The article discusses the challenges associated with the diagnosis of glioblastoma multiforme (GBM), which is a life-threatening form of cancer with low median survival rates. Medical image analysis is a challenge due to limited annotated data. However, transfer learning can help overcome this issue. In this study, deep learning models are trained to develop an automatic GBM diagnosis system that classifies the brain images as no tumor and glioma tumor. The performance of a customized convolutional neural network is compared to pretrained ResNet-50 and VGG-16 models based on metrics namely accuracy, specificity, sensitivity, precision, F1-score, and Matthews Correlation Coefficient (MCC). A critical analysis is conducted to determine the most efficient learning model for GBM detection based on the evaluated performance metrics. The study provides insights into developing a more efficient GBM diagnosis system, which can help in the development of novel therapeutic targets and treatment modalities for cancer.