Exploring video recognition models for force estimation in small bowel surgical retractions.

Area of Science:

• Medical technology
• Surgical robotics
• Computer vision in medicine

Background:

• Minimally invasive surgery (MIS) requires precise control to avoid tissue damage.
• Current force feedback systems often rely on costly hardware, limiting adoption in conventional laparoscopy.
• Haptic deficiency in MIS can impact surgical outcomes and surgeon skill assessment.

Purpose of the Study:

• To develop a vision-based method for estimating tissue retraction force during MIS.
• To assess surgeon skill and provide haptic feedback without specialized instruments.
• To overcome limitations of current force-sensing technologies in conventional laparoscopy.

Main Methods:

• An experimental setup was created to collect a force-vision dataset using laparoscopic instruments and a bowel phantom.
• ResNet and transformer-based computer vision models were evaluated for force estimation accuracy.
• Data collection involved novel procedures and force-signal preprocessing techniques.

Main Results:

• Vision-based models demonstrated generalization across varying phantom geometries and camera angles.
• ResNet-based models outperformed transformer models in force estimation accuracy.
• All evaluated models achieved real-time force estimation capabilities, with 3D ResNet being the fastest.

Conclusions:

• Vision-based force estimation from laparoscopic video is a viable and promising approach.
• The proposed method offers a low-cost, easily integrated solution for haptic feedback and skill assessment.
• This technology has the potential to enhance surgical training and patient safety in MIS.