Deep Learning Approaches for Brain Hemorrhage Classification: A Comparative Evaluation

The serious medical illness linked to blood vessel rupture, intracranial hemorrhage, requires immediate attention for the best results. The classification performance of computed tomography scan bleeding detection is the main objective of this investigation. Brain hemorrhage is the term for this condition. In this life-threatening medical emergency requiring skilled medical personnel, it is crucial to promptly identify the site of bleeding before administering treatment. This study proposes the use of deep learning models, including Convolutional Neural Networks (CNNs), namely ResNetV2, DenseNet, MobileNet, VGG16, and EfficientNet architectures, for brain hemorrhage classification. The accuracy and computational capacity of the popular CNN models are increased using a dataset of 200 head computed tomography scans with an augmentation technique. The efficacy of the CNN models is investigated in this study by analyzing popular optimizers, including Adam, NAdam, AdamW, LAMB, and Lion. The precision, recall, F1-score, specificity, accuracy, and loss of the training and testing sets for the specified model are used to evaluate the performance of the given CNN models. This comparative analysis indicates that the MobileNet model with the AdamW optimizer provides the best performance among the CNN models, achieving training and testing accuracies of 99% and 98%, respectively. It also demonstrates strong performance across a wide range of evaluation metrics. The comparative investigation contributes to the early detection and treatment of serious medical disorders associated with intracranial hemorrhage by adding to the expanding corpus of knowledge in medical image analysis.