Enhanced Brain Tumor Detection Via Hybrid CNN Transfer Learning and Deep Feature Fusion

1M LAHARI, 2C RAJA SEKHAR, 3KASIREDDY MANASA, 4NAGIRI MOHITH, 5KOUTHARAPU JAYA NIKHILESH

1Assistant Professor Dept of Information Technology, SV College of Engineering, Tirupati, India.

2B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

3B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

4B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

5B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

ABSTRACT:A brain tumor is an abnormal growth of cells within the brain that can be benign or malignant. It disrupts normal brain functions by exerting pressure on surrounding tissues and interfering with neural activity. Brain tumor classification using MRI images is a challenging yet critical task for early diagnosis and effective treatment. The existing brain tumor classification methods primarily use individual CNN architectures or traditional machine learning techniques. While CNNs effectively extract features from MRI images, they often suffer from limited receptive fields, homogeneous feature maps, and poor generalization on imbalanced datasets. Current models such as ResNet, MobileNet, and DenseNet typically neglect deep feature fusion, advanced augmentation, and hybrid learning, leading to reduced adaptability and accuracy across diverse tumor types.The proposed system introduces a hybrid CNN-based transfer learning framework that combines ResNet50V2, MobileNetV2, and DenseNet121. Deep features from these models are concatenated and refined through additional convolutional layers to capture complementary tumor characteristics. MRI-specific preprocessing, including region-focused cropping and affine transformations, enhances data quality and class balance. Selective fine-tuning further adapts pretrained models to MRI domain nuances. This hybrid approach improves classification performance—achieving higher accuracy, precision, recall, and F1-score compared to individual models—while sustaining computational efficiency.Keywords: machine learning, individual CNN architectures, ResNet, MobileNet, and DenseNet, deep feature fusion, advanced augmentation, hybrid learning, transfer learning.