Prediabetes Detection and Healthcare Accessibility: GlucAI

2025 HSHRC Finalist Proposal

By: Nisha Evangelista, 2025 HSHRC Finalist

Prediabetes affects approximately 96 million adults in the United States, yet over 80% of cases remain undiagnosed. (CDC, 2022). When left untreated, prediabetes can progress to type 2 diabetes, posing more significant health risks and more severe health complications. (American Diabetes Association, 2023). Early identification and intervention are critical, and lifestyle modifications at this stage can reduce case worsening by up to 58% (Diabetes Prevention Program Research Group, 2002). Sometimes, this critical phase could even mark the difference between life and death.

Despite the availability of diagnostic tools, limitations such as accessibility and lack of resources create considerable detection gaps, seen most commonly in underserved communities. Conventional diagnostic tools for prediabetes, such as fasting plasma glucose tests and hemoglobin A1c tests, require visits to medical facilities. Such requirements pose challenges for underserved communities, where healthcare infrastructure is limited. Research indicates that residents in rural areas of the United States are 14% less likely to receive preventative healthcare screenings compared to their urban counterparts (National Rural Health Association, 2021). The financial burden associated with diagnostic testing continues to amplify health disparities.

Studies demonstrate that biomarkers present in saliva can serve as reliable indicators of glucose dysregulation, or abnormalities in blood sugar stability. For instance, a correlation has been found between elevated levels of salivary glucose and prediabetic states (Patil et al., 2021). Additionally, a promising frontier has been emerging and transforming the future of healthcare: artificial intelligence. The integration of the AI diagnostic kit, GlucAI, would mitigate the issues above with both preventative and innovative measures through accessible technology.

GlucAI is a saliva-based diagnostic kit utilizing reagent-embedded test strips to detect prediabetes biomarkers. Using the GlucAI smartphone app, the test strip would be scanned and Convolutional Neural Networks (CNNs) would process the image to identify patterns and quantify glucose levels. Using frameworks like TensorFlow Lite, CNNs can be efficiently optimized to run directly on mobile devices. As an example, this would enable CNNs to process images, such as analyzing diagnostic test strips, without the need for external servers. TensorFlow Lite has the ability to compress CNNs, making them lightweight and suitable for resource-constrained environments.

With over 95% accuracy in analyzing diagnostic test images, (in optimal lighting and alignment conditions), CNNs also exhibit high precision and reliable results, even with a lack of resources. (Dastagir et al., 2024). In areas where individuals lack smartphones for scanning, the kits can be deployed in community centers or local clinics, where shared devices could be used to scan and process results. Health workers could also use their smartphones to assist those without devices. If transportation access is limited, the kits could be paired with standalone diagnostic readers, still allowing communities to benefit from the diagnostic capabilities.

To ensure affordability, the kits would utilize low-cost materials and focus on scalability through mass production. They are designed for single-use, minimizing costs while offering reliable diagnostics. The accompanying GlucAI smartphone app would provide immediate diagnostic feedback. The app would also function offline, overcoming barriers posed by unstable internet access in remote areas. Through awareness campaigns, community outreach, fundraising initiatives, and tech innovation, even teenagers can help expand the reach and impact of GlucAI, empowering individuals and communities to take charge of their health. The implementation of pilot programs in underserved regions could also provide necessary statistics on the efficacy and adoption of these kits. As an example, deploying 1,000 diagnostic kits in a low-resource community could identify an estimated 200 individuals with undiagnosed prediabetes, while simultaneously offering a critical opportunity for intervention, potentially saving a life.

Works Cited

American Diabetes Association. (2023). "Complications of Diabetes." Retrieved from https://diabetes.org/

Centers for Disease Control and Prevention. (2022). "Prediabetes Facts." Retrieved from https://cdc.gov/diabetes/basics/prediabetes.html

Dastagir, R. B., et al. (2024). "AI-Driven Smartphone Solution for Digitizing Rapid Diagnostic Test Kits and Enhancing Accessibility." arXiv, 27 Nov. 2024, https://doi.org/10.48550/arXiv.2411.18007.

National Rural Health Association. (2021). "Rural Health Disparities: Bridging the Gap." Retrieved from https://www.ruralhealthweb.org/

Patil, A., Joshi, R., & Sharma, P. (2021). "Salivary Biomarkers for Early Detection of Prediabetes." Journal of Clinical Biochemistry, 12(3), 145-153. "TensorFlow Lite." AI at Google Developers. Retrieved from https://ai.google.dev/edge/litert

WHO. (2020). "Universal Health Coverage Data." Retrieved from https://www.who.int/