Hyperspectral imaging for tissue differentiation using neural network in parotid surgery

Introduction In parotidectomy, continuous tissue differentiation and identification of risk structures, such as the facial nerve, is essential for complication-free surgery. Thus, there is a need for a comprehensive image-based method of tissue differentiation and identification. Hyperspectral imaging (HSI) is a potential method for this purpose, as it can objectively analyze and compare spectral characteristics of individual areas.

Methods So far, nine patients who underwent parotidectomy in our ENT clinic have been included in the evaluation. After successful dissection and identification of all anatomically relevant structures, an HSI image was taken and annotated on site. A distinction was made between up to ten tissue types. These data were used to train a neural network (CNN) for tissue identification. The acquired data was divided into a training and a validation set.

Results In the current results, six tissue types were examined. The sensitivity of tissue detection for these tissue types was 92% for muscle, 95% for vein, 98% for artery, 91% for nerve, 94% for parotid, and 96% for subcutaneous tissue when the data was split into 75% training and 25% validation data. This resulted in an overall target accuracy of over 90%.

Conclusion  This makes HSI analysis an option for tissue identification in the context of a parotidectomy and can accelerate the preparation here, while maintaining patient safety. However, more clinical data is needed to improve the algorithms. Thus, a reduction of tissue damage in the course of a parotidectomy could be achieved.

Publication History

Article published online:
12 May 2023