Subventions et des contributions :

Subvention ou bourse octroyée s'appliquant à plus d'un exercice financier. (2017-2018 à 2022-2023)

Nerve excitability testing is an innovative diagnostic tool that measures the electrical activity of muscles and nerves (i.e. electrodiagnosis) to make decisions about the relative health of the tissue. This particular electrodiagnostic test can provide unique insight into disease pathophysiology of nerves, but it has not been widely adopted. The largest barrier to using the test is the effective interpretation of the results. There is a need for an automated analysis system to assist clinicians with interpretation and differential diagnosis. Over the next five years I plan to work toward solving this issue.

One of the most promising approaches used to classify high dimensional data from a single individual is "machine learning". For example, a machine learning approach can be used to produce a classifier that would use all the attributes of the data from an individual, e.g. Mr. Jones, and determine the probability that Mr. Jones had healthy or unhealthy nerves, and subsequently, the likely pathology. First, a classifier is learned using standard techniques in reinforcement learning. However, this requires a very large dataset that includes the heterogeneity of the population of interest.

Our first task is to complete the acquisition of nerve health data from the arm and leg of neurologically healthy individuals. To date we have collected data from about 50 individuals and we require another 150 individuals to begin the theoretical analysis. Our goal is to develop an international database with contributions from our European, Australian and Japanese colleagues. The database will be set up with our initial deposit of 200 healthy controls and will be built with our collaborators at Atmist.com .

In tandem with our data collection we will use our current mathematical models of nerves to create a virtual database of all the Mr. and Mrs. Jones' and their assorted nerve health issues. This is an essential component to improving the performance of learned classifiers to detect specific pathology from electrodiagnostic data. We can construct a model that generates data consistent with someone with a neurodegenerative condition such as ALS. The machine learning classifier will be trained and then tested to detect and classify data consistent with a diagnosis of ALS, chemotherapy-induced nerve damage, spinal cord injury and other neurologic disorders.

The long-term goal is to provide a system in which clinicians and researchers around the world could upload the results of an electrodiagnotic test and be provided with a categorization and rank-ordered differential diagnosis using the machine learning algorithms developed over the next five years.