The future’s bright, the future’s Orange AI… If you Google ‘how much is AI worth?’, the AI overview tells you “The global artificial intelligence (AI) market size is estimated at approximately $244 billion to $391 billion in 2025 and is projected to reach over $3.4 trillion by 2033.” The uptake of AI in all aspects of our life is rising at an exponential rate – in fact, this week saw three AI-generated songs hit the top spots in Spotify and the American Billboard charts. AI in modern medicine is weaving its way into contemporary practice, evidenced by the fact that two of the articles in this issue’s journal reviews are focused on this very subject. The two editors’ choices look at the role of AI in swallowing assessment with videofluoroscopy and AI in the diagnosis of obstructive sleep apnoea. Both articles find that use of AI is a mixed bag – some aspects of diagnosis are very good. But it’s not good at the whole process, due to issues around the decision-making process in AI models, validation against diverse patient groups and challenges with integration of AI in routine work. Like it or loathe it, AI is clearly here to stay and will be a part of all our future practices. However, bearing in mind the words of the CEO of Alphabet (Google’s parent company) in a BBC interview in November, who stated that people should not “blindly trust” everything AI tools tell them, humans will always be needed in healthcare! As always, the editors would like to thank all of our journal reviewers for their contributions.
Nazia and Gaynor
It is widely accepted that videofluoroscopic assessment of swallowing (VFSS) provides the current gold standard for assessing and diagnosing oropharyngeal dysphagia. Over the last decade, there have been several developments in standardising the interpretation of VFSS, most notably the modified barium swallow impairment tool (MBSImP). This involves the clinician rating 17 components of swallow biomechanics, including safety (airway protection) and efficiency (remaining residue in the pharynx after swallow). However, these methods require considerable training and remain reliant on clinician skill, experience and interpretation based on their visuo-perceptual analysis of the fluoroscopy images. The authors of this paper review the current up-to-date evidence of the application of artificial intelligence (AI) to the process of analysing VFSS images. Following PRISMA guidelines for systematic reviewing, 30 English-language articles published between 2018 and 2024 met their inclusion criteria. They categorised the articles based on three common themes observed across the studies: swallowing phase analysis, segmentation or detection, and penetration-aspiration detection. They also describe the different AI metrics and frameworks used (machine learning and deep learning, evaluation metrics) across these studies. The review found that good progress has been made in airway invasion detection, swallowing phase analysis and localising key components such as the bolus and hyoid bone. There is much further work to be done before AI is a viable tool that can be used by clinicians and the lack of a public database of VFSS is currently the biggest challenge to advancing this work. This paper is a useful overview of the status of research in the field of AI applications to VFSS analysis.
