Artificial intelligence in dysphagia assessment: evaluating lingual muscle composition in head and neck cancer

Oropharyngeal dysphagia is a common and debilitating condition in head and neck cancer (HNC) patients. This study aimed to evaluate the relationship between tongue muscle composition (quantity and quality) and the risk of dysphagia in non-surgically treated HNC patients, using artificial intelligence (AI) analysis of pretreatment computed tomography (CT) scans.

A prospective analysis was conducted on 41 non-surgically treated HNC patients under-going curative radiotherapy. Tongue muscle quantity was measured as cross-sectional area (cm²) and as a percentage of body composition using AI-based segmentation of CT images. Muscle quality was assessed through Hounsfield Units (HU), representing muscle density. Dysphagia risk was evaluated with the validated EAT-10 questionnaire, considering scores ≥ 3 as indicative of increased risk.

A significant association was found between EAT-10 categorical scores and dysphagia risk (Chi² = 26.07, p < 0.0001). However, no significant correlation was observed between the percentage of tongue muscle and density (R = 0.081, p = 0.07). Patients with EAT-10 scores ≥ 3 had significantly larger percentages of tongue muscle area (mean 61.17 ± 10.44 cm²) compared to those with EAT-10 < 3 (mean 56.58 ± 5.77 cm²; p = 0.004). Additionally, higher tongue muscle density (HU) was associated with increased dysphagia risk (p = 0.046). A significant association was also observed between pre-treatment and post-treatment dysphagia, with patients who reported pre-treatment dysphagia (EAT-10 ≥ 3) continuing to experience higher post-treatment dysphagia (p = 0.009, R = 0.411). Biologically Effective Dose (BED) (p = 0.0042), advanced tumor stage (p = 0.004), and systemic treatment (p = 0.027) were further associated with increased post-treatment dysphagia risk.

The study demonstrates that non-surgically treated HNC patients with increased tongue area percentages and higher muscle density are at greater risk of dysphagia. Additionally, pre-treatment dysphagia was found to be a strong predictor of post-treatment dysphagia. The use of AI-based CT analysis provides a precise method for identifying patients at risk, allowing for timely interventions to improve swallowing function and quality of life.