7053SCN-Deep-Learning-and-Computer-Vision

Brain Tumor MRI Classification using Deep Learning: A Comparative Study of Custom CNN and EfficientNetB0

This project presents a comparative study of two deep learning approaches for multiclass brain tumor classification using MRI images. The objective is to automatically classify MRI scans into four categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor.

Two models were developed and evaluated: a Custom Convolutional Neural Network (CNN) built from scratch and a transfer learning model based on EfficientNetB0. The project includes comprehensive data preprocessing, image augmentation, model training, validation, and performance evaluation using metrics such as accuracy, loss, confusion matrix, and weighted F1-score.

Experimental results demonstrated that the EfficientNetB0 model significantly outperformed the Custom CNN, achieving 92.12% test accuracy and 91.96% weighted F1-score, compared to 75.00% accuracy achieved by the Custom CNN. The findings highlight the effectiveness of transfer learning for medical image classification tasks.

The project was implemented using Python, TensorFlow, Keras, NumPy, Matplotlib, and Google Colab GPU, providing an efficient and scalable solution for automated brain tumor diagnosis.

shaikha22/7053SCN-Deep-Learning-and-Computer-Vision

About

Resources

Stars

Watchers

Forks

Releases

Languages