Deep learning in the early diagnosis of acute aortic dissection

This review explores the potential of deep learning (DL) to enhance the diagnostic efficiency of acute aortic dissection (AAD), a life-threatening cardiovascular emergency characterised by high misdiagnosis rates. In contrast to a prior systematic review that evaluated DL approaches to aid diagnosis, this review focuses specifically on the early diagnostic phase, examining routinely available investigations, including the electrocardiogram, chest x-ray and tabular clinical data, to inform decisions regarding which patients should and should not proceed to advanced imaging. We analyse current diagnostic challenges contributing to AAD misdiagnosis and existing diagnostic decision support pathways. The review also examines healthcare data modalities and DL architectures capable of identifying complex patterns and relationships not readily recognised by clinicians. We summarise existing DL research in the early phase of AAD diagnosis aimed at improving diagnostic accuracy and offer insights into future research directions. Finally, challenges relevant to DL-based AAD diagnosis are discussed, including data scarcity, class imbalance due to its low prevalence, and the role of explainable AI.