Though endoscope-based surgery is highly beneficial regarding the recovery of the patients, it may also raise difficult tasks during special surgical actions. The main disadvantage is the lack of the tactile information for the surgeon, who consequently has to rely on the visual information only. As a special scenario, some organs of the human body, like ureters and arteries, have similar visual appearances, so distinguishing them is very challenging in real-time endoscopic surgery in these days. Thus, robust image processing-based solutions are highly welcome in supporting endoscopic keyhole-surgery. Accordingly, here, we present a semi-automatic tool, which is able to make the difference between ureters and arteries using a convolutional neural network (CNN). The CNN has been trained on 2000 images acquired during real endoscopic surgeries and tested on other 500 ones. We have achieved 94.2% classification accuracy regarding the correct recognition of the two types of organs.