Objectives: This study introduces a novel classification approach that combines convolutional neural network (CNN) and Raman mapping to differentiate between tongue squamous cell carcinoma (TSCC) and non-tumorous tissue, as well as to identify different histological grades of TSCC.
Materials and methods: In this study, 240 Raman mappings data from 30 tissue samples were collected from 15 patients who had undergone surgical resection for TSCC. A total of 18,000 sub-mappings extracted from Raman mappings were then used to train and test a CNN model, which extracted feature representations that were subsequently processed through a fully connected network to perform classification tasks based on the Raman mapping data.
Results: The experimental results indicated that the proposed method achieved competitive classification accuracy above 83%. To further validate the effectiveness of the Raman mapping, its performance was compared with Raman spectroscopy, demonstrating a competitive accuracy rate.
Conclusions: The promising outcomes from this application of CNN in Raman mapping suggest that this technique could be a reliable method for intraoperative assessment of surgical margins, potentially leading to shorter detection times.
Keywords: Raman mapping; convolutional neural network; intraoperative assessment; surgical margin; tongue squamous cell carcinoma.
© 2024 John Wiley & Sons Ltd.