The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here: https://www.microsoft.com/en-us/microsoft-365/windows/end-of-ie-support).

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

Spectral Unmixing

A key feature of photoacoustic imaging is its ability to illuminate tissue at multiple wavelengths, and thus record images with a spectral dimension. Spectral imaging allows sensing of intrinsic chromophores that can reveal physiological, cellular and subcellular functions. However, the identification of spectral signatures within images obtained at multiple wavelengths requires spectral unmixing techniques. In the image below we schematically demonstrate how a multispectral photoacoustic image is constructed and how spectra can be obtained from different regions of the image.

Spectral Unmixing - link to project.

Once a spectrum is extracted, spectral unmixing breaks that spectrum into a linear combination of so-called endmember spectra. Which endmember spectra should be included requires that some assumptions of what may be absorbing are made. One may, for example, suspect that in tissue, there should be some quantity of oxygenated and deoxygenated hemoglobin, melanin, water and fat. Once spectral unmixing is performed, the linear coefficients of each endmember provide some information on the amount of each endmember that is present in the volume represented by the extracted spectrum.

Several spectral unmixing techniques are available, such as adaptive matched filtering, independent component analysis, and principal component analysis. Preliminary results have shown that a modified variant of adaptive matched filtering shows promise. We will evaluate this further using a phantom, ex vivo, and in vivo measurements. 

The image below is taken from a recent publication of ours where the lateral dimensions of a melanoma skin tumor was determined with the aid of spectral unmixing. The dimensions agree quite well compared to the RGB image acquired of the same tumor (left). But the more important clinical implication is that we were able to determine the depth at each location to which the tumor extends (right).

Lateral dimensions of a melanoma skin tumor was determined with the aid of spectral unmixing. Photo.

Spectral unmixing is also useful to assess oxygen saturation of tissue at different depths. PA imaging can provide a detailed enough spectrum in every pixel, representing about 100 x 100 um2 of the tissue, down to a depth of a few cm. In a recent study, we managed to monitor the oxygen saturation in the top three skin layers of the fingertip while inducing ischemia via occlusion of the finger at its base. The image below demonstrates how the spectral unmixing operates on a measured PA spectrum as it breaks down the measured PA spectrum (black traces) into the best linear combination of absorption spectra representing melanin (yellow), oxygenated (red) and deoxygenated (blue) hemoglobin. By calculating the relative contribution of oxygenated Hb to the total Hb, oxygen saturation (sO2) can be acquired. During occlusion, it becomes clear that contribution of oxygenated Hb goes down, while also that of deoxygenated Hb goes up, albeit slightly. Once reperfusion occurs, oxygenated Hb rises and deoxygenated Hb falls. Noteworthy is that in all three stages, melanin content stays the same, which is consistent with what we would expect physiologically. 

Tissue composition after spectral unmixing. Graph.

Project participants

Azin Khodaverdi. Photo.
Azin Khodaverdi, MSc, PhD student
Magnus Cinthio, MSc, PhD
Associate professor Magnus Cinthio, MSc, PhD
Tobias Erlöv, MSc, PhD. Photo.
Tobias Erlöv, MSc, PhD
Aboma Mendasa, MSc, PhD. Photo.
Aboma Merdasa, MSc, PhD
Rafi Sheikh, MD, PhD. Photo.
Associate professor Rafi Sheikh, MD, PhD
Nina Reistad, MSc, PhD
Nina Reistad, MSc, PhD
Professor Malin Malmsjö, MD, PhD. Photo.
Professor Malin Malmsjö, MD, PhD