PhaKIR2024

在内窥镜视频中准确识别和定位手术器械是计算机和机器人辅助微创手术的重要基础。处理现实世界条件下的挑战，如变化的照明、器械和相机的模糊运动、出血和烟雾，是该领域的重要先决条件。PhaKIR挑战赛包括三个任务：1) 手术器械的实例分割；2) 关键点估计；3) 手术阶段识别。数据集包含13个真实的胆囊切除术视频，由经验丰富的医生在三家医院录制。每个视频都提供了详细的注释，包括手术器械的实例分割、关键点坐标和手术阶段。Accurate identification and localization of surgical instruments in endoscopic videos serves as a critical foundation for computer- and robot-assisted minimally invasive surgery. Addressing challenges in real-world scenarios such as varying illumination, blurry motion of instruments and cameras, bleeding, and smoke constitutes a key prerequisite for this research field. The PhaKIR Challenge includes three tasks: 1) Instance segmentation of surgical instruments; 2) Keypoint estimation; 3) Surgical phase recognition. The dataset contains 13 real-world cholecystectomy videos recorded by experienced surgeons at three hospitals. Each video is accompanied by detailed annotations, including instance segmentation masks for surgical instruments, keypoint coordinates, and surgical phase labels.

搜集汇总

背景与挑战

背景概述

PhaKIR2024是一个专注于腹腔镜手术的视频数据集，包含真实人类胆囊切除术视频，时长23-60分钟，录制于三个医院。其特点在于提供了像素级手术器械实例分割掩码（19类）、器械关键点坐标（每秒一帧采样）和手术阶段标注（8个阶段），全面覆盖器械定位和手术上下文，支持时间信息分析，适用于非商业研究用途。