The UWIN seminar series continue in February 2020 with a pair of short talks by Ariel Rokem and Samira Moorjani. The seminar is on Wednesday, February 12, 2020 at 3:30pm in Husky Union Building (HUB) 337. Refreshments will be served prior to the talks.
“Automated detection of glaucoma in multi-modal retinal images with interpretable deep learning”
Ariel Rokem, Senior Data Scientist, eScience Institute, University of Washington
“‘Movement-triggered electrical stimulation for strengthening cortical connections”
Samira Moorjani, Research Assistant Professor, Department of Physiology & Biophysics, University of Washington
“Automated detection of glaucoma in multi-modal retinal images with interpretable deep learning” (Ariel Rokem)
Glaucoma is a leading cause of irreversible blindness worldwide, affecting tens of millions of people. Automated systems to diagnose glaucoma could decrease the incidence of undiagnosed cases of glaucoma and facilitate earlier initiation of therapeutic mitigation strategies that can be effective in slowing the progression of the disease. We trained a machine learning model that can accurately detect the presence of glaucoma, using data from the UK Biobank study. Our model combines information from demographic, systemic medical data, and ocular data with deep learning (DL) models trained to analyze both color fundus photos (CFP) and retinal optical coherence tomography scans (OCT). Our model accurately detected glaucoma in a test set of held-out subjects with an area under the receiver operating characteristics curve (AUC) of 0.97. Using new methods for machine learning model interpretation provides additional insight into the features of the images and other variables used in determination of the predicted diagnosis.
“Movement-triggered electrical stimulation for strengthening cortical connections” (Samira Moorjani)
Activity-dependent electrical stimulation, inspired by spike-timing-dependent plasticity of neural circuits, has been employed in vivo in rats and monkeys for modulating synaptic connectivity. In my talk, I will describe a closed-loop stimulation paradigm in which electrical stimuli are triggered by volitional movements. When paired with successive use of the electrically-conditioned motor pathways, movement-triggered stimulation (MTS) led to lasting strengthening of cortical connections in behaving monkeys. Importantly, neither physical activity alone nor electrical conditioning alone produced similar effects. These experiments suggest that MTS creates a plastic landscape in which repetition of the conditioned movement drives cortical strengthening long after the stimulation has ended. Taken together, our data alludes to a crucial role of behavior in neural plasticity and provides support for combining MTS with physical therapy for strengthening motor pathways weakened by injury or disease.