A tumour is the development of aberrant cells. A timely identification of cells with potential malignant transformation is vital for improving patient survival. The identification of tumour through MRI is dependent on the expertise of healthcare experts for segmentation and identification of tumour regions. This method is prone to human error due to its high dependence on human expertise. The recent advancements and development in the domain and field of deep learning and artificial intelligence in medical imaging have substantially improved and enhanced the efficiency of early-stage tumour detection. One of the key steps in the early detection of tumour using new age technologies like deep learning and artificial intelligence is the extraction of relevant features. This work aims to develop a threshold-based skull stripping methodology with contour-based morphological cropping of medical MRI images. Also, data augmentation is applied to increase dataset for training and validation. Five different models, viz., VGG16, VGG19, InceptionV3, EfficientNetB4, and custom CNN, are trained on the augmented and preprocessed images. In the study, VGG16, along with thresholding-based skull stripping, achieved the maximum F1-score of 0.9426 and precision of 0.9274 in separating the tumour image from the normal images.