Pancreatic cancer is characterized by the low survival rate and the challenges associated with early diagnosis. Echoendoscopy (EUS) reports the highest diagnostic sensitivity, and EUS-guided biopsy is the gold standard to stage the disease. An automatic approximated location of pancreatic tumors may support and characterize the diagnosis, and audit the procedure. This paper presents a method for automatically identifying the approximate location of pancreatic tumors in recorded videos of Echoendoscopy procedures. The process involves preprocessing video frames by reorganizing the radially captured intensities, filtering out speckle noise, and enhancing contrast. Subsequently, frames containing tumors are detected using a dedicated classifier, such as SVM or Resnet18. Afterwards a OLO (You Only Look Once) model is employed to predict the approximate localization of tumors. To evaluate the methodology, a dataset of 66,249 frames was used, including 18 cases diagnosed with Pancreatic Cancer, 5 cases of pancreatitis, and 32 cases reported as having a healthy pancreas. A cross-validation scheme was applied, randomly selecting 33 patients for training, 4 for validation, and 18 for testing. The highest achieved mean Intersection over Union (IoU) value was 0.42, and the precision, with an IoU threshold of 0.1, reached 85.3%.