Deep Learning Based Brain Tumor Detection and Classification Using the BR35H MRI Dataset

Rupali Umbare, Dr. Shailesh Kumar, Dr. Ram Joshi

Abstract: Brain tumors are considered one of the most severe types of cancer, making their timely and precise identification essential for enhancing patient prognosis and treatment effectiveness. Conventionally, Manual examination is a major component of MRI-based diagnosis, which is time-consuming and prone to human mistake. This study presents a cutting-edge deep learning-based method to overcome these constraints for the automatically identifying and categorising brain tumours, utilizing the BR35H MRI dataset. The proposed system incorporates five distinct models: DenseNet and ResNet50 for classifying medical images, YOLOv5 and Faster R-CNN for identifying tumor regions, and LSTM networks to capture temporal and contextual features across MRI slices. The methodology involves preprocessing of MRI scans, data augmentation, model training, and evaluation Utilising standard metrics including mean Average precise (mAP), accuracy, F1-score, and precise recall. The models are evaluated for both computational efficiency and detection accuracy. A comparative analysis reveals that YOLOv5 is best appropriate for real-time scenarios because its high frame rates, whereas Accuracy is higher with faster R-CNN. for detailed localization. This paper aims to bridge classification and detection capabilities in a single framework. This study advances the field of AI-driven diagnostic systems for neuro-oncology, with particular relevance to under-resourced environments where access to expert radiologists is limited.

Keywords: Brain tumor detection, DenseNet, YOLOv5, Faster R-CNN, BR35H Dataset

How to Cite?: Rupali Umbare, Dr. Shailesh Kumar, Dr. Ram Joshi, "Deep Learning Based Brain Tumor Detection and Classification Using the BR35H MRI Dataset", Volume 14 Issue 8, August 2025, International Journal of Science and Research (IJSR), Pages: 398-402, https://www.ijsr.net/getabstract.php?paperid=SR25807191301, DOI: https://dx.doi.org/10.21275/SR25807191301