Abstract Background/Aims Diagnosing axial spondyloarthropathy (AxSpA) can be challenging due to the absence of specific clinical manifestations, unique biomarkers and radiological findings, often resulting in an incorrect or delayed diagnosis. Machine learning (ML) - a field of artificial intelligence (AI) - and its subset, deep learning (DL), have emerged as potential tools to address these challenges. ML is capable of analysing multidimensional data and offering insights to enhance clinical diagnosis, imaging interpretation, disease prognostication and management. This review aims to explore recent advances in the application of AI in AxSpA, with emphasis on early detection and diagnostic accuracy. Methods We conducted a systematic search on PubMed for peer-reviewed studies published within the past 5 years (2020-2025) that explored the use of ML and DL in AxSpA, with a focus on its diagnostic application. Case reports, conference abstracts and studies lacking primary data were excluded. Studies that fulfilled the inclusion criteria were examined and appraised at the abstract and full text stages. Results ML has demonstrated significant potential in aiding in the diagnosis of AxSpA. It has been increasingly employed in image feature recognition and medical image processing. Several studies have developed models that have successfully detected inflammatory changes associated with AxSpA MRI scans and even pelvic radiographs, with some models matching or outperforming experienced radiologists. Alternative applications of AI in AxSpA imaging included accurate grading of disease severity, predicting disease progression and response to treatment. ML methods have also been used to identify and understand the characteristics of people who are likely to be diagnosed with AxSpA in the future, with some models achieving a positive predictive value (PPV) of approximately 70-80% when applied to test data. This highlights the feasibility of population-level screening tools to reduce diagnostic delay. However, current studies have small sample sizes and predominantly consist of retrospective, single-centre studies, limiting the application of its results in a broader population. Conclusion ML can help facilitate early and accurate diagnosis, and is particularly beneficial in settings with limited resources, where specialist radiologist or advanced imaging such as MRI may not be readily available. Whilst there have been rapid advances in the use of ML in diagnosing AxSpA, current evidence is limited by small datasets and the lack of prospective, multicentre studies. Moreover, the algorithm is reliant on the quality and completeness of the available data; missing or mislabelled records can undermine the model’s performance. To move towards clinical application, multicentre studies and large-scale datasets are required, whilst also addressing the ethical and general principles of handling big amounts of data. Disclosure P. Millan: None. H. Chong: None. A. Moorthy: None.
Millan et al. (Wed,) studied this question.