Stereo Endoscopic Dataset

This dataset was created based on a clinical procedure named Radical Prostatectomy with lymphadenectomy using da Vinci Xi surgical system (Intuitive Surgical Inc., USA) at European Institute of Oncology (IEO, Milan, Italy). 
The videos were recorded with a 3D HD video recorder (HVO-3300MT, SONY, Tokyo).

The dataset is divided into 6 folders concerning the steps of the procedure: 
clips
dissection
irrigation
suction
suturing
traction

Each step is then divided into four phases of the procedure:
1 - collapse of the peritoneum
2 - prostate removal
3 - lymphadenectomy 
4 - anastomosis

Each frame is labeled as follows: Camera_step_NumberOfTheFrame
for example, for the first frame, left lens of the Da Vinci endoscope, in the dissection step we have: L_dissection_000000.jpg

In the folders, we also provide the text files created on Anvil (ground truth labels), which contain important surgical context information, including the steps, phases and types of surgical instruments.

This dataset could be used for:

1. Evaluation of surgical scene reconstruction models

2. Training of unsupervised learning based networks

3. Surgical workflow recognition based on stereo images or point clouds. 

To know more about this research, please see the manuscript:

FRSR: Framework for Real-time Scene Reconstruction in Robot-Assisted Minimally Invasive Surgery.

本数据集基于意大利米兰欧洲肿瘤研究所（European Institute of Oncology, IEO）使用美国直觉外科公司（Intuitive Surgical Inc.）的达芬奇Xi手术系统（da Vinci Xi surgical system）开展的根治性前列腺切除术联合淋巴结清扫术（Radical Prostatectomy with lymphadenectomy）临床手术流程构建。

所有手术视频均采用索尼（SONY）3D高清录像机（HVO-3300MT，东京，日本）录制。

本数据集共划分出6个对应手术步骤的文件夹，分别为：手术片段（clips）、解剖分离（dissection）、冲洗（irrigation）、抽吸（suction）、缝合（suturing）、牵引（traction）。

每个步骤文件夹下均进一步拆分为四个手术阶段：
1. 腹膜塌陷（collapse of the peritoneum）
2. 前列腺切除（prostate removal）
3. 淋巴结清扫（lymphadenectomy）
4. 吻合术（anastomosis）

每帧图像均按照 Camera_step_NumberOfTheFrame 格式命名，例如达芬奇内窥镜左镜下解剖分离步骤的首帧图像，命名为 L_dissection_000000.jpg。

各文件夹中还附带了由Anvil软件生成的真实标注标签（ground truth labels）文本文件，其中包含手术流程步骤、阶段类型以及手术器械类型等关键手术上下文信息。

本数据集可应用于以下研究场景：
1. 手术场景重建模型的性能评估
2. 无监督学习网络的训练
3. 基于立体图像（stereo images）或点云（point clouds）的手术流程识别。

如需了解该研究的更多细节，请参阅相关论文：
《FRSR：机器人辅助微创手术实时场景重建框架（FRSR: Framework for Real-time Scene Reconstruction in Robot-Assisted Minimally Invasive Surgery）》