A Novel DenseNet Framework for Brain Tumor Classification Enhanced by XgBoost and Fire Hawk Optimization

Nowadays, Brain tumours are the most crucial disease that is spreading rapidly all around the world.According to the statistics, more than a thousand people are losing their lives to this tumour in every country.The early prediction of tumours can help to diagnose and overcome the disease quickly.For an earlier prediction, there are Numerous research techniques like deep learning (DL) and machine learning (ML) models used for feature extraction and classification.In some cases, the hybrid models are also used for feature extraction and classification.However, the accuracy is not attained up to the level of satisfaction for various tumours like Glioma, pituitary, and meningioma.We propose a novel DenseNet architecture incorporating Self-Calibrated Squeeze-and-Excitation (SC-SE) for enhanced feature extraction and representation.The SC-SE DenseNet integrates SC Convolutions (SC-Conv) and SE Blocks within a DenseNet framework.SC-Conv dynamically recalibrates both spatial and channel-wise features, which improves the model's ability to adapt to diverse input variations.SE blocks are used to concentrate important channels for better feature learning.The extracted features from SC-SE DenseNet are classified using the XGBoost classifier.To further optimize performance, the Fire Hawk Optimizer (FHO) is used for feature selection and hyperparameter tuning.FHO aids in selecting the most relevant features from the input image and reduces dimensionality.Additionally, FHO is used to fine-tune the parameters of the XGBoost classifier to increase the classification accuracy.