ABHINAV V P (STM21CS004)
AHIN SURESH (STM21CS010)
DRUPAD M K (STM21CS028)
NIHEL VINOD MAROLI (STM21CS046)

This paper addresses critical gaps in brain tumor classification by proposing an advanced deep learning model capable of identifying multiple tumor types, including meningioma, glioma, pituitary tumors, and space-occupying lesions. A key innovation of this model is its ability to detect multiple tumors within a single MRI scan, a feature essential for more accurate and comprehensive diagnostics. The classification is performed using a convolutional neural network (CNN) model trained on an expanded dataset of 10,712 MRI images, while segmentation is conducted using the YOLOv10 model.
To enhance model stability, precision, and generalizability, we implemented advanced deep learning techniques, including data augmentation and regularization. This optimized approach refines the model’s performance and bolsters its applicability across varied
clinical scenarios. Evaluation on a test set demonstrates high class-specific accuracies and robust metrics, including precision, recall, F1-score, and AUC-ROC.
By accurately identifying diverse tumor types and detecting multiple tumors in individual MRI scans, this model shows substantial potential in advancing early, reliable, and precise brain tumor diagnostics. These findings underscore the value of deep learning in surpassing traditional diagnostic methods, paving the way for future research on neural network-based classification systems for complex tumor detection and broadening the scope of automated medical imaging analysis.