Sali, Saghil (2025) Development of an ultrasound sleeve for real-time diagnosis of fracture in the lower limb. Research Master thesis, Victoria University.

Abstract

Fracture detection in clinical and emergency settings predominantly relies on X-ray imaging, which, despite its efficacy, is constrained by radiation exposure, high costs, and the necessity for centralised medical facilities. These limitations necessitate the development of a real-time, wearable diagnostic solution capable of providing immediate, non-invasive fracture assessment in diverse settings such as sports medicine, field emergencies, and remote healthcare. The Capacitive Micromachined Ultrasonic Transducer (CMUT)-based ultrasound sleeve developed in this study aims to address these constraints by offering real-time fracture detection with enhanced sensitivity and portability. This research investigates whether CMUT arrays, coupled with real-time machine learning algorithms, can provide high-resolution diagnostic imaging comparable to conventional methods while maintaining wearability and accessibility. To achieve this objective, a structured research methodology was implemented, incorporating sensor fabrication, microelectronic integration, AI-driven ultrasound signal processing, and clinical feasibility assessments. To ensure optimal performance, the initial phase of this research focused on the design and fabrication of CMUT sensor arrays, emphasising material selection, cavity uniformity, and electrode stability to maintain signal clarity. The electrical and acoustic properties of the transducers were meticulously examined through resonance frequency calibration, impedance analysis, and signal-to-noise ratio (SNR) testing. Results indicate that the proposed CMUT arrays demonstrated higher sensitivity and broader bandwidth compared to traditional piezoelectric transducers, confirming their viability for real-time medical ultrasound applications. While theoretical calculations of resonance frequency, capacitance, and pull-in voltage closely corresponded with experimental observations, minor deviations due to fabrication inconsistencies were identified and mitigated through iterative design refinements. Following sensor development, the research explored AI-driven ultrasound data processing, incorporating Support Vector Machines (SVMs) and real-time signal analysis techniques. Key ultrasound parameters such as time-of-flight (TOF), frequency response, and amplitude variation were analysed to assess their role in fracture detection. A dataset comprising simulated lower limb models with artificial fractures was developed to train and test the machine learning model. Various SVM classifiers, including linear, radial basis function (RBF), and polynomial kernels, were applied to optimise classification accuracy. Feature selection techniques such as Pearson's correlation, t-tests, and hill-climbing methods were utilised to refine the dataset, reducing the feature set from 19 to 4. To attain classification accuracy of 70-80% was achieved using the polynomial SVM kernel, demonstrating the efficacy of CMUT-based AI systems for automated fracture detection. To evaluate real-world usability, the wearable sleeve prototype underwent testing in a simulated diagnostic environment, with a focus on signal stability, real-time feedback, and clinical feasibility. The system was assessed across two cohorts—one utilising biofeedback-enhanced ultrasound monitoring and the other employing standard imaging techniques. Statistical analysis revealed that the biofeedback-integrated CMUT sleeve provided more expeditious and accurate diagnostic feedback, significantly improving fracture detection efficiency compared to conventional handheld ultrasound devices. Further refinements were implemented to enhance power optimisation, noise reduction, and electrode adhesion, ensuring greater reliability and readiness for future clinical trials. This research successfully demonstrates the feasibility of a CMUT-based wearable ultrasound sleeve for real-time fracture detection. Through the integration of high-resolution CMUT sensors with AI-driven data processing, the system enhances accuracy, portability, and accessibility in orthopaedic diagnostics. The findings highlight the potential impact of wearable ultrasound technology in emergency care, sports medicine, and point-of-care diagnostics, paving the way for future clinical applications and widespread adoption.

Search Google Scholar

Repository staff login