Bio-Integrated Artificial Intelligence: Synergizing Nanomaterials, Machine Learning Algorithms, and Self-Powered Devices for Early-Stage Healthcare Diagnostics

Abstract

In an era driven by technological advancements, the intersection of functional materials, artificial intelligence (AI), and bioelectronics holds unprecedented promise for transformative healthcare applications. The ever-growing demand for non intrusive, real-time health monitoring solutions has spurred a surge in research towards innovative technologies. This thesis addresses the imperative need for multifaceted advancements in functional materials, focusing on their integration into bio-integrated intelligent systems for early-stage disease diagnosis, particularly chronic obstructive pulmonary disease (COPD), cardiovascular disease, vocal cord disorder, and Parkinson's disease. Utilizing piezoelectric and triboelectric phenomena, self-powered devices emerge as a prominent avenue for achieving bio integrated systems in an easy and efficient manner. The scientific quest commences with an investigation of nylon-11 nanofibers under alternate voltage biasing to activate piezoelectric δ‟-phase, where piezoelectricity is probed with piezoresponse force microscopy (PFM). By leveraging electrospinning under negative bias polarity, these nanofibers exhibit a remarkable ~3-fold enhancement in the piezoelectric charge coefficient (d33~ -27 pm/V) than conventional positive bias in electrospinning. It sets the stage for the development of self-powered devices with enhanced mechano-sensitivity. It is further revealed that the increased molecular dipole alignment and crystallinity, as elucidated by polarized Fourier transform infrared spectroscopy, plays a pivotal role in superior piezoelectric characteristics. The resultant piezoelectric nanogenerator (PNG) exhibits a paradigm shift in bio-sensing capabilities, enabling real-time tracking of physiological events, such as arterial, carotid pulses, and facial movements. With the flipping of the dipoles by using alternate bias polarity in electrospinning, the surface potential of these nanofibers has been further probed with Kelvin probe force microscopy (KPFM). This study reveals a prominent change in their triboelectric properties moving from tribopositive nature to tribonegative. This novel approach led to the foundation of a single-material-driven triboelectric nanogenerator (S-TENG). Exhibiting remarkable mechano-acoustic sensitivity (~27,500 mV Pa–1), ~6 times higher than conventional electret microphones, S-TENG exhibits its capabilities as a voice recognition sensor, demonstrating promising performance in the detection of laryngeal disorders. vi The voyage for scientific advancements continues by inducing an electroactive δ‟ phase in nylon-11 films by easy and scalable solution processable technique, enabling the fabrication of flexible and wearable piezoelectric sensors for continuous physiological monitoring. Owing to its high mechano-sensitivity (~225 mV N−1) and rapid responsivity (~4 ms), It has been further integrated seamlessly with the Internet of Things (IoT) and deep learning algorithms for real-time monitoring of cardiovascular risks with accuracy > 94%, leveraging variations in physiological conditions for different age and the body mass index (BMI) as descriptors. Moving towards sustainable solutions, we introduce a wearable gadget fabricated from recycled components of nylon-66. This eco-friendly, flexible gadget efficiently tracks diverse body movements, and gait analysis, leading to the early-stage assessment of Parkinson's disease and other kinesiological disorders. This study comprised the investigation of various ML algorithms, including support vector machine (SVM), k-nearest neighbor (KNN), k-means clustering, and pattern recognition. Encompassing the feature of micropattern learning, the pattern recognition algorithm is found to be the best, with ~98 % accuracy. Delving into the convergence of printed electronics and AI/ML programming, electronic skins (e-skin) emerge as a promising methodology for crafting ultrasensitive electromechanical sensors, stretchable and conformable electronics. In the realm of healthcare, this technology proves instrumental in precisely capturing therapeutic biomarkers for the early prediction of diseases. Employing the screen printing technique and polyvinylidene fluoride-tetrafluoroethylene (PVDF-TrFE) ink, we materialize a flexible piezoelectric polymer interface that exhibits unparalleled sensitivity and durability. Having strong directionality, stability over a large operating temperature range (30–90 ˚C), broad frequency (~6 kHz), ultra sensitivity (~5.77 V Pa−1), and high signal-to-noise ratio (38–55 dB). This lays the foundation for an innovative voiceprint biometrics technology, adept at capturing distinctive voiceprints with an accuracy exceeding 96% for hoarseness and other vocal cord disorders identification. To further explore the potential of e-skin sensors in the respiratory domain, a wearable electronic skin respirometer has been demonstrated. Owing to the precise conversion factor (~12 mL/mV), signal-to-noise ratio (~58 dB), and low limit of detection down to ~100 mL contributes to its effectiveness in recording diverse breathing signals. This non-intrusive marvel monitors lung function with vii ~95% accuracy in detecting chronic obstructive pulmonary diseases (COPDs), particularly emphysema and chronic bronchitis. The e-skin devices have been further utilized to monitor diverse physiological cues, offering insights into the human emotional landscape with implications for various health conditions. Emotions are closely tied to cognitive behavior and are indicative of conditions such as anxiety, depression, and stress. We have employed multi-modal sensing involving multiple physiological parameters to identify cognitive behaviors that are associated with diverse neurological disorders. Leveraging Long Short-Term Memory (LSTM) neural networks and reinforcement Q-learning, the translation of these cues into emotional responses not only aids in understanding health conditions linked to emotional well being but also contributes to development of cognitive robotics, marking strides toward artificial general intelligence (AGI). In the last section of the thesis, we have explored the power of quantum entanglement for processing vast, complex, and random datasets promptly. We delved into the realm of quantum computing and quantum machine learning by harnessing the properties of the feature maps we have demonstrated to solve intricate datasets efficiently with enhanced accuracy and speed. Thus, the present thesis collectively contributes to the evolving landscape of bio-integrated artificial intelligence, bridging the gap between traditional diagnostics and the dynamic needs of modern healthcare systems.