Diseases of the gastrointestinal (GI) tract are among the most common pathologies in the world population and are responsible for thousands of deaths every year. This work proposes an automatic method for detecting regions with GI Tract abnormalities, intending to reduce the number of lesions missed in Wireless Capsule Endoscopy (WCE) video exams by expert endoscopists. By taking advantage of convolutional neural networks (CNNs) and YOLO detection models, the proposed method not only increases the reliability of pathological detection in WCE images, but also sets a new benchmark in this field. Our results for binary classification between healthy and pathological images are promising, with an accuracy of 87.8%, precision of 91.6%, recall of 89.1% and F1-Score of 90.3%. In addition, the detection model showed an Intersection over Union (IoU) of 31.33% among all the images classified as pathological. The impact of this research is significant, as it provides a method capable of detecting GI Tract diseases in WCE images and contributing to better clinical decision-making and patient care.