SERV-CT (SERV-CT: A disparity dataset from CT for validation of endoscopic 3D reconstruction)

手术场景的内窥镜立体重建会产生特定的问题，包括缺乏清晰的角落特征、高度镜面反射的表面特性以及血液和烟雾的存在。这些问题给立体重建本身和标准化数据集生产带来了困难。我们提出了一个基于锥形束 CT (SERV-CT) 的立体内窥镜重建验证数据集。在内窥镜视野内放置两具离体小猪全躯干尸体，在 CT 扫描中可以看到内窥镜和目标解剖结构。手动对齐内窥镜的后续方向以匹配立体视图，并计算基准差异、深度和遮挡。 CT 扫描的要求将每个离体样本的立体对数限制为 8 个。对于第二个样本，获取 RGB 表面以帮助对齐平滑、无特征的表面。重复的手动对齐显示了大约 2 个像素的 RMS 视差精度和大约 2 mm 的深度精度。提供了一个简化的参考数据集，由内窥镜图像对组成，具有相应的校准、差异、深度和遮挡，涵盖了大部分内窥镜图像和一系列组织类型，包括光滑的镜面，以及深度的显着变化。 SERV-CT 数据集为手术应用提供了易于使用的立体验证，具有平滑的参考差异和深度，覆盖了大部分内窥镜图像。

Stereoscopic reconstruction of endoscopic imaging in surgical scenarios presents specific challenges, including the lack of distinct corner features, highly specular surface properties, and the presence of blood and smoke. These challenges impose difficulties for both stereoscopic reconstruction itself and the production of standardized datasets. We present a validation dataset for stereoscopic endoscopy reconstruction based on Cone Beam CT (SERV-CT). Two ex vivo full-body pig cadavers were placed within the endoscopic field of view, with the endoscope and target anatomical structures visible in the CT scans. The subsequent orientation of the endoscope was manually aligned to match the stereoscopic views, and ground truth disparity, depth, and occlusion were calculated. CT scanning requirements limited the number of stereoscopic pairs per ex vivo specimen to 8. For the second specimen, RGB surface scans were acquired to aid in aligning smooth, featureless surfaces. Repeated manual alignments demonstrated an RMS disparity accuracy of approximately 2 pixels and a depth accuracy of roughly 2 mm. A simplified reference dataset is provided, consisting of pairs of endoscopic images with corresponding calibrations, disparities, depths, and occlusion information. It covers most endoscopic views and a range of tissue types, including smooth specular surfaces, as well as significant variations in depth. The SERV-CT dataset provides an easy-to-use stereoscopic validation resource for surgical applications, with smooth ground truth disparities and depths, covering most endoscopic imaging scenarios.

搜集汇总

背景与挑战

背景概述

SERV-CT是一个基于锥形束CT的内窥镜立体重建验证数据集，提供内窥镜图像对及其校准、视差、深度和遮挡信息，覆盖平滑镜面组织和深度变化，旨在解决手术场景中重建的特定问题。该数据集具有约2像素的视差精度和2毫米的深度精度，便于手术应用中的立体验证。