Brain umours (BT) are highly prevalent and dangerous disease, have extremely short prognoses at the most malignant grade. Merging the MRI modes results in hybrid images with information that is used to classify tumours. The obtained images are luxurious to gain and hard to store, the diagnostic process consume a substantial amount of time. The magnetic resonance imaging (MRI) provides clean structural data, but it is time-consuming. To overcome these issues, a novel deep learning-based model is proposed for the early detection and classification of brain tumour from MRI images. Initially, the input images are pre-processed utilizing Clifford gradient to improve the quality of the image. Then, hybrid dilated CNN model (HD-CNN) is employed for extracting the features in the pre-processed brain image. Afterward, the extracted features are fed into the Attention Link Net model to classify the four cases of brain tumour. According to the test result, the proposed model has a 99.23% accuracy rate. The Attention Link Net high accuracy than Alex Net, Dense Net, and Res Net which obtains 0.15%,0.31%, and 0.25% while having a significantly lower computational cost than other networks. The proposed model improves overall accuracy by 2.35%,0.94%, and 2.42%, over the VGG19, DNN, and DCNN, respectively.