Development of a flexible endoscopic system using leached image guides for visible–near-infrared hyperspectral imaging

In this study, we developed a flexible endoscopic system using leached image guides (hereafter, the LIG-based endoscope) for visible–near-infrared (VIS–NIR) hyperspectral imaging (HSI). Its performance was evaluated by comparison with a conventional flexible endoscope using a silica-based bundle fiber (hereafter, the BF-based endoscope). The developed system uses a leached-image-guide fiber structure with 2,508 fibers and improved structural uniformity. A multilayer objective lens was also introduced to suppress chromatic aberration from the visible to the near-infrared region. As a result, wavelength-dependent focal adjustment using a liquid tunable lens and affine transformation for magnification correction, both of which were required in the conventional system, became unnecessary. This simplified the data acquisition and preprocessing workflow. Functional evaluations showed a marked improvement in fiber structural uniformity, with the standard deviation reduced from 32 to 4. Spatial resolution increased by approximately 32%, and exposure time in the near-infrared range (900–1600 nm) was shortened. In addition, the allowable bending radius improved from 25 mm to less than 10 mm, resulting in higher flexibility. In the six-class resin identification experiment, the developed system achieved an average classification accuracy of 93.3% in the wavelength range of 1000–1600 nm. This result indicates that spectral discrimination performance was maintained despite the redesign of the endoscope. In addition, pixel-wise classification based on mouse-organ hyperspectral data showed accuracy and recall values around or above 80% for major organs under many conditions. These results suggest that the system has spectral discrimination capability not only for resin materials but also for biological tissues. Overall, the developed LIG-based endoscope provides a practical device platform for industrial material discrimination and shows initial feasibility for biological tissue imaging. The results support its potential for both medical endoscopy and industrial inspection in confined or complex environments.