SP 8.05 Machine Learning-Assisted Diffuse Reflectance Spectroscopy for Intra-Operative Identification of Parathyroid Glands In Vivo
Nikita Rani Chander, Mufaddal Moonim, Ara Darzi, Christopher Peters, Daniel Elson, Aimee Di MarcoAbstract
Aims
Reliable intra-operative identification of parathyroid glands is critical for optimal surgical outcomes, but challenging due to their small size, variable appearance, and proximity to visually similar tissues. Diffuse reflectance spectroscopy (DRS) is an optical modality that characterises tissue based on structural and biochemical properties. Previous ex vivo studies have shown high accuracy of DRS for differentiating parathyroid tissue, with preliminary in vivo data demonstrating its feasibility for intra-operative use. This study expands on prior work to determine the in vivo performance of DRS for parathyroid identification.
Methods
A prospective in vivo study was conducted in patients undergoing neck endocrine surgery. Spectral measurements were acquired intra-operatively using a sterilised handheld DRS probe applied directly to exposed tissues. Spectra were analysed with a supervised machine-learning framework using the extreme gradient boosting classifier. Model performance was assessed using stratified five-fold cross-validation.
Results
Of 100 patients enrolled, 75(75%) were female, with a median age of 53(18-79) years. A total of 4,620 spectra were acquired from 95 normal parathyroids, 41 pathological parathyroids, 91 fat samples, 30 thyroid samples, and 16 lymph nodes. Classifier accuracy for identification of parathyroid glands compared to other tissues is summarised below. Normal ParathyroidPathological ParathyroidFat95.5%96.9%Thyroid96.6%91.3%Lymph Node99.4%97.8%
Conclusions
This study demonstrates that DRS can accurately identify parathyroid tissue intra-operatively in a complex surgical environment. These findings support the potential role of DRS as a real-time adjunct to assist surgeons in parathyroid identification and preservation during neck endocrine surgery.