[cid:image001.jpg@01D87F0B.A0C39650] BME Summer Seminar Series Wednesday, June 15th, 2022 9:30-10:30AM EST Via Zoom Meeting - link below* Evaluation links: Neal Patel: https://purdue.ca1.qualtrics.com/jfe/form/SV_0q66STHRzd5tO9E Andrew Sivaprakasam: https://purdue.ca1.qualtrics.com/jfe/form/SV_7NvEfyH8UY881yC Denoising and Super-Resolving 4D Flow MRI Using Physics-Guided Neural Networks Neal Patel (V. Rayz, advisor) [https://engineering.purdue.edu/ResourceDB/ResourceFiles/image243503/alter?bo...] Abstract: Time-resolved 3D phase-contrast MRI (4D Flow MRI) enables in vivo measurements of blood flow and cerebrospinal fluid (CSF) flow in large vessels and channels. While these measurements can be obtained without assumptions about the flow geometry and boundary conditions, unlike patient-specific computational fluid dynamics (CFD), the resulting velocity data is affected by limited spatiotemporal resolution and noise. As a result, the noisy and under-resolved 4D Flow MRI data affects the accuracy of estimated hemodynamic flow metrics, especially those relying on near-wall velocity gradients such as wall shear stress (WSS). These metrics have been associated with vascular pathologies such as atherogenesis and cerebral aneurysm formation and progression, as well as CSF pathologies such as normal pressure hydrocephalus. Neural networks have been proposed to denoise and super-resolve 4D Flow MRI data, but standard neural nets do not ensure physical consistency of reconstructed high-resolution flow fields. Here, we investigate the use of physics guided neural networks (PGNN) for denoising and super-resolving 4D Flow MRI data in blood flow within cerebral aneurysms and CSF flow within the third and fourth cerebral ventricles. We assess our PGNN using two datasets of high- and low-resolution paired data. High resolution data was obtained from CFD in cerebral aneurysm and ventricle geometries. These high-resolution data were converted to a complex MR signal and downsampled in space and time. Gaussian noise was added, and the resulting synthetic MR signal was then converted back into a low-resolution noisy velocity field. The high-resolution low-resolution pairs were split into training, validation, and test datasets. We show that PGNN outperform traditional neural networks in reconstructing denoised high-resolution velocity fields when tested on both synthetic and in vitro data. Further, we show that by including a divergence-based regularization term, we can better resolve near wall velocities. Place and Time Processing of Pitch in the Impaired Cochlea Andrew Sivaprakasam (M. Heinz/H. Bharadwaj, advisor) [https://engineering.purdue.edu/ResourceDB/ResourceFiles/image218947/alter?bo...] Abstract: Sensorineural hearing loss occurs in 15% of American adults and current treatment protocols are often guided by limited and archaic diagnostics. Sensorineural hearing loss is an umbrella diagnosis that is the result of several underlying deficits difficult to quantify non-invasively. Understanding how specific patterns of damage to the cochlea or auditory nerve variably impair the perception of different sound features is critical to improve treatments for hearing-impaired individuals. The history of auditory research has led to considerable insight as to how anatomic components of the auditory periphery, namely inner hair cells (IHCs), outer hair cells (OHCs), and the cochlear synapse function together to transduce, amplify, and code simple sounds. However, there exists considerable gaps in our knowledge of how these peripheral components maintain the fidelity of more complex auditory phenomena and perception. Pitch, the perceived "highness" or "lowness" of a given sound, is a complex psychoacoustic phenomenon. Pitch cues are used to listen to and compose music and to process vowels, identify talkers, and convey emotion. Without intact pitch perception, conversation becomes emotionless, a symphony becomes a cacophony. After centuries of study, our knowledge of the underlying neurophysiology of pitch still remains mostly hypothetical, with leading theories describing pitch perception in terms of place (tonotopicity) and time (nerve firing patterns) coding by the auditory system. My preliminary results suggest carboplatin-induced IHC damage may result in substantial alteration of the transduction of sound in the cochlea to neural firing of the auditory nerve, quantified non-invasively through envelope-following responses to pitch stimuli. Noise-induced cochlear synaptopathy seems to result in milder deficits. Neither pathology implies a distorted representation of cochlear place. Planned further study includes OHC-related hearing-loss pathologies in parallel with behavioral and EEG experiments in human subjects to quantitatively investigate the functional implications of specific profiles of hearing loss on pitch perception. *Join Zoom Meeting https://purdue-edu.zoom.us/j/92168739019?pwd=cnBOaFBxN3FWVVVTWC8wWWhaekdIdz0... Meeting ID: 921 6873 9019 Passcode: biomedical Liz Rowen She/Her Graduate Program Assistant Weldon School of Biomedical Engineering Martin C. Jischke Hall of Biomedical Engineering 206 S. Martin Jischke Drive West Lafayette, IN 47907-2032 o: 765-494-1197 [7054E290]<https://www.purdue.edu/?utm_source=signature&utm_medium=email&utm_campaign=purdue> -- Bmeroundtable-list mailing list Bmeroundtable-list@ecn.purdue.edu https://engineering.purdue.edu/ECN/mailman/listinfo/bmeroundtable-list