Abstract:
"Brain tumor detection from Magnetic Resonance Imaging (MRI) scans is a critical aspect of
clinical diagnosis and treatment planning in neuro-oncology. Manual interpretation of MRI images
by radiologists is time-consuming and prone to human error. This thesis addresses the challenge
by proposing an innovative approach that combines autoencoder-based dimensionality reduction
with ensemble Convolutional Neural Networks (CNNs) for improved brain tumor detection. The
autoencoder extracts essential features from high-dimensional MRI images while reducing
dimensionality, enhancing computational efficiency, and aiding subsequent classification.
Simultaneously, the ensemble CNN architecture aggregates the predictive power of multiple
models, mitigating overfitting risks and enhancing overall robustness.
This research used an autoencoder to condense MRI images, then combined multiple CNN models
in an ensemble for better classification. Testing on diverse datasets showed higher accuracy and
efficiency than current methods. This system promises effective brain tumor detection, aiding
clinical workflows and enhancing patient outcomes in neuro-oncology.
The brain tumor detection system developed in this project achieved compelling performance
metrics on MRI images. With an accuracy of 90.6%, precision of 0.93, recall of 0.876, F1 score
of 0.903, and an AUC of 0.91, the system demonstrates robustness and effectiveness in identifying
tumors. By leveraging autoencoder-based dimensionality reduction and an ensemble of CNN
models, the approach not only enhances diagnostic accuracy but also streamlines clinical
workflows. These results underscore the system's potential as a valuable tool for early tumor
detection, treatment planning, and disease monitoring in neuro-oncology, contributing to improved
patient care outcomes. "
