Comparison of clinical nasal endoscopy, optical biopsy, and artificial intelligence in early diagnosis and treatment planning in laryngeal cancer: a prospective observational study.

Background: Laryngeal cancer accounts for approximately 2% of all cancers globally and is considered one of the most aggressive types of head and neck cancer. Prompt diagnosis is crucial to improving survival and function. Direct laryngoscopy and imaging modalities are conventional diagnostic methods. However, laryngeal cancer diagnosis can be delayed, and early subtle mucosal changes can be missed. Flexible nasal endoscopy, particularly when integrated with artificial intelligence and optical biopsy methods, has shown promise in the early detection of laryngeal cancer. Yet, there is little literature on the combined experiences of these modalities. Methods: This prospective observational study involved 142 patients with suspected laryngeal cancer. All included patients underwent flexible nasal endoscopy with topical anesthesia. The patients were assessed using one or more optical biopsy techniques (Narrow Band Imaging [NBI], SPIES, or ISCAN), depending on available equipment and whether the lesions were visible. AI algorithms were retrospectively applied to endoscopic images to categorize lesions as cancerous or non-cancerous depending on vascular, textural, and color characteristics. The AI model was trained on a different pre-annotated dataset, and the images from the study cohort were not used to train the AI model - this methodologically ensures no bias has been introduced into the evaluation. Histopathology was used as the reference standard. Diagnostic performance was calculated using sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: The study revealed superior sensitivity (95.2%) and specificity (96.5%) with AI-enhanced endoscopy compared to conventional endoscopy (89.6%, 92.4%), respectively. Optical biopsy methods provided better visualization of lesions; however, not all patients had all three modalities in a single procedure. Diagnostic delay was shortened with a median time of 15 to 7 days (<0.001). Inter-rater agreement was strong overall (κ=0.84), with hoarseness having the most reliability, most likely due to better exposure of the glottis. Conclusions: AI and selectively applied optical biopsy methods improved diagnostic accuracy in nasal endoscopy and reduced time delays for the early detection and management of laryngeal cancer. Further study in multicenters will allow for further validation of this work.