A Wearable Multisensor Platform for Neurofibromatosis Surveillance Using Bioimpedance, Bioelectrical Signals, and TinyML

Abstract

Neurofibromatosis comprises genetically distinct tumour-predisposition disorders that require lifelong, risk-adapted clinical surveillance. Current follow-up relies principally on clinical examination and imaging, which are essential but intermittent and may not capture short-term changes in peripheral-nerve function or local tissue electrical properties. This paper presents the design and bench-level evaluation of a wearable, non-invasive proof-of-concept platform that combines multifrequency bioimpedance sensing, amplified peripheral bioelectrical-signal acquisition, embedded feature extraction, and TinyML-based risk-pattern classification. An AD5933 impedance-converter circuit is used to estimate impedance magnitude and phase, whereas an AD620 instrumentation-amplifier stage conditions low-amplitude nerve- or electromyography-like signals. An ESP32 microcontroller performs data acquisition, filtering, feature computation, model inference, and Bluetooth Low Energy transmission. The proposed feature set includes mean impedance, phase angle, root-mean-square amplitude, and zero-crossing rate. Bench testing with simulated impedance loads and generated bioelectrical waveforms demonstrated stable acquisition and real-time wireless display. However, the reported thresholds and classification outputs are engineering labels for prototype verification and are not validated diagnostic criteria. Patient recruitment, clinical reference standards, sensitivity, specificity, calibration, and longitudinal outcome analysis remain necessary before any clinical interpretation. The present work therefore establishes an embedded-systems architecture and feasibility workflow rather than a clinically validated neurofibromatosis diagnostic device.