Kidney edge detection in laparoscopic image data for computer-assisted surgery : Kidney edge detection

Area of Science:

• Medical Imaging
• Robotic Surgery
• Computational Anatomy

Background:

• Robotic-assisted kidney surgery benefits from augmented reality visualization.
• Accurate soft-tissue registration is crucial for visualizing subsurface structures.
• Current registration methods are computationally intensive and require large target surfaces.

Purpose of the Study:

• To develop an efficient deep learning method for kidney edge detection.
• To overcome limitations of traditional registration methods in robotic surgery.
• To provide a foundational step for enhanced surgical scene augmentation.

Main Methods:

• A deep learning network was designed to predict kidney edges from laparoscopic images.
• The network utilized depth information and distance fields for edge prediction.
• The method was evaluated on expert-labeled in vivo data from the EndoVis 2017 Kidney Boundary Detection challenge.

Main Results:

• The proposed method achieved a median precision-like of 39.5 px, recall-like of 143.3 px, and intersection over union (IOU) of 0.3.
• Leave-one-out cross-validation was employed to assess network performance.
• Performance was evaluated against the established state-of-the-art.

Conclusions:

• The deep learning approach effectively predicts kidney edges, enhancing surgical visualization.
• High occlusion in surgical scenes can lead to decreased intersection over union (IOU) scores.
• All associated code, data, and models are publicly available to facilitate further research.