Spectral Analysis of AVF Signals for Early Dysfunction Detection: Towards AI-Based Home Monitoring

Arteriovenous fistula (AVF) dysfunction is a critical complication in hemodialysis patients, often requiring timely intervention to avoid severe clinical consequences. This work presents a signal processing pipeline based on spectral analysis for early detection of AVF dysfunction, using acoustic signals recorded via a portable, wearable acquisition system. The custom-designed sensor platform captures mechanical vibrations from the AVF site, enabling the collection of real-world vascular signals in both stenotic and non-stenotic conditions. The signals are analyzed using deterministic techniques, including Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), and Mel spectrograms, from which twelve frequency-domain features are extracted and evaluated. Results indicate that features such as Area Under the Curve (AUC), f95, and skewness show strong discriminatory power between pathological and healthy cases. These findings demonstrate the feasibility of lightweight, embedded-compatible algorithms for future integration into AI-based, real-time monitoring platforms. The proposed approach lays the groundwork for continuous, at-home AVF monitoring through wearable biomedical technologies.