Brain tumour segmentation in medical image analysis is a challenging task because precision is crucial in the process of diagnosis and treatment. The current research applies a sophisticated method that utilizes Convolutional Neural Networks (CNNs) in conjunction with gradient-weighted Class Activation Mapping (Grad-CAM) to enhance the detection accuracy of brain tumours. By virtue of the implemented complex architecture of EfficientNetB1, our technique shines at solving the complex problems of medical image data processing. Grad-CAM makes a precious input into CNN by supplying visual interpretations of the attention-paying areas of CNN, empowering doctors to make the right diagnoses. We introduce a model that is based on a great number of brain tumour images with confirmed labels and learns to differentiate different tumour types based on their specific patterns. From our comparative analysis, we can see that there is a significant improvement in tumour detection accuracy, with our model reaching even as high as 99.67%. This one is more effective than the VGG16 model that delivers 85%-90% accuracy and ResNet50 model that has 90%-97% accuracy. In particular, the EfficientNetB1 model provides an accuracy range in the interval of 96%-98%, which clearly shows the efficiency of our proposed technique, since this would result in better treatment outcomes for patients.