A Survey of Methods and Architectures for Convolutional Autoencoder with Dual-Key Transformer Network Based Smart E-Health Application for the Prediction of Tuberculosis Using Serverless Cloud Computing

Abstract

Tuberculosis (TB) remains one of the leading causes of mortality worldwide, necessitating efficient, scalable, and accurate diagnostic systems. Recent advancements in artificial intelligence (AI), particularly deep learning (DL), have enabled automated TB detection using medical imaging and clinical data. This survey explores state-of-the-art methods and architectures integrating Convolutional Autoencoders (CAE), dual-key transformer networks, and serverless cloud computing for smart e-health applications. CAEs play a critical role in unsupervised feature extraction and dimensionality reduction, improving model robustness and handling noisy medical datasets. Transformer architectures, particularly Vision Transformers, enhance global feature learning through attention mechanisms, achieving high accuracy levels (up to 98%) in TB detection tasks. Hybrid CNN-transformer models further improve classification by combining local and global feature representations. The integration of serverless cloud computing enables real-time processing, scalability, and cost efficiency, facilitating remote healthcare delivery and automated diagnosis systems. Cloud-based AI models have demonstrated diagnostic accuracy exceeding 93% in TB detection using deep learning approaches. This survey critically analyses existing methods, identifies research gaps, and highlights future directions for integrating CAE, transformer networks, and cloud computing in intelligent TB prediction systems