A deep convolutional neural network-based method for laryngeal squamous cell carcinoma diagnosis

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Laryngeal squamous cell carcinoma (LSCC) is a common respiratory tract tumor.
• Current LSCC diagnosis relies on laryngoscopy and pathology.
• Deep learning shows potential for accurate clinical diagnoses.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for LSCC diagnosis.
• To assess the CNN's performance on narrow-band imaging (NBI) endoscopy and pathological images.
• To improve the accuracy and speed of LSCC diagnosis.

Main Methods:

• Developed a CNN model trained on 4,591 laryngeal NBI scans and 3,458 pathology images.
• Images were divided into training, validation, and testing sets (70:15:15 ratio).
• Validated the model using independent test cohorts from multiple institutions.

Main Results:

• The CNN achieved high diagnostic accuracy, with Area Under the Curve (AUC) values up to 0.994 in the pathology group and 0.966 in the NBI group.
• Validation and testing datasets showed strong performance for both NBI and pathology image analysis.
• Independent testing confirmed the model's robustness across different institutions.

Conclusions:

• The CNN model demonstrated excellent performance in diagnosing LSCC from NBI and pathology images.
• AI-assisted diagnosis can lead to more accurate and rapid detection of LSCC.
• This technology offers a valuable tool for improving LSCC diagnostic workflows.