Deep learning–integrated multilayer thermal gradient sensing platform for real-time blood flow monitoring

Professor Kyeongha Kwon’s research team at KAIST has developed a wireless wearable sensing platform that enables real-time, noninvasive monitoring of blood flow by integrating multilayer thermal sensing with deep learning. The system simultaneously measures blood flow rate and vessel depth, overcoming a major limitation of conventional thermal sensing approaches, and further improves continuous blood pressure estimation when combined with photoplethysmography (PPG).

Blood flow is a key physiological indicator of cardiovascular health, tissue perfusion, and hemodynamic stability. However, accurate blood flow monitoring has typically relied on bulky clinical instruments such as Doppler ultrasound, limiting its use in continuous daily monitoring. Although wearable thermal sensors have emerged as a promising alternative, conventional single-layer devices cannot distinguish whether changes in temperature arise from blood flow itself or from differences in vessel depth beneath the skin. This makes accurate quantification difficult in real-world use.

To address this challenge, the KAIST team designed a multilayer thermal gradient sensing module that captures thermal signals at different heights above the skin. The platform uses two sensing layers with upstream, downstream, and reference thermistors arranged around a central thermal actuator, enabling it to capture both lateral and vertical thermal gradients generated by subsurface blood flow. A neural network processes six thermal inputs in real time to estimate both blood flow rate and vessel depth simultaneously. The wireless device further integrates sensing electronics with Bluetooth Low Energy communication, enabling real-time data transmission to a smartphone interface for deep learning-based analysis.

The platform demonstrated strong performance in both benchtop and on-body validation, measuring blood flow rates from 1 to 10 mm/s with an accuracy of ±0.12 mm/s and resolving vessel depths from 1 to 2 mm with a depth resolution of ±0.07 mm.

The researchers further showed that the platform can enhance continuous cuffless blood pressure monitoring when integrated with a PPG sensor. The combined thermal sensing–PPG approach tracked rapid blood pressure changes during a Valsalva maneuver more accurately than a PPG-only method, particularly during dynamic physiological transitions. Blood pressure estimation error was reduced by as much as 72.6%, highlighting the platform’s strong potential for wearable cardiovascular monitoring in daily life.

This technology could support early detection of hemodynamic abnormalities, personalized monitoring for patients with chronic cardiovascular conditions, and clinical applications such as skin graft surveillance and acute shock detection. This work was supported by the National Research Foundation of Korea, the BK21 FOUR program, IITP, and Samsung Advanced Institute of Technology (SAIT).