SP8.05 Diffuse reflectance spectroscopy for diagnosing malignancy in thyroid nodules

Abstract Aims Thyroid nodules are common; although 80-90% are benign, accurately differentiating benign from malignant nodules can be challenging. Diagnostic surgery is often required for indeterminate nodules; this may later be judged as under- or over-treatment depending on histological result. Diffuse reflectance spectroscopy (DRS) is an optical technique capable of assessing tissue morphology, functionality and biochemical composition, with studies showing its utility in differentiating cancerous from non-cancerous tissue across multiple organs. This study aims to assess if DRS can differentiate benign from malignant thyroid nodules. Methods This is a prospective study whereby patients undergoing thyroid surgery were prospectively recruited. Freshly resected ex-vivo specimens were analysed with a handheld DRS probe; spectral data was validated against histopathology. Binary classification was performed with the extreme gradient boosting supervised machine learning classifier. Data were divided into training and testing sets with a stratified 5-fold cross-validation method, and the classifier evaluated. Results Of 54 patients, 37 (69%) were female; median age 51.5 (25-80) years. A total of 6964 spectra were collected across 28 benign and 28 malignant nodules. In nodules pre-operatively confirmed as benign or malignant, the classifier achieved a diagnostic accuracy of 88.3%. In pre-operatively indeterminate nodules, accuracy was preserved at 88.4%. Conclusions This study provides ex-vivo validation of the DRS system for differentiation of benign and malignant thyroid nodules, with an accuracy exceeding that of cytology (for which the PPV is 17-68%). This represents a significant step toward the development of a tool to improve the diagnosis and management of indeterminate thyroid nodules.