Discrimination of Vocal Folds Lesions by Multiclass Classification using Autofluorescence Spectroscopy

ObjectivesThe diagnosis of vocal fold cancer currently relies on invasive surgical biopsies, which can compromise laryngeal function. Distinguishing between different types of laryngeal lesions without invasive tissue sampling is therefore crucial. Autofluorescence spectroscopy (AFS) has proved to be efficient as a non-invasive detection technique but has yet to be fully exploited in the context of a multi-class tissue analysis. This study evaluates whether AFS can be used to discriminate between different types of laryngeal lesions in view of assisting in vocal fold surgery and preoperative investigations. Materials and methodsEx vivo spectral autofluorescence scans were recorded for each sample using a 405-nm laser excitation. A total of 1308 spectra were recorded from 29 vocal fold samples obtained from 23 patients. Multiclass analysis was conducted on the spectral data, classifying lesions either as normal, benign, dysplastic, or carcinoma. The results were compared to histopathological diagnosis. ResultsThrough an appropriate selection of spectral components and a cascading classification approach based on artificial neural networks (ANN), a classification rate of 97% was achieved for each lesion class, compared to 52% using autofluorescence intensity. ConclusionThe study demonstrates the effectiveness of AFS combined with multivariate analysis for accurate classification of vocal fold lesions. Comprehensive spectral data analysis significantly improves classification accuracy, even in challenging situations such as distinguishing between malignant and premalignant or benign lesions. This method could provide a way to perform in situ mapping of tissue states for minimally-invasive biopsy and surgical resection margins.