Method and system for reflectance imaging of peripheral nerves
Methods and systems useful for machine learning assisted imaging and detection of peripheral nerves comprising reflectance imaging spectroscopy. The method can be conducted label-free and in real-time.
1 . A system for real time imaging a peripheral nerve in a tissue sample in a surgery, the system comprising:
a light source configured to irradiate the tissue sample,
a photodetector configured to detect reflected light at a wavelength of 410-490 nm from the tissue sample, and
a computer configured to:
generate one or more images from the detected reflected light;
use a first deep learning model to classify the one or more images so as to identify one or more nerve-related extracted images, wherein each of the one or more nerve-related extracted images is an image with a presence of nerve;
use a second deep learning model to perform segmentation of the peripheral nerve in each of the one or more nerve-related extracted images for highlighting one or more specific anatomical structures of the peripheral nerve to a surgeon during the surgery;
additionally use the first deep learning model to determine if the second deep learning model for nerve segmentation provides one or more confusing nerve segments; and
if it is determined that the second deep learning model provides the one or more confusing nerve segments, alert the surgeon to conduct a surgical procedure with extra caution to prevent damage to nerve during the surgery.
2 . The system according to claim 1 , wherein the light source comprises coherent light, metal-halide lamp, LED light, mercury lamp, superluminescent diodes, or broadband light sources that provide light across a wide range of wavelength.
3 . The system according to claim 1 , wherein the system further comprises a band pass filter to remove reflected light outside of 410-490 nm wavelength.
4 . The system according to claim 1 , wherein the photodetector is a stereomicroscope.
5 . The system according to claim 1 , wherein the first deep learning model is DenseNet201 and the second deep learning model is DoubleUNet.