Small intestinal viability assessment using dielectric relaxation spectroscopy and deep learning

Intestinal ischemia is a serious condition where the surgeon often has to make important but difficult decisions regarding resections and resection margins. Previous studies have shown that 3 h (hours) of warm full ischemia of the small bowel followed by reperfusion appears to be the upper limit for viability in the porcine mesenteric ischemia model. However, the critical transition between 3 to 4 h of ischemic injury can be nearly impossible to distinguish intraoperatively based on standard clinical methods. In this study, permittivity data from porcine intestine was used to analyze the characteristics of various degrees of ischemia/reperfusion injury. Our results show that dielectric relaxation spectroscopy can be used to assess intestinal viability. The dielectric constant and conductivity showed clear differences between healthy, ischemic and reperfused intestinal segments. This indicates that dielectric parameters can be used to characterize different intestinal conditions. In addition, machine learning models were employed to classify viable and non-viable segments based on frequency dependent dielectric properties of the intestinal tissue, providing a method for fast and accurate intraoperative surgical decision-making. An average classification accuracy of 98.7% was obtained using only permittivity data measured during ischemia, and 96.2% was obtained with data measured during reperfusion. The proposed approach allows the surgeon to get accurate evaluation from the trained machine learning model by performing one single measurement on an intestinal segment where the viability state is questionable.

Automated summary

This study analyzed the characteristics of different degrees of intestinal ischemia/reperfusion injury using permittivity data and proposed a method for assessing intestinal viability based on dielectric relaxation spectroscopy. The results showed that dielectric parameters can be used to distinguish healthy, ischemic, and reperfused intestinal segments. Additionally, machine learning models were used to classify intestinal tissue based on its frequency dependent dielectric properties, providing a fast and accurate method for intraoperative decision-making.