Indocyanine green quantification in full robotic esophagectomy using an unsupervised learning approach

Area of Science:

• Surgical Oncology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Intraoperative indocyanine green (ICG) imaging assesses gastric conduit perfusion during esophagectomy.
• Delayed perfusion identified by ICG correlates with anastomotic leakage.
• Current ICG assessments exhibit high interuser variability, necessitating quantitative analysis.

Purpose of the Study:

• To present a quantitative indocyanine green analysis using an unsupervised, self-organizing map (SOM) cluster network.
• To evaluate the utility of this quantitative approach in robotic-assisted minimally invasive esophagectomy.

Main Methods:

• Included 70 patients undergoing robotic-assisted minimally invasive esophagectomy with ICG imaging.
• Recorded videos of ICG imaging were clustered using an unsupervised SOM network.
• An ICG perfusion score was determined and correlated with anastomotic leakage.

Main Results:

• The ICG perfusion score, patient age, and body mass index correlated with increased anastomotic leakage risk (univariate analysis).
• Comorbidities and arteriosclerosis severity did not significantly differ between patients with and without leakage.
• The developed quantitative ICG analysis provided a predictive score for anastomotic leakage.

Conclusions:

• Unsupervised learning for quantitative ICG analysis may aid in predicting anastomotic leakage after robotic esophagectomy.
• Further validation in prospective, randomized controlled studies is required to clarify the clinical value.
• This quantitative approach holds potential for improving surgical decision-making and patient outcomes.