Reproducibility Package for CastorNet and Baseline CNNs on BRISC‑2025 and Brain Tumor Datasets

This repository contains the full reproducibility package for the manuscript: “CastorNet: A Lightweight Brain Tumor Classification Network Integrated with Graphical User Interface Software for a Clinical Prediction System.” Contents BRISC‑2025 Models: Trained weights, training curves (accuracy vs. epochs, loss vs. epochs), logs, and confusion matrix summaries for CastorNet and baseline CNNs (VGG16, VGG19, ResNet‑50, Inception‑V3, MobileNet‑V2). BRISC‑2025 Ablation: Single Path and Swish+Double architectural variants of CastorNet, each organized by 5‑fold cross‑validation (t1–t5) with models, curves, logs, and results. Brain Tumor Dataset (Kaggle/Figshare): Comparative experiments across six CNN architectures, organized by fold (t1–t5) with trained models, training curves, logs, and confusion matrices. Purpose All files are provided to ensure full reproducibility of experiments. The package includes: Trained models (.keras) for CastorNet and baselines. Training curves (accuracy and loss vs. epochs) for each fold. Logs documenting hyperparameters, optimizer, learning rate, and training progress. Results, including confusion matrices and classification reports. Ablation studies demonstrating the robustness of CastorNet variants. Keywords Brain tumor MRI, CastorNet, deep learning, convolutional neural network, explainable AI, reproducibility, medical imaging, BRISC‑2025, Kaggle, Figshare.