Advances in Real-Time 3D Reconstruction for Medical Endoscopy.

Area of Science:

• Computer Vision
• Medical Imaging
• Surgical Technology

Background:

• Recent advancements in computational power and computer vision have spurred progress in fields like autonomous driving and robotics.
• These advancements offer potential for adaptation to medical endoscopy, addressing challenges such as featureless surfaces, variable lighting, and deformable anatomy.

Purpose of the Study:

• To provide a comprehensive overview of current real-time 3D reconstruction methods for medical endoscopy.
• To categorize and compare existing techniques, including monocular, binocular, trinocular, and multiocular approaches, as well as active and passive methods.
• To discuss relevant error metrics, hardware considerations (GPU vs. FPGA), and best practices for dataset utilization in evaluating 3D reconstruction performance.

Main Methods:

• Categorization of 3D reconstruction methods based on the number of cameras (monocular to multiocular) and sensing approach (active vs. passive).
• Inclusion of studies on both flexible and non-flexible endoscopes to ensure comprehensive coverage.
• Consideration of publications evaluated on diverse datasets, including medical datasets, KITTI, and Middlebury, to incorporate relevant related methods.

Main Results:

• The review systematically categorizes various real-time 3D reconstruction techniques applicable to medical endoscopy.
• It highlights the adaptability of computer vision techniques to endoscopic challenges and discusses performance evaluation metrics.
• The comparison extends to hardware choices, such as Graphics Processing Units (GPUs) versus Field-Programmable Gate Arrays (FPGAs), for camera-based 3D reconstruction.

Conclusions:

• Real-time 3D reconstruction in medical endoscopy benefits significantly from advancements in computer vision and computational power.
• A structured approach to categorizing methods and evaluating them using appropriate datasets and metrics is crucial for progress.
• Further research adapting techniques from related fields, like autonomous driving, holds promise for enhancing endoscopic visualization and surgical navigation.