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

Intraoperative indocyanine green imaging has emerged as a powerful tool for assessing gastric conduit perfusion during open and minimally invasive esophagectomy. Although delayed perfusion correlates with the development of anastomotic leakage, indocyanine green assessments have high surgeon-dependent interuser variability. Therefore, quantitative indocyanine green analysis is recommended. We present a quantitative indocyanine green analysis using an unsupervised, self-organizing map cluster network during robotic-assisted minimally invasive esophagectomy.

In total, 70 patients treated with robotic-assisted minimally invasive esophagectomy, intraoperative indocyanine green imaging, and prophylactic endoluminal vacuum therapy were included in the study. The occurrence of anastomotic leakage, cycles of endoluminal vacuum therapy, patient comorbidities, and arteriosclerosis shown on preoperative computed tomography scans was recorded. The recorded videos of intraoperative indocyanine green imaging were clustered using an unsupervised, self-organizing map network, and an indocyanine green perfusion score was determined.

The indocyanine green perfusion score, as well as patient age and body mass index, correlated with an increased risk of anastomotic leakage in the univariate analysis. Other comorbidities and the extent of arteriosclerosis in preoperative computed tomography scans did not differ in patients with and without anastomotic leakage.

An unsupervised learning approach to quantify intraoperative indocyanine green imaging could aid the prediction of anastomotic leakage after robotic-assisted minimally invasive esophagectomy in future treatments. However, the value of this approach needs to be clarified in a randomized, controlled prospective study.