Objective Fine Vessel Assessment Using a Novel Endoscopic Ultrasound Detective Flow Imaging-based Computer Vision Software to Differentiate Between Normal and Abnormal (Acute, Chronic, and Steatotic) Pancreas: A Pilot Study

Aims Detective flow imaging (DFI) is a relatively new endoscopic ultrasound (EUS) imaging technology helpful in visualization of fine vessels (FV) with low velocities in the absence of motion artifacts. Microvasculature changes have been seen in both acute pancreatitis (AP) and chronic pancreatitis (CP). Up until now, DFI FV detection has been measured in a subjective way, based on the visual impression of the endoscopists. In the present study, we developed a computer vision (CV) software aimed to provide an average number of FV observed during EUS-DFI and the maximum number of FV detected (Figure 1), in order to establish an objective measurement for DFI evaluation. In the present study, we aimed to establish a cut-off value of the number of FVs detected through the CV software to differentiate between normal (NP) and abnormal pancreas (AP, CP, and pancreatic steatosis (PS)) [1].

Methods A single-center, observational, prospective, pilot study performed at the Instituto Ecuatoriano de Enfermedades Digestivas, Ecuador. Four cohorts (NP, AP, CP, and PS) were included in the study. Patients over 18 years-old who required EUS evaluation underwent EUS-DFI assessment. A p-value<.05 will be considered statistically significant.

Results A total of 23 patients were included in the study. The mean age was 54.3 (SD: 16.6) and 14 (60.9%) were female (Table 1). Among the included patients, 4/23 (17.4%) normal pancreas, 3/23 (13%) had PS, 4/23 (17.4%) had AP, and 12/23 (52.2%) had CP. The median number of FV observed in NP was 6.5 (4.0-11.0), whereas the median number of FV in abnormal pancreas was 16 (13.5-17.0). The median number of FV observed for PS, AP and CP were 16 (15-17.5), 15.5 (12.8-17.3), and 16 (12.8-17.0), respectively (p=.04). The median maximum number of FV observed during EUS-DFI evaluation was 36.5 (35-51.3) for NP and 40 (29.5-48.0) for abnormal pancreas (p=0.773). The maximum number of FV observed in PS, AP, and CP are 29 (28-38.5), 34.5 (27.5-44.0), and 41 (35-49) CP (p=.486). A cut-off value was determined at 9.5 for the average of FV observed during EUS-DFI, achieving a sensitivity, specificity, PPV, NPV, and observed agreement of 95%, 75.0%, 95.0%, 75%, and 91.0%, respectively. The software obtained an area under the receiver operator curve (AUROC) of 0.822 to differentiate between normal and abnormal pancreas.

Conclusions In the present pilot study, the DFI-EUS-based AI software was capable to differentiate between normal pancreas and abnormal pancreas with high observed agreement. There was no difference between AP, CP and PS regarding the average number of FV.

Publikationsverlauf

Artikel online veröffentlicht:
27. März 2025

• References

• 1 Pitkäranta P, Kivisaari L, Nordling S, Nuutinen P, Schroder T.. Vascular changes of pancreatic ducts and vessels in acute necrotizing, and in chronic pancreatitis in humans. Int J Pancreatol 1991; 8 (1): 13-22 PMID: 2033315