Hyperspectral Image Segmentation and Clustering for Intraoperative Diagnosis of Intestinal Ischemia

Abstract

We present an intraoperative hyperspectral imaging (HSI) approach for the assessment of intestinal ischemia, integrating deep learning-based segmentation with unsupervised spectral clustering. Sixteen patients were enrolled, of whom hyperspectral datasets from three representative cases were analyzed in detail. The developed convolutional neural network (1D-CNN) demonstrated high classification performance between intestinal tissue and background, enabling consistent delineation of clinically relevant regions. For ischemia assessment, oxygen saturation was estimated using a two-wavelength method, and principal component analysis combined with Gaussian mixture modeling (PCA-GMM) was applied to spectral features. The clustering revealed progressive separation of intact, moderately ischemic, and severely ischemic tissue, consistent with intraoperative surgical observations. This pipeline enables pixel-wise visualization of ischemic alterations with quantitative metrics, providing surgeons with objective and spatially resolved information to support resection decisions. The proposed framework demonstrates feasibility in clinical practice and establishes a foundation for large-scale validation and integration into intraoperative decision-support systems.