SmartBio: An AI-Enabled Smart Medical Device for Early Cancer Detection using Variational Autoencoders and Multimodal Sensor Integration

Abstract

This research explores the capability of a generative AI model called Variational Autoencoder (VAE), leveraging device sensors such as breath acetone and sweat biomarkers to identify life-threatening diseases, such as cancer, diabetes, and heart disease at earlier stages and help address metabolic issues. These sensors are intended to be integrated into smart devices such as wearable fitness trackers or smartwatches. The sweat biomarker sensor collects data from perspiration, including lactate, glucose, cortisol, and sodium levels. The breath acetone sensor measures the concentration of acetone in exhaled breath a byproduct of fat metabolism that reflects metabolic state. Both sensors can help assess mitochondrial quality, a core parameter for predicting diseases like cancer, diabetes, and cardiovascular disorders. The work demonstrates the efficacy of the system, achieving a training accuracy of 92%, testing accuracy of 89%, and an anomaly detection rate of 90%, with a low false positive rate of 5%. A reconstruction error threshold of 0.1 was empirically determined to differentiate between normal and abnormal patterns. The system’s architecture built on Azure cloud and edge infrastructure supports secure data storage, low-latency inference, and personalized health recommendations via mobile interfaces. Overall, SmartBio offers a proactive and scalable solution for personalized metabolic health monitoring, paving the way for early intervention and lifestyle-driven disease prevention.