Patients will be recruited independently in each of the centers participating in the study, according to the specific organization of each service (hospital and home). Patients will be selected based on the study inclusion/non-inclusion criteria. The included patients will receive oral information about the study. The informed consent will be obtained before any investigation.

Each photograph will focus on the intravascular catheter. The investigators will respect the main non-inclusion criterion: not to show any peripheral identification sign close to the insertion point of the catheter that cannot be masked when the photograph is taken. In this context, jewellery, clothing, tattoos, scars, and birthmarks represent identifying features. Individual access accounts with secure, randomly generated passwords will be provided to investigators.

The photographs will be anonymous, and any identification of the person concerned will be impossible.

Three medical experts have been selected to review the photo collected. Each expert medical assesses the presence of local signs of infection on the photographs by annotating them directly via a dedicated software. They will annotate local signs: redness, perfusion extravasation, necrosis, hematoma, edema, non-purulent discharge, and purulent discharge.

A convolutional neural network model will determine the probability of local sign presence. Each picture will be annotated to determine the main characteristics of the catheter. A dataset preparation with photo cropping will be performed for modelling.

Overall classification accuracy of the learning model compared to the assessment of three independent medical experts on the detection of the presence of redness greater than or equal to 5 mm at the catheter insertion site.

The precision metric focuses on Type-I errors(FP). A Type-I error occurs when we reject a true null Hypothesis.