Precise Recognition With Enhanced Vision of Endocrine Neck Targets

Iatrogenic injuries to the parathyroid glands during thyroid surgery or to the recurrent laryngeal nerve (RLN) do still occur, requiring often specialized management. Recently, it has been demonstrated that the parathyroid gland shows a significant autofluorescence. Using a commercially available Near-InfraRed (NIR) camera (Fluobeam®, Fluoptics©, France), the parathyroid glands can be clearly visualized by contrast-free fluorescence imaging. However it lacks real-time quantification of the fluorescence intensity. The hyperspectral imaging (HSI), which is a technology that combines a spectrometer to a camera system, examines the optical properties of a large area in a wavelength range from NIR to visual light (VIS). It provides spatial information real time, in a contact-free, non-ionizing manner. The HSI technology would add the spatial information, thus enormously enhancing the intraoperative performance. The aim of the proposed study is to identify the spectral features of the important neck target structures, in particular the parathyroid glands, using an appropriate deep learning algorithm, to perform an automated parathyroid recognition. Additionally, this study proposes to compare the detection rate of the hyperspectral based parathyroid recognition with the already existing NIR autofluorescence based recognition.

The major challenge in thyroid and parathyroid procedures, is the safe identification of the recurrent laryngeal nerve (RLN) and the localization of the parathyroid glands (to be preserved or to be selectively removed). Iatrogenic injuries to the parathyroid glands during thyroid surgery (resulting in transient or permanent hypocalcemia) or to the RLN (resulting in hoarseness, dysphonia, dyspnea) do still occur, requiring often specialized management. The percentage of incidental parathyroidectomies, in specialized endocrine centers, is around 16%. In these cases, it is more likely to observe clinical relevant hypocalcemia than after planned parathyroidectomy for hyperparathyroidism. Therefore, there is a critical need for an intra-operative method enabling a precise, real-time parathyroid identification. Recently, it has been demonstrated that the parathyroid gland shows a significant autofluorescence, which is caused by the optical properties of a still unknown intrinsic fluorophore. When the gland is excited by a light source with a wavelength ranging from 750-785 nm, it emits a fluorescence peak around 820 nm. Taking advantage of this property, Falco et al., using a commercially available NIR camera (Fluobeam®, Fluoptics©, France), could clearly visualize the parathyroid glands by contrast-free fluorescence imaging and could easily discriminate them from the thyroid and the surrounding tissue. The drawback with this autofluorescence-based imaging is that it lacks real-time quantification of the fluorescence intensity. The hyperspectral imaging (HSI), which is a technology that combines a spectrometer to a camera system, examines the optical properties of a large area in a wavelength range from near infrared (NIR) to visual light (VIS). It provides diagnostic information about the tissue physiology, composition and perfusion. The fact that the HSI produces pictures, thus providing spatial information real time, in a contact-free, non-ionizing manner, makes it potentially a very valuable tool for the intraoperative use. HSI has exhibited its great potential in the medical field especially in the diagnosis of various neoplasia (e.g. of the cervix, breast, colon, brain), in the detection of perfusion pattern in patients with peripheral arterial disease and in the area of wound diagnostic. As previously shown, it is possible to discriminate the thyroid from the parathyroid glands according to the spectral characteristics, but the HSI technology would add the spatial information, thus enormously enhancing the intraoperative performance. In collaboration with the University of Leipzig, Germany, the investigators performed a clinical pilot trial on 8 patients, which showed promising results. Hyperspectral images during benign endocrine surgery procedures were able to demonstrate that thyroid and parathyroid have specific hyperspectral signatures. Furthermore, the parathyroid glands showed usually less oxygenated than the thyroid. A discrimination of the parathyroid glands based on these characteristics is proven to be possible. The aim of the proposed study is to identify the spectral features of the important neck target structures, in particular the parathyroid glands, using an appropriate deep learning algorithm, to perform an automated parathyroid recognition. Additionally, this study proposes to compare the detection rate of the hyperspectral based parathyroid recognition with the already existing NIR autofluorescence based recognition.