Speaker
Description
Atherosclerotic disease of the carotid artery is a chronic pathological condition characterized by the progressive accumulation of lipid-rich plaques within the arterial wall, resulting in structural remodeling, altered hemodynamic conditions, and luminal narrowing, thereby posing a high risk of ischemic stroke. In this context, early identification of carotid stenosis is crucial to enable timely intervention and reduce the likelihood of ischemic events. To this end, several imaging modalities are currently used to screen the carotid artery, including computed tomography angiography (CTA), magnetic resonance angiography (MRA), and catheter-based angiography (CBA). However, except for carotid ultrasound (CU), these techniques are either invasive or require ionizing radiation or contrast agents. From this perspective, the present work presents an in-silico framework designed to explore the potential of infrared thermography (IRT) based approaches for detecting carotid stenosis from skin thermal footprints of the vessel. The underlying rationale is that atherosclerotic plaque-induced alterations in vascular geometry and blood fluid dynamics modify local heat transport mechanisms, thereby producing surface temperature variations in the surface temperature distribution. A synthetic thermographic dataset was generated using 3D multiphysics simulations (through GPU-accelerated finite difference method with immersed boundary methods), in which pulsatile blood flow and heat transfer were solved in a coupled manner. Precisely, the model resolved fluid dynamics within the vessel, conductive heat transport in the vessel and in the surrounding tissue, and convective heat exchange at the skin-air interface. The simulation framework implemented a structured parametrization of three boundary physical parameters. Blood inflow rate (Qin), convective heat transfer coefficient (h), and ambient air temperature (Tair) were varied according to a one-factor-at-a-time protocol, while all other parameters were held constant. This approach produced 36 virtual subjects, each one simulated until a periodic regime is attained. Then the unsteady dynamics within 5 seconds, corresponding to 50 consecutive frames, capturing the cardiac cycles driven thermal response, are saved. For each subject, three physiopathological configurations were simulated, corresponding to 0% (healthy), 30%, and 70% carotid stenosis, resulting in a set of 1800 in-silico thermographic images representing vessel-induced thermal footprints on the skin surface. Furthermore, the simulations indicated that stenosis-related surface temperature changes range in the order of hundredths of a degree Celsius, which fall within the measurable sensitivity actual IRT systems. Additionally, a VGG16 model was finetuned on the in-silico thermographic dataset within a group k-fold cross-validation framework to distinguish between the three stenotic scenarios, reaching a test accuracy of 95%. These findings suggest that IRT, combined with deep learning (DL) algorithms, may effectively capture subtle hemodynamic and thermal changes caused by atherosclerotic plaques, supporting its feasibility as a non-invasive tool for early carotid artery assessment.
