Lab Home | Phone | Search
	Center for Nonlinear Studies
	Home  ▶ People   Current   Executive Committee   Postdocs   Visitors   Students   Research   Publications  ▶ Conferences   Workshops   Sponsorship  ▶ Talks   Seminars   Postdoc Seminars Archive   Quantum Lunch   Quantum Lunch Archive   P/T Colloquia   Archive   Ulam Scholar     Postdoc Nominations  ▶ Student Requests   Student Program  ▶ Visitor Requests   Description   Past Visitors  ▶ Services   General     History of CNLS     Maps, Directions   CNLS Office   T-Division   LANL				Monday, January 22, 20243:00 PM - 4:00 PMCNLS Conference Room (TA-3, Bldg 1690) Colloquium Machine learning applied to cardiovascular and photoacoustic microscopy data Georgios TsironisSchool of Engineering and Applied Sciences, Harvard University and Department of Physics, University of Crete, Greece Machine learning (ML) is a growing field poised to change the way we practice cardiovascular medicine, providing new tools for interpreting data and making decisions. Cardiac digital images or biosignals defy traditional statistical methods and require the deployment of ML. In the first part of this talk we show that ML classifiers trained using anthropometric features and ECG-derived features, can: (a) detect abnormal left ventricular geometry, even before the onset of left ventricular hypertrophy; (b) detect hypertension using 12-lead-ECG-derived features; and (c) detect hypertension in populations without cardiovascular disease from single-lead-ECGs. The latter classifiers can be useful in raising awareness in people with undiagnosed hypertension. We then present a computational method to simulate the dynamics of action potential propagation using the three-variable Fenton-Karma model and account for both normal and damaged cells through spatially inhomogeneous voltage diffusion coefficient. In the second part we focus on frequency domain photoacoustic microscopy (FD-PAM) that constitutes a powerful cost-efficient imaging method integrating intensity-modulated laser beams for the excitation of single-frequency PA waves. Nevertheless, FD-PAM provides extremely small signal to noise ratios (SNR), which can be up to 2 orders of magnitude lower than the conventional time-domain (TD) systems. To overcome this inherent SNR limitation of FD-PAM, we utilize a U-Net neural network aiming at image augmentation without the need of excessive averaging or the application of high optical power. In this context, we improve the accessibility of PAM, as the system's cost is dramatically reduced and we expand its applicability to demanding observations while retaining sufficiently high image quality standards. In the final part we transform classical data into quantum through a density matrix approach and compare classical and quantum machine learning methods in data applications. Host: Avadh Saxena (T-4)