Glioma, the most common primary brain tumor, has a 5-year survival rate of approximately 5–10%, particularly in high-grade glioma. Accurate diagnosis of glioma is crucial, as any misdiagnosis could have serious consequences for patients and potentially impact their chances of survival. Evaluating magnetic resonance images manually is a highly challenging task, and a definitive diagnosis still depends on surgical pathology. As a result, there is a demand for efficient digital techniques that can enhance the precision of tumor diagnosis. While there are numerous artificial intelligence methods available for brain tumor classification, it is worth noting that there is a lack of classification studies utilizing reinforcement learning algorithms with various architectures in Matlab. The present study aims to utilize reinforcement learning algorithms to diagnose glioma by applying different architectures and brain MR datasets. The proposed method involved conducting three different convolutional neural network architectures (DenseNet201, VGG16, and ResNet50) in Matlab to detect glioma brain tumors. Then, Q learning, an effective algorithm of reinforcement learning for problem solving, is integrated and trained. In the proposed approach, MR images including a total of 1203 glioma images and 396 normal brain images were evaluated. The results show that using the ResNet50 architecture and then the Q learning algorithm, the glioma diagnosis and detection accuracy is higher than the other architectures with 91% prediction.