Confident texture-based laryngeal tissue classification for early stage diagnosis support

Area of Science:

• Medical Imaging
• Computational Pathology
• Oncology

Background:

• Early diagnosis of laryngeal squamous cell carcinoma (SCC) is crucial for patient outcomes.
• Computer-assisted diagnosis (CAD) for laryngeal SCC remains underexplored despite diagnostic challenges.

Purpose of the Study:

• To evaluate texture-based machine learning algorithms for classifying early-stage cancerous laryngeal tissue.
• To assess the reliability of classification using a confidence measure.

Main Methods:

• A dataset of 1320 patches from endoscopic videos of 33 SCC patients was created, representing four tissue classes.
• Texture-based machine learning algorithms were employed for tissue classification.
• A confidence measure was used to refine classification accuracy.

Main Results:

• The best performing feature achieved a median classification recall of 93%.
• Excluding low-confidence patches increased the median recall to 98%.
• The results demonstrate high reliability for the proposed computer-assisted diagnosis approach.

Conclusions:

• This study advances the state-of-the-art in computer-assisted laryngeal diagnosis.
• The findings support the development of endoscope-integrated systems for early SCC detection.
• The proposed method shows promise for improving diagnostic accuracy and patient care.