Synthetic Exhaled Breath Data-Based Edge AI Model for the Prediction of Chronic Obstructive Pulmonary Disease

Diseases that affect the respiratory system are one of the main causes of death across the globe. There is a need for a personalized, easy to use and convenient mechanism to self-detect a potentially contagious disease, thus limiting the spread of infections. The integration of Artificial Intelligence (AI) and Internet of Things (IoT) provides a great opportunity to bring detection and monitoring of respiratory diseases at home. However, the development of efficient AI models has been hindered by the lack of datasets for the targeted biomarkers, with privacy concerns limiting open data access and sharing. Starting from an existing small dataset of COPD, this study leverages the emerging synthetic data technology to artificially augment its size to have adequate data for training a TinyML model for predicting COPD using a Neural Network model, thus improving the inference accuracy of portable noninvasive self-diagnostic kits for respiratory disease. An online platform, Mostly AI is used to synthetically enhance data, the platform uses deep neural networks with inbuilt mechanisms that retain valuable information while providing a good as a real anonymous dataset. Next, the Keras Neural Network is used to train the edge AI models. The performance of the model was evaluated and the results show that when the same training parameters were applied, the model trained from synthetic data performed with an accuracy almost similar to that based on real open datasets. The use of synthetic data will complement the few breaths that are collected in healthcare facilities. This will enable the training of efficient AI models for respiratory diseases.