What question did this study set out to answer?

The aim is to develop an automated system for classifying brain tumors from MRI images using deep learning techniques.

May 2, 2026SHILAP Revista de lepidopterología0 citationsOpen Access

Automated Brain Tumor Classification using Deep Convolutional and Transfer Learning Networks

VPVinodkumar PatilShri Vile Parle Kelavani Mandal MPManisha PatilShri Vile Parle Kelavani Mandal DDDr Pramod DeoreShri Vile Parle Kelavani Mandal

Key Points

The aim is to develop an automated system for classifying brain tumors from MRI images using deep learning techniques.
Merged two publicly available MRI tumor datasets for diversity.
Developed a baseline convolutional neural network and compared it with four transfer learning models.
Pre-processed images to a standard size of 224 × 224 pixels for analysis.
ResNet50V2 achieved 97.20% accuracy with high precision and recall.
Transfer learning models outperformed the baseline CNN in classification accuracy.
Demonstrated effectiveness of using pre-trained networks for tumor identification.

Abstract

Brain cancers are some of the fastest-growing and most deadly types of neurological diseases in medicine. Early detection with accuracy is very important to improve the survival of patients. Manually reading MRI scan images is a slow process. It requires a special skills person and can differ from one observer to another. It is in this context that the automatic computer-aided diagnosis system has emerged as a vital research area. In this work, we use deep learning-based methods to classify various types of brain tumors using MRI. We developed a baseline convolutional neural network and compared it with four transfer learning models: MobileNetV2, VGG16, VGG19, and ResNet50V2. To ensure data diversity and robustness, we merged two publicly available MRI tumor datasets and normalized, balanced, and pre-processed the data to a constant 224 × 224-pixel size for each image of the four categories: glioma, meningioma, pituitary tumor, and no tumor. The experimental results show that the transfer learning approach performs significantly better than the CNN baseline. ResNet50V2 became highly effective, providing 97.20% accuracy, high precision, and excellent recall. These findings demonstrate that combining pre-trained neural networks with integrated datasets can provide better results and a scalable framework for automated brain tumor identification and classification.

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper