<![CDATA[Brain hemorrhage is a critical medical condition that requires early and accurate detection to improve patient recovery outcomes. However, conventional image classification methods for brain hemorrhage still face limitations in terms of accuracy and efficiency. To address this issue, this study proposes optimizing the LeNet model using various activation functions—ReLU, Sigmoid, Tanh, and Swish—to enhance classification performance. Several optimization strategies were applied, including data augmentation techniques (flipping, rotation, shearing, rescaling) and fine-tuning of hyperparameters, to improve model generalization. Experimental results indicate that the model utilizing the Swish activation function achieves the most stable overall performance, with an accuracy of 55%, recall of 54%, precision of 54%, F1-score of 54%, and a ROC AUC value of 0.45. Although this performance is still below clinical application standards, the findings serve as an initial step toward exploring activation function optimization in CNN architectures. Further research is needed to significantly enhance classification accuracy and enable clinical viability.]]>
Copyrights © 2025
