Multi-class classification algorithm for optical diagnosis of oral cancer

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Spectroscopy

Background:

• Oral cavity cancer diagnosis often relies on invasive biopsies.
• Accurate and early detection of oral squamous cell carcinoma (OSCC) is crucial for patient outcomes.
• Non-invasive diagnostic tools are needed to improve screening efficiency.

Purpose of the Study:

• To develop a direct multi-class spectroscopic diagnostic algorithm for oral tissue discrimination.
• To differentiate high-grade cancerous, low-grade cancerous, precancerous (leukoplakia), and normal squamous tissue sites.
• To apply the theory of total principal component regression (TPCR) for enhanced diagnostic accuracy.

Main Methods:

• Utilized in vivo autofluorescence spectral data from oral cavity patients.
• Developed and validated a diagnostic algorithm based on total principal component regression (TPCR).
• Employed leave-one-out cross-validation for the training dataset and an independent dataset for validation.

Main Results:

• The TPCR-based algorithm achieved high classification accuracy for four tissue types.
• Training data classification accuracies: 94% (high-grade SCC), 100% (low-grade SCC), 100% (leukoplakia), 91% (normal).
• Independent validation accuracies: 90% (high-grade SCC), 90% (low-grade SCC), 85% (leukoplakia), 88% (normal).

Conclusions:

• The developed spectroscopic algorithm demonstrates satisfactory performance in classifying oral tissue types.
• TPCR-based spectral analysis offers a promising non-invasive approach for oral cancer diagnostics.
• This method can aid in distinguishing between various grades of oral lesions and normal tissue.