SugarVoice

The goal of the Sugar Voice Project is to develop deep learning models capable of detecting voice changes in diabetic patients. A non-invasive, remotely accessible, and real-time screening tool holds immense potential for early disease detection and complication prevention. Following this, we will evaluate the clinical effectiveness of these machine learning models using real-world cases. We hypothesize that leveraging audio data for screening could greatly aid in diabetes detection. This cost-effective, non-invasive test can be conveniently conducted at home, widening its accessibility to diverse populations. The potential impact of this innovation in enhancing diabetic patient identification cannot be emphasized enough.

Publications

Our Drive

The urgency behind the Sugar Voice Project stems from the profound impact of diabetes, a chronic condition disrupting glucose regulation globally. With over 422 million individuals affected by diabetes mellitus (DM), projections indicate a startling rise in prevalence, exacerbating the burden of undiagnosed cases by 25% by 2030 and 51% by 2045.

Alarming statistics reveal that approximately 25% of diabetes cases remain undetected, contributing to increased in-hospital mortality rates and poorer outcomes for patients with acute critical illnesses. The escalating prevalence of undiagnosed diabetes, which has surged by 82% over five years, incurs a substantial financial burden exceeding $31.7 billion.

Globally, diabetes consumes a staggering 12% of total healthcare expenditure, amounting to an astonishing $673 billion. Furthermore, the National Institutes of Health (NIH) reports 8.5 million undiagnosed adults aged 18 years and older, representing 23% of the adult diabetic population.

Given the challenges associated with laboratory blood tests for diabetes detection, particularly among unaware individuals, the elderly, and the disabled, there is an imperative need for innovative, non-invasive screening methods. The Sugar Voice Project emerges as a response to this pressing need, aiming to harness the potential of voice analysis as a novel tool for early diabetes detection and intervention."

Progress

The current clinical diagnosis of diabetes poses significant financial burdens and inconvenience, especially for elderly individuals. Recognizing the need for alternative solutions to address these challenges, our pioneering work began. This innovation holds the promise of revolutionizing diabetes diagnosis, potentially alleviating the substantial costs associated with repeated testing.

Through our preliminary investigations, we've uncovered distinct voice characteristics differentiating individuals with diabetes from those without the condition. Notably, diabetic patients exhibit higher pitch, increased shimmer, elevated Cepstral Peak Prominence (CPP), heightened jitter, and a lower Harmonic-to-Noise Ratio (HNR) compared to non-diabetic counterparts."