Laparoscopic hyperspectral imaging for in vivo detection of the vagal nerve in upper gastrointestinal surgery

TY - JOUR

T1 - Laparoscopic hyperspectral imaging for in vivo detection of the vagal nerve in upper gastrointestinal surgery

AU - Köhler, Hannes

AU - Ilgen, Annalena

AU - Pfahl, Annekatrin

AU - Stelzner, Sigmar

AU - Mehdorn, Matthias

AU - Jansen-Winkeln, Boris

AU - Melzer, Andreas

AU - Gockel, Ines

AU - Moulla, Yusef

N1 - © The Author(s) 2025

PY - 2025/8/11

Y1 - 2025/8/11

N2 - Background Accurate intraoperative detection of nerves in critical anatomic regions during oncologic resection, such as the vagal nerve, is crucial. The vagal nerve regulates many internal organs and is at risk during surgeries, potentially leading to severe complications. While different types of intraoperative neuromonitoring methods exist for functional assessment, Hyperspectral Imaging (HSI) offers a non-invasive, real-time alternative for morphologic identification. This study is, to our knowledge, the first to assess laparoscopic HSI for imaging and delineating the vagal nerve during minimally invasive surgery. Methods This prospective cohort study examined the vagal nerves of 19 patients undergoing Ivor Lewis esophagectomy using a laparoscopic HSI system with a 30-degree optic during the thoracic part of the procedure. Measurements on the exposed vagal nerve collected spectral data from 500 to 995 nm at each pixel. Based on these reflectance spectra, four different state-of-the-art machine learning methods for binary and multi-class tissue differentiation were evaluated. The classifiers were validated using Leave-One-Patient-Out Cross-Validation (LOOCV) and k-fold CV. Results The spectra of the tissue classes azygos vein, pleura, lung, vagal nerve, and esophagus showed high similarity, with wide inter-patient variability. All tested machine learning classifiers showed similar accuracy in differentiating vagal nerve tissue. Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) outperformed Logistic Regression (LR) and Multilayer Perceptron (MLP), with LDA showing the highest F1 score (harmonic mean of precision and recall) for binary classification (0.85), and SVM excelling in multi-class classification (0.74). Reflectance spectra without further pre-processing provided the best results for tissue differentiation. Conclusion The first application of HSI to detect the vagal nerve during minimally invasive surgery has shown promising classification results for the five tissue classes considered and highlighted the technical challenges for clinical use. Further clinical research is needed to explore the full potential of HSI to improve the reliability of nerve classification during surgery.

AB - Background Accurate intraoperative detection of nerves in critical anatomic regions during oncologic resection, such as the vagal nerve, is crucial. The vagal nerve regulates many internal organs and is at risk during surgeries, potentially leading to severe complications. While different types of intraoperative neuromonitoring methods exist for functional assessment, Hyperspectral Imaging (HSI) offers a non-invasive, real-time alternative for morphologic identification. This study is, to our knowledge, the first to assess laparoscopic HSI for imaging and delineating the vagal nerve during minimally invasive surgery. Methods This prospective cohort study examined the vagal nerves of 19 patients undergoing Ivor Lewis esophagectomy using a laparoscopic HSI system with a 30-degree optic during the thoracic part of the procedure. Measurements on the exposed vagal nerve collected spectral data from 500 to 995 nm at each pixel. Based on these reflectance spectra, four different state-of-the-art machine learning methods for binary and multi-class tissue differentiation were evaluated. The classifiers were validated using Leave-One-Patient-Out Cross-Validation (LOOCV) and k-fold CV. Results The spectra of the tissue classes azygos vein, pleura, lung, vagal nerve, and esophagus showed high similarity, with wide inter-patient variability. All tested machine learning classifiers showed similar accuracy in differentiating vagal nerve tissue. Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) outperformed Logistic Regression (LR) and Multilayer Perceptron (MLP), with LDA showing the highest F1 score (harmonic mean of precision and recall) for binary classification (0.85), and SVM excelling in multi-class classification (0.74). Reflectance spectra without further pre-processing provided the best results for tissue differentiation. Conclusion The first application of HSI to detect the vagal nerve during minimally invasive surgery has shown promising classification results for the five tissue classes considered and highlighted the technical challenges for clinical use. Further clinical research is needed to explore the full potential of HSI to improve the reliability of nerve classification during surgery.

KW - Hyperspectral imaging (HSI)

KW - Minimally invasive surgery (MIS)

KW - Vagal nerve

KW - Machine learning

U2 - 10.1007/s00464-025-12028-1

DO - 10.1007/s00464-025-12028-1

M3 - Article

C2 - 40789779

SN - 0930-2794

JO - Surgical Endoscopy

JF - Surgical Endoscopy

ER -