . The method of claim 15 , comprising:

training a conditional diffusion model for virtual staining using training data comprising:

photoacoustic microscopy images acquired with ultraviolet laser excitation; and

corresponding histologically-stained images,

wherein the conditional diffusion model is trained using a loss function comprising at least one of reconstruction loss, perceptual loss, adversarial loss, structural preservation loss, or segmentation consistency loss; and

training a semantic segmentation network for tumor margin detection using annotated datasets comprising labeled tumor regions and non-tumor regions.