Hi All,
We’re looking for motivated folks to collaborate with.
We’re interested in using machine learning to quantitatively identify tissues/materials within the body for medical imaging (x-rays).
We have a new method that uses pre-image response signals. It can be thought of as being analogous to mass spectrometry, but using x-rays, instead of ambiguous grayscale images. We achieve this by creating vectors across energy bins that are similar to hands on a clock, but instead of telling the time, we can tell what tissues the photons have passed through.
We then place materials in a 3D cube, and based on their position, we can infer the atomic number of materials/tissues in an x-ray image. We obtain this with a simple k-nearest-neighbor (KNN) algorithm. Please see below as an example of our work. The red dot represents breast tissue, which was left out of the training dataset, where said method accurately identified the tissue’s atomic number (~6.8).
Importantly, our process provides explainable, transparent, and reproducible results because of physics-embedded geometry - critical for patient trust and FDA clearance. Our physics-embedded vectors contain photon-tissue interaction information, much like the hands on a clock inadvertently contain Newton’s laws of gravity and Kepler’s angular momentum as the earth traverses around the sun.
Feel free to check out our website: kairossensors.com
Shoot me an email if you think you can help: Kendon.Shirley@KairosSensors.com
