Background: Swallowing assessment aims to evaluate safety and efficiency across the phases of deglutition. While current instrumental assessments are able to capture the consequences of impaired swallowing physiology, e.g. aspiration, POCUS has emerged as a non-invasive additional modality to reveal possible contributing causes to dysphagia in the oral and pharyngeal phases. A recent POCUS study has found that the variability and concurrency of swallowing movement sequencing might be the key factors that indicate age-related changes and impairments in swallowing physiology (Lam et al., 2024). The diagnostic value of these factors, however, remains unclear. Meanwhile, the sophistication and application of (X)AI might provide further insights in this regard.

Purpose: The present study aims to explore (i) the dysphagia detection accuracy of movement sequencing features acquired through POCUS, and (ii) the contribution of these factors in swallowing physiology classification using off-the-shelf machine learning algorithms and XAI techniques.

Methods: Ninety-five individuals, including 42 healthy younger adults, 37 healthy older adults, and 16 dysphagic adults, were recruited in the community from July 2021 to October 2023. Each participant performed 3 repeated trials of dry swallow and prescribed/thin water boluses of various volumes. Swallowing movements were acquired using B-mode ultrasound with a curvilinear transducer placed at the mid-sagittal plane of the submental region. Sequence analysis (Lam et al., 2024) was used to extract the variability and concurrency of swallowing movements. The training and testing of classification algorithms were then conducted using stratified K-fold cross validation.

Results: K-nearest neighbour outperformed other algorithms with an Area Under the Receiver Operating Characteristic Curve of .957 in differentiating dysphagic individuals from other groups. XAI measures also echoed with previous findings in Lam et al (2024).

Conclusion: The application of XAI to movement sequencing features acquired from POCUS is a promising technique for informing clinical decisions in dysphagia management.

Lam, W. Y. S., Kwong, E., Chan, H. W. T., & Zheng, Y. P. (2024). Using Sequence Analyses to Quantitatively Measure Oropharyngeal Swallowing Temporality in Point-of-Care Ultrasound Examinations: A Pilot Study. Journal of Clinical Medicine, 13(8), 2288.