Optical imaging employs light as a medium to non-invasively look into the cellular and molecular functions inside a body. It has significant in vivo applications like assisting in surgeries or examining the brains. A group of researchers at Purdue University is now bringing the technology a step further, to shed lights on how certain diseases will affect brain’s activities and help surgeons to better map out tumors found within the body.
Optimizing optical imaging
The Purdue researchers proposed a diffusion model based optimization approach to enhance localization of contrast between fluorescent agents introduced into the body in search for tumors and/ or blood vessels within tissues and light that’s being absorbed. They believe this will overcome one of the major challenges in fluorescence imaging, which is light becomes overtly diffused and minimizes the amount of information surgeons receive to design and carry out surgical procedures to remove tumors.
The technology not only helps surgeons to determine position and depth of tumors, which are not easily attainable with present technologies, but also provides more detailed information on the neuron activities within the brain. This may also help doctors in detecting diseases like Parkinson’s. The findings are now published in journal IEEE Transactions on Medical Imaging.
Apart from looking at fluorescent inhomogeneities, some researchers are also relying on artificial intelligence (AI) to smarten up optical imaging. Researchers from University of Michigan, Columbia University, New York University, University of Miami, and Invenio Imaging, built a deep convolutional neural network using over 2.5 million samples from 415 patients to categorize tissues into 13 histologic categories most commonly found in brain tumors. The network was then validated using data from another 278 patients.
AI helps to see the invisible
In order to generate real-time and accurate intraoperative diagnosis of brain tumors, researchers leverage on stimulated Raman histology (SRH) imaging techniques before the microscopic images are processed and analyzed by the AI system. Results indicated that diagnosis involving AI had achieved 94.6% accuracy level as compared to 93.9% attained by human pathologist interpretations.
According to the research team, surgeons are often restricted to act on what they can see but AI has enabled them to see even the “invisible” or what normally they failed to notice. This will not only quicken and perfect a diagnosis but also minimize the possibility of a misdiagnosis. They believe AI will make cancer surgeries safer and more effective in the long run as AI becomes able to predict key molecular changes in brain tumors. Related findings were published in Nature Medicine earlier.