Enhancing Multi-Class Tumour Detection with DenseNet121: A Deep Learning Approach in Medical Images

Abstract

Technology has played a pivotal role in revolutionizing healthcare through digital transformation. Healthcare systems now include a vast infrastructure of medical information systems, electronic devices, medical records, wearable and smart devices and mobile devices. The progress in medical infrastructure, coupled with advances in computational methods, has enabled the researchers and practitioners to develop new solutions. The present work intends to articulate the essentiality of enacting CNN deep learning with Transfer Learning technology for efficient brain tumour detection and classification. In the proposed work, a strong data set including 2,870 training images and 394 test images considered with four class detection; MRI scans respectively tumour classification data and site segmentation of the tumour and some augmentation data are used to enable model implementation to improve training as well as testing. Fine training the deep learning model, where dense neural architecture (DenseNet-121) is trained using pre-processed MRI images. The obtained training and validation accuracies are 99.98% and 98.72%, respectively. Evaluation metrics, including precision, recall and F1 score, confirm the exceptional performance of the model in identifying and classifying brain tumours. Visualizations such as confusion matrices, training/validation loss and accuracy plots provide comprehensive insights into the typical training process and its overall effectiveness. These results demonstrate the potential of the proposed approach to significantly advance the diagnosis of brain tumours, ultimately benefiting patients.