Development of a machine learning-based tension measurement method in robotic surgery.

Area of Science:

• Robotics in Surgery
• Machine Learning Applications
• Biomechanical Engineering

Background:

• Anastomotic leaks complicate up to 10% of over 300,000 annual colorectal surgeries in the U.S.
• Current intraoperative tension assessment relies on subjective metrics, increasing complication risks.
• Objective measurement of colonic mechanical tension is needed to mitigate surgical risks.

Purpose of the Study:

• To assess the feasibility of a novel objective method for measuring mechanical tension in ex vivo porcine colons.
• To develop and validate a machine learning algorithm for estimating tissue tension.
• To provide a quantitative metric for surgical tension to prevent complications.

Main Methods:

• Utilized the da Vinci Research Kit (dVRK) for robotic manipulation.
• Developed a long short-term memory (LSTM) neural network algorithm to estimate pulling forces.
• Applied upward forces to ex vivo porcine colon segments and compared algorithm-estimated forces with ground-truth measurements from a force sensor.

Main Results:

• The machine learning algorithm achieved force estimation accuracy up to 88%, with an average of 74%.
• Estimated and measured forces demonstrated a very strong correlation (Spearman's Correlation ≥ 0.80).
• The system successfully measured forces ranging from 0 to 17.2 N in short experimental durations.

Conclusions:

• A machine learning algorithm was developed to objectively estimate colonic tension, closely approximating ground-truth sensor data.
• This represents the first study to objectively measure and report tissue tension in Newtons using robotic assistance.
• The proposed method is adaptable for various tissue types, potentially reducing surgical complications and mortality.