Gender, Smoking History, and Age Prediction from Laryngeal Images

Area of Science:

• Otolaryngology
• Artificial Intelligence
• Medical Imaging

Background:

• Flexible laryngoscopy is crucial for diagnosing laryngeal diseases and identifying potentially malignant lesions.
• Machine learning (ML) models show promise for automated diagnosis using laryngeal images.
• Incorporating patient demographic data can enhance ML diagnostic performance but manual entry is time-consuming.

Purpose of the Study:

• To develop deep learning models for predicting patient demographic information directly from laryngoscopic images.
• To improve the performance of automated laryngeal disease detection models by integrating predicted demographic data.
• To establish a benchmark for deep learning models on a novel laryngoscopic image dataset.

Main Methods:

• Development and application of deep learning models (CNNs and Transformers) to predict gender, smoking history, and age from laryngoscopic images.
• Creation of a new, comprehensive laryngoscopic image dataset for ML analysis.
• Benchmarking the performance of eight classical deep learning architectures.

Main Results:

• Achieved overall prediction accuracies of 85.5% for gender, 65.2% for smoking history, and 75.9% for age.
• Demonstrated the feasibility of using deep learning to infer demographic information from laryngeal images.
• Identified and benchmarked the performance of various deep learning models on the new dataset.

Conclusions:

• Predicting patient demographics from laryngoscopic images using deep learning is feasible and can enhance diagnostic models.
• The developed models and dataset offer a pathway to improve automated laryngeal disease detection by incorporating patient data.
• This approach offers a more efficient alternative to manual data entry for improving AI-driven diagnostic tools in otolaryngology.