Design of a compact wearable functional electrical stimulation system for foot drop correction with AI driven foot lift detection

Abstract

Foot drop is the inability to lift the front part of the foot. It happens due to conditions such as stroke, spinal cord injury, and multiple sclerosis. Functional Electrical Stimulation (FES) can correct foot drop by applying electrical pulses to targeted nerves. However, most existing FES systems do not properly balance the electrical charge of these pulses. It is important to apply the electrical pulses only when the person lifts their foot. A foot sensor is used to detect the foot lift event. However, current sensors use a cable which creates cable complexity and they are uncomfortable to wear. To address these issues, we have designed a compact, wireless, charge-balanced, trapezoidal FES system with an Al-driven foot lift detection system. The thesis is divided into six chapters. Chapter 1 serves as an introduction, providing an extensive literature review, defining the problem, and outlining the specific objectives of this research. In Chapter 2, we have discussed the development of the circuit for the FES system. We explored various circuit designs before finalizing one. This chapter provides a detailed description of all system components, explaining the process from the initial schematic design to the final PCB layout. We have also presented the results of our circuit testing and examined the characteristics of the stimulation, particularly focusing on pulse width and frequency. Chapter 3 focuses on the power management system, detailing how we effectively supplied power to all components. We have developed positive and negative boost converter to power the high power amplifier. Chapter 4 focuses on the detection of foot lift in the FES system. It covers the ethical clearance and the procedures for data collection from patients. This chapter also details the development of a machine learing algorithm specifically designed for foot lift detection within the FES system. We trained the model using three different approaches and implemented the final machine learning model on an ESP32 microcontroller. In Chapter 5, we have discussed the incorporation of all the subsystems that we have developed in previous chapters. We have also tested the device and calculated the accuracy. The conclusions and future perspectives of this work have been discussed in Chapter 6.