(003) Artificial Intelligence in Diagnostic Imaging in Otorhinolaryngology: A Systematic Review of the Current Evidence

Introduction: To evaluate the application of artificial intelligence (AI) in imaging diagnosis in otorhinolaryngology, highlighting its potential to improve diagnostic accuracy and support clinical practice, as well as to analyze limitations and perspectives for clinical adoption.

Methods: A systematic review was conducted using the PubMed, Scopus, and Web of Science databases, including studies published between 2022 and 2025 in English and Portuguese that evaluated the use of AI in imaging diagnosis in otorhinolaryngology. The search strategy included the descriptors and keywords “artificial intelligence,” “machine learning,” “medical imaging,” “otolaryngology,” “laryngeal lesions,” and “head and neck surgery,” as well as corresponding MeSH terms. Diagnostic studies reporting accuracy, sensitivity, and specificity were included. Study screening and methodological assessment were independently performed by two reviewers.

Results: Eight studies were included. AI models demonstrated high diagnostic performance, with mean sensitivity of 91%, specificity of up to 94%, and accuracy ranging from 80.6% to 99.7%. In laryngeal cancer, deep learning algorithms, particularly YOLO-based models, correctly localized and classified vocal fold carcinomas in 71% to 78% of validation and test cases. Ensemble methods achieved accuracy rates of 78% for carcinoma and 70% for benign lesions. Detection speed of up to 63 frames per second supports feasibility for real-time laryngoscopy. Studies also reported increasing use of AI in early neoplasm detection, nasal polyp characterization, differentiation of benign and malignant lesions, and analysis of CT, MRI, and endoscopic images.

Conclusions: Artificial intelligence shows strong potential as a complementary tool in imaging diagnosis in otorhinolaryngology, contributing to improved diagnostic accuracy and reduced interobserver variability. Nevertheless, more robust studies, methodological standardization, and external validation are necessary before large-scale clinical implementation.