Efficient Brain Tumor Multiclass Classification using Transfer Learning with DenseNet169

Tumours in or around the brain are the main cause of brain cancer, which is a prevalent illness throughout the globe. Due to incorrect tumour type identification and delayed diagnosis, adult brain cancer has a significant fatality rate while having a lower frequency and incidence than malignancies such as breast, skin, and lung. Reducing false detections and ensuring early diagnosis are crucial for improving patient survival rates. Recent deep learning (DL) approaches have shown exceptional performance, particularly in classification tasks. This study introduces a Transfer Learning architecture using DenseNet169, addressing overfitting by initializing weights and combining dense layers with global average pooling (GAP). Using a Brain MRI Dataset for multiclass classification (no tumor, glioma, meningioma, pituitary), the model achieved an average accuracy of 0.9535, ROC AUC Score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 9 6 3}$</tex>, Cohen Kappa Score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 3 7 6}$</tex>, and Matthews Correlation Coefficient of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 3 7 9}$</tex>, demonstrating its potential as a decision-support tool for radiologists diagnosing brain cancer patients.