Xception-based Deep Learning Model for Precise Brain Tumour Classification
Goldy Verma
Abstract
Brain tumour, being one of the major health diseases, it becomes one of the major factors for accurately categorizing the brain tumour which is required for analysing the strategies for the treatment and patients outcomes. This work uses a fine-tuned Xception model to classify brain tumours into three types: pituitary tumours, meningiomas, and gliomas. Originating from glial cells, gliomas are very aggressive; meningiomas and pituitary tumours, although usually benign, can cause major medical problems because of their location and effects on hormonal balance. Deep learning, particularly the usage of convolutional neural networks (CNNs) like Xception, has lately showed considerable potential in medical image analysis. This work made use of a 2,000 MRI image dataset from the Kaggle platform, fairly split among the three tumour classes. Resizing, normalising, and data augmentation helped the images to be pre-processed so improving model performance. Trained and validated, the fine-tuned Xception model produced results with a training accuracy of about 98% and high precision, recall, and F1scores spanning overall classes. Analysing potential significance, the training and validation losses, accuracies, and confusion matrices helps to further validate the model by thereby improving diagnosis and treatment outcomes. These results show how well the fine-tuned Xception model classifies brain tumours, therefore paving the way for its application in clinical practice to raise patient care and diagnosis accuracy.