Algorithm 1: Brain Tumor Classification using ConvNeXT Model

Input:  • Brain MRI dataset (images and labels)  • Batch size  • Training epochs

Output:  • Tumor class label C ∈ {Glioma, Meningioma, Pituitary, No_tumor}  • Performance metrics

// Step 1: Dataset Loading and Preprocessing (1)Load MRI images using image_dataset_from_directory. (2)Apply dataset transformations: resize images to 224 × 224, convert to tensors, normalize using ImageNet statistics and augment using on-the-fly. (3) Split the dataset into: ○ 80% training ○ 20% validation (4)Prefetch all datasets for performance optimization. // Step 2: Model Initialization (5)Import the ConvNeXt Base model pre-trained on ImageNet. (6)Replace its classifier head with a new linear layer to output four classes. // Step 3: Training Setup (7)Define the loss function as cross-entropy loss. (8)Initialize the optimizer (Adam) with a learning rate of 1 × 10−4. (9)Move the model and data to the GPU if available. // Step 4: Training Loop (10)For each epoch: (11)Set the model to training mode. (12)For each batch in the training set: (13)Forward propagate input images through the model. (14)Calculate the loss. (15)Backpropagate gradients and update model weights. (16)Calculate epoch training loss and accuracy. // Step 5: Validation Loop (17)Set the model to evaluation mode. (18)For each batch in the validation set: (19)Forward propagate validation images, record loss, and predictions. (20)Compute validation loss and accuracy. // Step 6: Benchmarking and Explainability (21)Calculate accuracy, precision, recall, F1-score, AUC, Kappa (22)Measure and record inference speed (FPS), model size (MB), GPU/CPU memory footprint, and power consumption. (23)Apply Grad-CAM++ and gradient SHAP for model interpretation. // Final Classification (24)Apply SoftMax classifier → C (25) Return: ○ Final tumor class C ○ Performance metrics ○ Visual overlays