In March 2020, the World Health Organisation (WHO) identified COVID-19 as a global pandemic brought about by the SARS-CoV-2 virus. RT-PCR continued to be the first line of diagnosis, but its slow turnaround emphasized the need for faster, complementary diagnostic modalities. Imaging techniques like chest X-rays (CXR), computed tomography (CT) scans, and ultrasound were soon in high demand; yet manual assessment was time-consuming and susceptible to faults. To overcome these challenges, artificial intelligence (AI), and more so deep learning (DL), has proven to be a revolutionary instrument through feature extraction automation and enhanced diagnostic accuracy. This survey differentiates itself from reviews that have already been conducted by providing an exhaustive review of state-of-the-art DL architectures engineered for COVID-19 analysis on various imaging modalities, also highlighting under-investigated aspects like prognostics, rehabilitation assistance, and the role played by uncertainty quantification (UQ) in achieving clinical trustworthiness. The survey identifies areas of research gaps, such as the scarcity of multimodal datasets, difficulties in generalising models across populations, and the absence of standardised evaluation benchmarks. Systematically resolving these gaps, this work highlights the practical significance of AI-based computer-aided diagnosis (CAD) systems toward accelerating faster, more robust, and scalable pandemic response tools. In addition, it gives researchers and clinicians a blueprint for developing AI-based healthcare, allowing for both short-term use in managing COVID-19 and long-term relevance in future public health emergencies.

COVID-19 Systematic Review Deep Learning Medical Imaging Artificial Intelligence Classification Segmentation