Publication Details
Abstract
Detecting brain tumors is still a big problem in medical diagnostics. This is made even harder by the fact that there aren't many labeled datasets, MRI imaging quality varies, and high diagnostic accuracy is needed in clinical situations where time is of the essence. Conventional manual diagnosis is time-intensive and prone to observer variability, whereas independent deep learning models frequently encounter generalization challenges, especially within limited and imbalanced medical datasets. To tackle these challenges, this study presents InceptiNeuro-V2B1, a hybrid transfer learning-based CNN system for accurate and instantaneous brain tumor classification. The suggested approach utilizes MobileNetV2 as a pre-trained foundational model, meticulously adjusted to differentiate between glioma, meningioma, pituitary tumors, and normal brain scans, employing sophisticated preprocessing and real-time data augmentation techniques. The model is part of a web app built on Flask that lets users upload MRI images directly and get real-time predictions on tumors. Using a public brain MRI dataset for evaluation, Inceptineuro-V2B1- V2B1 got a test accuracy of 95.1% and high precision, recall, and F1-scores for all types of tumors. The system guarantees strong performance even on imbalanced datasets by using transfer learning, a lightweight architecture, and clinical usability. InceptiNeuro-V2B1 is a decision-support tool for real-world medical settings that is scalable, ready to be deployed, and efficient. It improves early diagnosis and treatment planning in neuro-oncology.