Deep convolutional neural networks for tongue squamous cell carcinoma classification using Raman spectroscopy

Area of Science:

• Oncology
• Biomedical Engineering
• Spectroscopy

Background:

• Accurate intra-operative assessment of tumor margins is critical for effective tongue squamous cell carcinoma (TSCC) treatment.
• Traditional histopathological analysis can be time-consuming and may not always provide real-time feedback during surgery.

Purpose of the Study:

• To develop and validate a novel classification method for discriminating between TSCC and non-tumorous tissue using deep convolutional neural networks (ConvNets) and fiber optic Raman spectroscopy.
• To assess the diagnostic performance of the proposed method in terms of sensitivity and specificity.

Main Methods:

• Collected Raman spectral data from 24 tissue samples (12 patients) of TSCC and adjacent non-tumorous tissue.
• Employed a deep ConvNets architecture with 6 blocks (each containing a convolutional and max-pooling layer) to extract nonlinear features from Raman spectra.
• Utilized a fully-connected network for the final classification task.

Main Results:

• The proposed ConvNets-based method achieved high sensitivity (99.31%) and specificity (94.44%) in classifying TSCC.
• Demonstrated competitive classification accuracy compared to existing state-of-the-art methods.
• The model effectively extracted nonlinear feature representations from Raman spectra for accurate discrimination.

Conclusions:

• The developed deep ConvNets model offers a promising, accurate, and rapid method for intra-operative evaluation of TSCC resection margins.
• This technique has the potential to improve surgical outcomes and enhance patient survival rates by ensuring complete tumor removal.