RTAEI: Robust Tabular Autoencoder Interpolator to Gastric Cancer Innovative Detection for Deep Learning Empowered Healthcare Electronics

Abstract

The recent integration of Artificial Intelligence (AI) into smart consumer electronics and sustainable healthcare has shown promising outcomes. However, challenges persist in utilizing medical data, mainly due to its imbalance. Imbalanced data may lead to the model’s inability to adequately learn the relationship between labels and features for the minority samples, resulting in unreliable prediction outcomes. What’s more, the existing gastric cancer detection methods are invasive, costly, and not suitable for widespread use. In this paper, a deep learning algorithm with Robust Tabular AutoEncoder Interpolar (RTAEI) is designed to screening for gastric cancer patients. First, we collected questionnaire survey data and crucial biochemical indicators from patients to form our dataset, which covers dietary habits, and H pylori infection status. Second, the Robust Tabular AutoEncoder Interpolator generated tabular data to balance the dataset. This involved encoding sparse data into a dense latent space using robust variational autoencoders and generating synthetic samples through SMOTE. Third, the generated data was input into subsequent models to predict the risk of gastric cancer. Experiments show that combining RTAEI and deep learning models, the highest AUC reached 0.792, makes it a non-invasive, convenient, and rapid method for gastric cancer screening.