Lesion Localization | Translational Neuroimaging

Artificial Intelligence in Epilepsy Imaging

Why has Artificial Intelligence impacted many aspects of our everyday life yet is almost entirely absent from the routine clinical assessment of people with epilepsy? We explore the combination of medical and computer science research and, specifically, how we can develop Artificial Intelligence that translates into a gain for patients’ health.

State of the Art FCD Detection: A critical Review

Can we combine imaging and

non-imaging information?

Most of the currently available models solely use MRI data to

detect lesions. But presurgical evaluation is a much more complex

process where many non-imaging modalities play an essential role.

How can AI models use all the available information from imaging

and non-imaging modalities?

In a recent publication, we review current approaches for the automated detection of Focal Cortical Dysplasia (FCD) and highlight present challenges and future research directions to translate such research into clinical practice.

Find it under: https://doi.org/10.1111/epi.17522

Quantifying human performance in lesion localization

Developing clinically impactful AI requires understanding the human capabilities to find lesions in the brain and determine their extent. We have found that FCD lesions are, by normal standards, incredibly hard to find!

Diagnostic Benefit of Computational Models

In the foreseeable future, algorithms will be first and foremost used as decision support systems, so developing AI that finds many lesions is only half the picture. Such models could contribute in different ways to the diagnostic process, so what is their benefit for a clinician tasked with finding a lesion?

What is our ground truth?

Artificial Intelligence models thrive on large and well-annotated datasets. But data generated in clinical routine is often sparse, noisy, and incomplete. Can AI still learn anything meaningful and novel in such an environment?