Oral Cancer Discrimination and Novel Oral Epithelial Dysplasia Stratification Using FTIR Imaging and Machine...

Area of Science:

• Biomedical Optics
• Spectroscopy
• Computational Pathology

Background:

• Oral squamous cell carcinoma (OSCC) is a significant global health concern.
• Early detection of oral precancerous lesions, such as oral hyperkeratosis (HK) and oral epithelial dysplasia (OED), is crucial for improving patient outcomes.
• Current diagnostic methods can be subjective and require invasive procedures.

Purpose of the Study:

• To evaluate the efficacy of Fourier transform infrared (FTIR) imaging combined with machine learning for the detection and classification of oral lesions.
• To differentiate between HK, OED, and OSCC using spectral data.
• To develop a potential risk stratification strategy for precancerous OED samples.

Main Methods:

• FTIR imaging was performed in a transmission model on biopsy samples of HK, OED, and OSCC.
• Spectral data preprocessing and cluster analysis were conducted to generate representative spectra.
• Machine learning models, including Partial Least Squares Discriminant Analysis (PLSDA), Support Vector Machines Discriminant Analysis (SVMDA), and Extreme Gradient Boosting Discriminant Analysis (XGBDA), were trained and tested.

Main Results:

• Exploratory analyses showed good spectral separation between HK and OSCC, with OED spectra overlapping with either HK or OSCC.
• The PLSDA model achieved 100% sensitivity and 100% specificity in differentiating HK and OSCC.
• The PLSDA model classified OED samples into HK-grade, OSCC-grade, or borderline categories, indicating potential risk stratification.

Conclusions:

• FTIR imaging coupled with machine learning provides a powerful tool for objective and accurate early detection of oral cancer.
• The developed PLSDA model demonstrates high performance in classifying oral lesions and stratifying risk for precancerous conditions.
• This technique holds promise for non-invasive or minimally invasive early diagnosis of oral cancer, potentially improving patient prognosis.