Heidelberg colorectal data set for surgical data science in the sensor operating room

Area of Science:

• Computer Vision
• Medical Imaging
• Surgical Data Science

Background:

• Image-based tracking of surgical instruments is crucial for surgical data science.
• Existing methods for instrument detection, segmentation, and tracking often fail on challenging images and lack generalization.
• There is a need for robust datasets to benchmark these algorithms.

Purpose of the Study:

• To introduce the Heidelberg Colorectal (HeiCo) dataset, the first publicly available resource for benchmarking medical instrument detection and segmentation algorithms.
• To emphasize the robustness and generalization capabilities of these algorithms.
• To facilitate the development of more reliable surgical instrument tracking systems.

Main Methods:

• The Heidelberg Colorectal (HeiCo) dataset comprises 30 laparoscopic videos from three types of surgery.
• It includes sensor data from medical devices and detailed annotations.
• Annotations cover surgical phase labels and instrument instance-wise segmentation masks for over 10,000 frames.

Main Results:

• The HeiCo dataset enables comprehensive benchmarking of medical instrument detection and segmentation algorithms.
• It specifically focuses on evaluating method robustness and generalization.
• The dataset has been successfully utilized in international competitions (Endoscopic Vision Challenges 2017 and 2019).

Conclusions:

• The HeiCo dataset is a valuable resource for advancing the field of surgical instrument tracking.
• It addresses the limitations of previous datasets by focusing on challenging scenarios and generalization.
• This dataset will drive the development of more accurate and reliable computer vision tools for surgery.