System integration, signal processing and application of smart compression wearable system

Abstract

Compression therapies have a historical tradition spanning thousands of years. In contemporary practice, compression garments are extensively utilized for managing chronic venous diseases, scar management, orthopedic applications, body shaping, sportswear, and other uses. The development of imperceptible, multifunctional, long-term wearable smart compression garments is highly desirable, as these can monitor pressure variations and provide valuable information during therapy, potentially enhancing therapeutic efficacy. However, traditional compression garments lack real-time pressure measurement capabilities. Several laboratory investigations into smart compression garments have been conducted by various researchers, but these prototypes are often single-functional, bulky, and unsuitable for prolonged wear. Consequently, this study proposed an innovative integrated smart compression wearable system, comprising compression garments, embedded fabric capacitive pressure sensors, edge control units, interactive user interfaces, and dedicated software applications. This system offers numerous advantages, including high sensitivity to low pressure, mechanical flexibility, integration with external systems, rapid internal component upgrades and simplified repairs, configurable software parameters and modular hardware components, and a low cost of US$ 33. To accurately measure static and dynamic pressure on the human body, textile capacitive pressure sensors with a sandwich structure were designed, tested, and integrated. These sensors feature a dielectric layer filled with polydimethylsiloxane mixed with carbon black powder and roughened by abrasive papers. The laboratory-fabricated sensors exhibit several benefits, including facile fabrication, low cost, satisfactory accuracy and repeatability, high conformity to curved surfaces, adjustable integration with compression garments, high sensitivity below 50 mmHg, energy-efficient design, and rapid response times. The sensing performance of the smart compression garment system is compromised by parasitic capacitances caused by surrounding electromagnetic interference, proximity effects, and deformation on curved surfaces, which manifest as frequency-overlapped and non-stationary noise in real-world applications. Traditional deterministic and stochastic signal processing methods are inadequate for improving denoising performance. Therefore, a novel encoder-decoder deep neural network architecture was developed to enhance noise reduction and achieve high-resolution pressure detection. This architecture employs stacked autoencoders for the encoders and is trained for various tasks for the decoders, offering high interpretability, cost-effective unsupervised learning, and compatibility with edge control units with limited computational and storage capacities. Empirical validations included 100-minute flat surface pressure recordings with seven standard weights, producing 22,498 capacitance-pressure data pairs for algorithm training and testing. The algorithms achieved a root mean square error (RMSE) in pressure measurement of 0.7891 mmHg, a 38% improvement over traditional polynomial regression methods. Subsequent fine-tuning over 1000 epochs with the substitution of various frozen encoders resulted in an RMSE of approximately 0.9 mmHg. Additionally, the system's effectiveness was verified through 12-minute curved surface pressure recordings over ten days, producing 75,888 capacitance-pressure data pairs. The textile capacitive pressure sensors were attached to position B, B1, and C on a medium-sized wooden mannequin leg, with a sphygmomanometer used to tightly wrap the sensor on the leg and exert seven different pressure levels. The algorithms exhibited RMSEs of 0.0283 mmHg, 0.0633 mmHg, and 0.0387 mmHg for positions B, B1, and C, respectively, representing improvements of 98%, 97%, and 98% over traditional polynomial regression methods. Continuous monitoring of lower extremity muscle function is essential, given the critical role these muscles play in posture maintenance, locomotion, and dynamic movements. However, conventional assessment techniques, such as electromyography and physiological cross-sectional area measurements, often lack the capacity to deliver accurate, real-time data while maintaining user comfort and practicality in both clinical and community environments. To address these challenges, this study introduced an application of the proposed smart compression stocking system. A clinical validation study was conducted involving twelve healthy young adults who performed maximum voluntary isometric contractions of ankle plantarflexion under standardized conditions. Muscle force data were collected simultaneously using the smart compression stocking system and a calibrated Humac NORM dynamometer, which served as the reference standard. Statistical analysis demonstrated strong linear correlations between the outputs of the two systems, with correlation coefficients exceeding 0.92. Further, two-way analysis of variance demonstrated that the ankle joint angle (p = 0.055) had a more significant impact on measurement outcomes compared to inter-participant variability (p = 0.290). These results validate the smart compression stocking system as a reliable and practical tool for monitoring lower extremity muscle force during isometric contractions. The system holds significant potential for applications in clinical assessment, rehabilitation monitoring, and sports performance evaluation. In summary, this thesis comprehensively and systematically studied, designed, implemented, and optimized wearable capacitive pressure sensors, tailored system integration of the smart compression garment system, unified signal processing methods, and evaluated the system by healthcare applications. The aim is to deliver cost-effective pressure management and relevant healthcare information for anyone, anytime, anywhere.