Real-Time Anatomy Recognition During Single-Port Retroperitoneal Surgery: A Prospective Comparison Between AI and...

Area of Science:

• Robotic Surgery
• Artificial Intelligence in Medicine
• Surgical Anatomy

Background:

• Real-time anatomical recognition in robot-assisted surgery can enhance intraoperative decision-making.
• The study presents the first clinical test of an AI algorithm for identifying key anatomical structures during robot-assisted retroperitoneal renal and adrenal procedures.

Purpose of the Study:

• To clinically evaluate an AI algorithm for real-time anatomical recognition during robot-assisted retroperitoneal procedures.
• To compare the AI's recognition timing and patterns against the operating surgeon.
• To assess the impact of patient BMI and prior abdominal surgery on AI recognition time.

Main Methods:

• A deep learning algorithm was trained and retrained to identify six anatomical structures: psoas muscle, ureter, kidney, renal artery, renal vein, and inferior vena cava (IVC).
• The system was prospectively evaluated in 15 patients, comparing AI recognition timing (Δ-time) with the surgeon's.
• Recognition patterns and the influence of patient BMI and surgical history were analyzed.

Main Results:

• Surgeons were significantly faster in recognizing most structures, frequently being the first to identify the psoas muscle, renal artery, and kidney.
• The AI system achieved simultaneous recognition in some cases but was rarely the first to identify structures.
• Higher BMI and previous abdominal surgery were associated with longer AI recognition times for the renal artery.

Conclusions:

• The AI algorithm can recognize key anatomical structures in real time during robot-assisted surgery.
• The surgeon consistently outperformed the AI in recognition timing.
• The AI shows promise as a supportive tool for intraoperative guidance and situational awareness, with future improvements needed for autonomous recognition.