This paper investigates the use of Deep Learning techniques for semanticsegmentation of central venous catheters (CVC) in chest X-rays. The goal isto assist in identifying these devices to determine their positioning, thereby reducing complications associated with invasive procedures. Initially, we addresschallenges such as the small relative size of the CVC devices in the imagesand data imbalance. For this, we utilize di!erent backbones, such as Resnetsand E"cientNets, in addition to making adjustments in image resizing. Toimprove the robustness and generalization capacity of the models, we applydata augmentation techniques. We implemented an ensemble of models, combining the results of various individual architectures, which proved e!ectiveby surpassing isolated models in various performance metrics. An additionalscript was developed to identify the presence of the CVC in the ensemble’spredictions by analyzing the count of active pixels and contour detection. Thefinal results demonstrated that the ensemble approach enhances the accuracyand reliability of CVC detection. Future research should focus on exploringthe classification of device positioning as a subsequent step, aiming to furtherimprove the clinical applicability of these techniques.