Artificial intelligence (AI) has revolutionized dentistry, especially in radiology. It offers advanced data analysis, prediction, and decision support tools, increasing diagnostic accuracy and improving patient outcomes. This article aims to conduct a systematic review of the contribution of AI in the radiological diagnosis of maxillary lesions. An electronic exploration was conducted on four databases (PubMed, ScienceDirect, Google Scholar, and Web of Science) to identify retrospective diagnostic studies published between 2014 and 2024 relating to the radiological diagnosis of maxillary lesions by AI. Two reviewers carried out the bibliographic searches, the selection of studies, the extraction of data, and the quality assessment. Among the 315 articles initially identified, 13 retrospective diagnostic studies were retained because they met the eligibility criteria. These studies included a total of 16,578 panoramic images and nearly 20,789 volumes and sections of cone beam computed tomography (CBCT) for the training and validation of the chosen architectures. 22 software and architectures were used, such as You Only Look Once (YOLO) (v2, v3 and v5), Google Inception v3, EfficientDet-D3, VGG-16 or DenseNet121. The full text of the articles was carefully analyzed, and relevant data such as sensitivity, specificity, accuracy, F1 score, positive predictive value, and negative predictive value, as well as diagnostic time, were extracted. Direct comparisons with experienced radiologists were performed to evaluate the sensitivity, specificity, and diagnostic time of the AI, revealing a significant increase in accuracy and diagnostic time, with a difference of at least 23 min. AI diagnosis is, therefore, a method that has similar or even better results than a specialist in a significantly shorter time.