High-Speed Ultraviolet Photoacoustic Microscopy for Histological Imaging with Virtual-Staining assisted by Deep Learning

Intraoperative surgical monitor analysis is still a big challenge in tumor resection surgeries because it requires a diagnostic tool to confirm the complex excision of cancerous tissue accurately and quickly. It's then possible to avoid repeated surgeries due to a positive surgical margin. Our high-speed open-top ultraviolet photoacoustic microscopy can provide labeled histological images of tissue samples within 18 minutes with an imaging area of five by five millimeter squares, showing great potential for intraoperative surgical margin analysis.

Demonstrating the procedure will be Xiufeng Li, a final year PhD student from my laboratory. Start with setting the optical illumination for the microscopic system by using a Q-switch diode-pumped solid-state laser with a wavelength of 266 nanometers as an excitation source. Install two convex lenses to expand the laser beam and place a pinhole close to the focal point of the first convex lens for spatial filtering.

Use a 1D galvanometer mirror to reflect the laser beam upward. With the help of the objective lens, focus the laser beam on a sample. Fix the ring shaped focused UT with the active area facing upward into a lab made water tank with an optically transparent window at the bottom covered by a thin quartz coverslip, then attach the water tank to a two-axis manual stage of the microscope to control the lateral position of the UT.When done, turn on the UV laser and adjust the position of the UT to allow the laser beam to pass through the center of the UT, then turn off the UV laser and fill the water tank with deionized water to fully immerse the UT.Connect the output of the UT to two amplifiers to get a total gain of 56 decibels and then connect the output of the second amplifier to a data acquisition card, or DAQ, installed in the computer.

Next, attach a sample holder to a z-axis manual stage connected to xy-motorized stages followed by placing four pieces of double-sided tape surrounding the empty hole of the sample holder. Later, proceed to align the system by attaching black tape to a glass slide and then placing the glass slide to cover the hole of the sample holder with the black tape facing downward. After pressing the glass slide on the sample holder, lower the sample holder to immerse the glass slide into the water.

Disconnect the ring shaped UT and amplifiers, then connect the UT to the pulser/receiver and the output of the pulser/receiver to an oscilloscope. Operate the pulser/receiver into the pulse echo mode with the pulse amplitude of six and the gain at 20 decibels. Once the parameters are set, adjust the z-position of the sample holder, define the position of the acoustic focal plane where the ultrasonic signals are maximum.

Change the pulse/receiver to the transmission mode before setting the gain to 60 decibels, then enable the laser output and adjust the z-position of the objective lens to maximize the photoacoustic or PA signals measured by the oscilloscope. To make the generated PA signal symmetric and maximum, adjust the lateral position of the ring-shaped UT, then adjust the z-position of the sample holder to maximize the PA signals. Repeat the adjustments for the z-position of the objective lens and the sample holder to optimize the symmetry and amplitude of the PA signals.

When the PA signals are optimized, record the time delay or the time taken by the PA waves to reach the UT on the oscilloscope. Move the sample holder to image different positions of the black tape. Adjust the sample holders flatness such that the PA signals generated from each position of the black tape have the same time delay as the one measured earlier.

After the system alignment is complete, turn off the laser and connect the UT to the two amplifiers. To prepare a formalin-fixed and paraffin-embedded mouse brain slice, fixed the harvested brain in 10%neutral buffered formalin. After 24 hours, process the fixed brain by dehydration with graded alcohol, clearing with xylene and embedding with paraffin.

Use a microtome to obtain five micrometer thick slices of the embedded brain. Place the sample slices on the quartz slides to dry in an oven at 60 degrees Celsius for one hour. Later, de-paraffinize the brain sections with a clearing agent to avoid high background signals of paraffin.

To prepare a fresh mouse brain slice, wash the harvested mouse brain with PBS and then cut a five millimeter thick slice of the brain sample by hand followed by washing the brain slice with PBS to remove the blood on the cross-section. For the sample placement, prepare a lab made sample tank with a UV transparent polyethylene membrane with a thickness of 10 micrometers, then add a drop of water to the membrane and place the biological sample on the sample tank to cover the water. Next, place the sample containing tank on the sample holder ensuring to cover the empty hole of the sample holder.

Set the UV laser to the external trigger mode and use the lab built lab view program to set the scanning parameters as described in the manuscript. Start the trial scan on a small region by setting the number of moving steps of the x and y-axis motors, then adjust the z-axis manual stage to place the sample on the focal plane for obtaining maximum PA signals. After moving both x and y-axis motors to the desired starting point and setting the scanning region by setting the xn and yn values, start the image acquisition program.

When the images are required, turn off the laser and remove the sample holder. Store fresh biological tissues in 10%neutral buffered formalin. Use the collected PA signals to reconstruct the maximum amplitude projection image with the help of a lab-built image processing algorithm.

The representative analysis shows the UV-PAM and H&E images of an FFPE mouse brain slice. In the zoomed in UV-PAM image, the individual cell nuclei could be resolved. The corresponding cell nuclei were found in the standard H&E stained images, showing the high accuracy of the current system for cellular imaging.

A deep learning algorithm was applied to transfer the gray scale UV-PAM image to a virtual H&E stained image. It is important to adjust the position of the ring-shaped ultrasonic transducer to allow the UV light to pass through a center during the system alignment. This ensures that a detectable photoacoustic signal can be obtained and further optimized when the laser is on.

Our classification algorithm can be further incorporated to classify tumor and normal regions in the images, serving as an assistive imaging and a diagnostic platform for medical professionals.