Alam, Md AshrafulHossain, Shah Md. ShakhawathAlam, F M TahoshinFaiyaz, Hazra Mohammed Ahnaf2024-09-082024-09-08©20242024-05ID 18101133ID 18101030ID 17241014http://hdl.handle.net/10361/24008Cataloged from PDF version of thesis.Includes bibliographical references (pages 40-43).This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science, 2024.The early and accurate diagnosis of brain tumors is a critical challenge in medi cal imaging, significantly impacting treatment outcomes and patient survival rates. Despite the advancements in imaging technologies, the interpretation of MRI scans remains a complex and subjective task. This research introduces a novel cross modality deep learning approach aimed at enhancing the performance of multiclass brain tumor classification by leveraging superior imaging representations to guide and improve the analysis of less effective modalities. Our methodology involves the development of a guidance model that utilizes the robust representations de rived from high-quality imaging modalities to enhance the diagnostic accuracy of more practical but less efficient modalities. Specifically, we employed deep learn ing techniques to process and analyze MRI and histology data, including Convolu tional Neural Networks (CNNs) such as ResNet50, EfficientNetB0, InceptionV3, and DenseNet121. The guidance model integrates these representations to construct an ensemble model that achieves superior performance. The results demonstrate that our guidance model significantly improves the diagnostic accuracy of the subordinate modality. In the case of brain tumor classification, the model not only surpasses the performance of models trained solely on the superior modality but also achieves com parable results to those utilizing both modalities during inference with the guidance ensemble accuracy of 94.61%. Compared to this, other models such as Efficient NetB0 achieved 94% and DenseNet121 achieved 93% test accuracy. This approach offers a practical and efficient solution for enhancing diagnostic accuracy while mini mizing the reliance on more costly and less accessible imaging technologies. Overall, our cross-modality deep learning model represents a substantial advancement in the field of medical imaging, providing a more accurate, reliable, and cost-effective method for the diagnosis of brain tumors.43 pagesenBrac University theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission.Convolutional neural networksNeuro-oncologyMagnetic resonance imagingEnsemble modelsGuidance modelDeep learning (Machine learning).Brain--Tumors.Thesis