AI-powered mind reading: Women-only university in Karnataka targets early epilepsy detection

The university, facing severe financial difficulties and once on the verge of closure, recently floated a tender for advanced hardware, including processors and NVIDIA A100 graphics cards, essential for accelerating their LDM-based research in brain image reconstruction.

Professor Ramesh said the university is undertaking this project with all its limited resources.

"The use of AI-based LDMs to map brain activity to high-resolution images offers major advantages in medicine and society. It can aid in the early detection and diagnosis of neurological disorders such as Alzheimer's, Parkinson's, and schizophrenia by identifying changes in brain activity patterns,” said Ramesh.

“It can also contribute to the development of brain-computer interfaces, potentially enabling paralysed people to communicate with the outside world,” he said.

The project is being implemented with funding from the state government's Vision Group on Science and Technology (VGST) under the research excellence scheme.

How does it work?

LDM is a type of generative model that can create highly realistic images from noise. The team proposes to use LDM to interpret the complex patterns of brain activity captured through fMRI.

fMRI is a non-invasive brain imaging technique that measures changes in blood flow in the brain, which correlate with neural activity. By analysing these patterns, the LDM can generate images representing the visual information the brain is processing. Leveraging LDMs can also reduce computational costs without sacrificing performance.

“AI-based LDM extracts and parses data from fMRI and electroencephalograms [EEG], disseminating knowledge that facilitates the reconstruction of fMRI and EEG data and aids in innovating new biomarkers. These new biomarkers can help identify the type of epilepsy in infants, allowing for the correct treatment to be administered at the right time,” said Ramesh.

Biomarkers are indicators that can be used to track brain function and health. “LDM architecture works in a back-and-forth direction, with many weights assigned during the reconstruction of high-quality fMRI and EEG images,” he said.

Ramesh said although the department is not partnering with any hospitals, it will consult with them to collect fMRI and EEG data.

Speaking about the challenges of obtaining fMRI and EEG data, Ramesh said, “AI-based computational models are not rule-based; they are training-based and are inherently a black box. Many challenges must be addressed to achieve accurate results.”

The university’s AI advancements extend beyond groundbreaking research in brain activity. Last year, the it made headlines when its journalism department introduced Asha, an AI news anchor, on its campus news channel, Akka.

AI is making significant strides in medical science, particularly in disease detection. Earlier this year, researchers from Harvard and Google mapped a small section of the human brain, capturing each neuron and the intricate networks they form. This fragment was taken from a 45-year-old woman who underwent brain surgery for epilepsy.

This groundbreaking brain map reveals around 57,000 neurons, 230 millimeters (9 inches) of blood vessels, and 150 million synapses — the connection points between neurons.

Meanwhile, the Bengaluru-based start-up BrainSightAI is harnessing the power of AI and machine learning to create personalised brain maps.

The company's VoxelBox platform uses AI and machine learning-based neuroimaging solutions to create personalised brain maps, providing visual representations of the functional and structural connections within the brain.