Background: Arterial blood supply deficiency and venous congestion both play a role in anastomotic complications. Our aim was to evaluate a software-based analysis of the fluorescence signal to recognize the patterns of bowel ischemia. Methods: In 18 pigs, two clips were applied on the inferior mesenteric artery (group A: n = 6) or vein (group V: n = 6) or on both (group A–V: n = 6). Three regions of interest (ROIs) were identified on the sigmoid: P = proximal to the first clip; C = central, between the two clips; and D = distal to the second clip. Indocyanine Green was injected intravenously. The fluorescence signal was captured by means of a near-infrared laparoscope. The time-to-peak (seconds) and the maximum fluorescence intensity were recorded using software. A normalized fluorescence intensity unit (NFIU: 0-to-1) was attributed, using a reference card. The NFIU’s over-time variations were computed every 10 min for 50 min. Capillary lactates were measured on the sigmoid at the 3 ROIs. Various machine learning algorithms were applied for ischemia patterns recognition. Results: The time-to-peak at the ischemic ROI C was significantly longer in group A versus V (20.1 ± 13 vs. 8.43 ± 3.7; p = 0.04) and in group A–V versus V (20.71 ± 11.6 vs. 8.43 ± 3.7; p = 0.03). The maximal NIFU at ROI C, was higher in the V group (1.01 ± 0.21) when compared to A (0.61 ± 0.11; p = 0.002) and A–V (0.41 ± 0.2; p = 0.0005). Capillary lactates at ROI C were lower in V (1.3 ± 0.6) than in A (1.9 ± 0.5; p = 0.0071), and A–V (2.6 ± 1.5; p = 0.034). The K nearest neighbor and the Linear SVM algorithms provided both an accuracy of 75% in discriminating between A versus V and 85% in discriminating A versus A–V. The accuracy dropped to 70% when the ML had to identify the ROI and the type of ischemia simultaneously. Conclusions: The computer-assisted dynamic analysis of the fluorescence signal enables the discrimination between different bowel ischemia models.
Quero, G., Lapergola, A., Barberio, M., Seeliger, B., Saccomandi, P., Guerriero, L., Mutter, D., Saadi, A., Worreth, M., Marescaux, J., Agnus, V., Diana, M., Discrimination between arterial and venous bowel ischemia by computer-assisted analysis of the fluorescent signal, <<SURGICAL ENDOSCOPY>>, 2019; 33 (6): 1988-1997. [doi:10.1007/s00464-018-6512-6] [http://hdl.handle.net/10807/204464]
Discrimination between arterial and venous bowel ischemia by computer-assisted analysis of the fluorescent signal
Quero, Giuseppe;
2019
Abstract
Background: Arterial blood supply deficiency and venous congestion both play a role in anastomotic complications. Our aim was to evaluate a software-based analysis of the fluorescence signal to recognize the patterns of bowel ischemia. Methods: In 18 pigs, two clips were applied on the inferior mesenteric artery (group A: n = 6) or vein (group V: n = 6) or on both (group A–V: n = 6). Three regions of interest (ROIs) were identified on the sigmoid: P = proximal to the first clip; C = central, between the two clips; and D = distal to the second clip. Indocyanine Green was injected intravenously. The fluorescence signal was captured by means of a near-infrared laparoscope. The time-to-peak (seconds) and the maximum fluorescence intensity were recorded using software. A normalized fluorescence intensity unit (NFIU: 0-to-1) was attributed, using a reference card. The NFIU’s over-time variations were computed every 10 min for 50 min. Capillary lactates were measured on the sigmoid at the 3 ROIs. Various machine learning algorithms were applied for ischemia patterns recognition. Results: The time-to-peak at the ischemic ROI C was significantly longer in group A versus V (20.1 ± 13 vs. 8.43 ± 3.7; p = 0.04) and in group A–V versus V (20.71 ± 11.6 vs. 8.43 ± 3.7; p = 0.03). The maximal NIFU at ROI C, was higher in the V group (1.01 ± 0.21) when compared to A (0.61 ± 0.11; p = 0.002) and A–V (0.41 ± 0.2; p = 0.0005). Capillary lactates at ROI C were lower in V (1.3 ± 0.6) than in A (1.9 ± 0.5; p = 0.0071), and A–V (2.6 ± 1.5; p = 0.034). The K nearest neighbor and the Linear SVM algorithms provided both an accuracy of 75% in discriminating between A versus V and 85% in discriminating A versus A–V. The accuracy dropped to 70% when the ML had to identify the ROI and the type of ischemia simultaneously. Conclusions: The computer-assisted dynamic analysis of the fluorescence signal enables the discrimination between different bowel ischemia models.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
