A Systematic Review of Deep Learning Algorithms for the Early Detection of Chronic Disease

Cancer, cardiovascular diseases, diabetes mellitus, heart disease, liver disease and neurological disorders are the leading causes of death and disability worldwide. Hence, it is important to diagnose early and accurately. Recent advances in deep learning have significantly improved early disease diagnosis, where features can be learned automatically from different health care data, such as medical images, physiological signals, electronic health records, laboratory measurements and genetic data. This study is a systematic review of algorithms using deep learning for early detection of chronic diseases following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. A systematic literature search with pre-defined search strategies was followed by research selection, quality assessment, data extraction and qualitative synthesis. Deep learning architectures commonly used in the major chronic disease domains are Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long ShortTerm Memory (LSTM) networks, Gated Recurrent Units (GRUs), Autoencoders, Generative Adversarial Networks (GANs), Vision Transformers (ViTs), Graph Neural Networks (GNNs) and hybrid models. The paper describes the preprocessing, validation, performance evaluation criteria, research directions and datasets available. It also covers the pros and cons of current approaches. The results indicate that deep learning has great potential to improve the automated detection of chronic diseases. However, its wide application is still limited by problems of annotated dataset shortage, class imbalance, model interpretability, external validation, data privacy and clinical implementation. In this paper, we propose ideas to develop more reliable, accurate, interpretable and therapeutically beneficial deep learning systems for early detection of chronic diseases. It offers a comprehensive analysis of the existing literature and assists in identifying knowledge gaps.