Towards automated spectroscopic tissue classification in thyroid and parathyroid surgery

Background: In (para-)thyroid surgery iatrogenic parathyroid injury should be prevented. To aid the surgeons’ eye, a camera system enabling parathyroid-specific image enhancement would be useful. Hyperspectral camera technology might work, provided that the spectral signature of parathyroid tissue offers enough specific features to be reliably and automatically distinguished from surrounding tissues. As a first step to investigate this, we examined the feasibility of wide band diffuse reflectance spectroscopy (DRS) for automated spectroscopic tissue classification, using silicon (Si) and indium-gallium-arsenide (InGaAs) sensors. Methods: DRS (350–1830 nm) was performed during (para-)thyroid resections. From the acquired spectra 36 features at predefined wavelengths were extracted. The best features for classification of parathyroid from adipose or thyroid were assessed by binary logistic regression for Si- and InGaAs-sensor ranges. Classification performance was evaluated by leave-one-out cross-validation. Results: In 19 patients 299 spectra were recorded (62 tissue sites: thyroid = 23, parathyroid = 21, adipose = 18). Classification accuracy of parathyroid–adipose was, respectively, 79% (Si), 82% (InGaAs) and 97% (Si/InGaAs combined). Parathyroid–thyroid classification accuracies were 80% (Si), 75% (InGaAs), 82% (Si/InGaAs combined). Conclusions: Si and InGaAs sensors are fairly accurate for automated spectroscopic classification of parathyroid, adipose and thyroid tissues. Combination of both sensor technologies improves accuracy. Follow-up research, aimed towards hyperspectral imaging seems justified.