Advanced Spectroscopy Laboratory

Retinal Laser Lesion Segmentation

by Eddie Gil

Healthcare providers face a data deluge. This makes quick and efficient decision making challenging. Having automated assist systems can alleviate this issue and save lives. Computer aided diagnostics describes techniques  to assist medical professionals not replace them. One such example is in retinal lesion segmentation.

High power lasers have become easily accessible. Retinal Laser damage may occur under a variety of conditions. However, identifying retinal lesions is challenging and typically requires a panel of ophthalmologists. We trained a fully convolutional network to perform automated segmentation of retinal laser lesions in fundus images. We trained a similar network to perform semantic segmentation of the images as well to display whether damage was, photothermal, photomechanical, or photo chemical.

Fundus images input to the segmentation network alongside their groudtruths and the network output.

Preprocessed image alongside semantic segmentaiton result from network. Color in the segmentation image maps to photothermal, photomechanical, and photochemical damage.

Brillouin Scattering

by Sean O’Connor and Dominik Doktor

Understanding physiological mechanisms at the organelle level is crucial to the development of novel treatments. Viscoelastic properties of materials are of particular interest. Impulsive Stimulated Brillouin Scattering (ISBS) is an emerging nonlinear spectroscopy technique which can be used to probe these viscoelastic properties.  The advantage of this technique is that it requires less acquisition time than Spontaneous Brillouin Scattering ( SpBS).

Our focus is to improve the spatial resolution of ISBS systems to cellular and organelle level imaging. This can later be used to reveal insights about mitochondria, cytoskeletons, etc.

 

Schematic for laser system

Schematic for optical phenomena. Ballmann (2017). Doctoral Dissertation

Compressed Hyperspectral Raman

by Mark Keppler

Hyperspectral microscopy captures spatial information about a scene in a series of images, where each image covers a different wavelength over a large portion of the electromagnetic spectrum.

Key Points:

• Scanning can be replaced with a digital micromirror device (DMD).
• DMD spatially encodes a scene.
• Reduces acquisition time to a single-shot.
• Hyperspectral image stacks are reconstructed by compressed sensing algorithms.

Schematic of the hyperspectral raman imaging system.

Barium Sulfate Spectrum acquired from 1 line in the hyperspectral raman image.

resulting 2D reconstruction of acquired image.