Deep-Learning Virtual Superior Mesenteric Artery Modeling for Risk Stratification in Pancreas Surgery

Area of Science:

• Medical Imaging
• Surgical Anatomy
• Artificial Intelligence in Medicine

Background:

• Patient-specific anatomy of the superior mesenteric artery (SMA) is crucial for pancreatic surgery.
• Deep learning models offer virtual 3D visualization of SMA anatomy and branching patterns.
• Correlating SMA features with patient outcomes is essential for surgical success.

Purpose of the Study:

• To assess deep-learning-based virtual SMA modeling for 3D visualization.
• To correlate SMA anatomical features with intraoperative and postoperative outcomes.
• To enhance preoperative planning and intraoperative navigation in pancreatic surgery.

Main Methods:

• Analyzed CT scans of 124 patients undergoing pancreatic resection.
• Utilized deep learning software for SMA course, branching, caliber, and aortic angle analysis.
• Segmented SMA into SMA1, SMA2, and SMA3 regions relative to the pancreas.
• Employed logistic and linear regression to link anatomical measurements with perioperative outcomes.

Main Results:

• Observed differences in SMA measurements between patient populations.
• Mean calibers: SMA1 (7.05 mm), SMA2 (6.20 mm), SMA3 (5.69 mm).
• Mean branches: SMA2 (2.21), SMA3 (4.52).
• Fewer SMA2 branches associated with postoperative pancreatic fistula (POPF) and severe complications (Clavien-Dindo grade ≥ III).
• Increased distance between pancreas and SMA linked to POPF in minimally invasive cases.

Conclusions:

• Deep learning-based 3D SMA reconstruction accurately assesses pancreatic-SMA relationships.
• Identified specific SMA anatomical features predictive of surgical outcomes.
• SMA modeling improves preoperative planning, intraoperative navigation, and outcome prognostication.