,
Professor, Department of Electronics and Communication Engineering, P. A. College of Engineering and Technology , Pollachi , India
,
Assistant Professor, Department of Artificial Intelligence and Data Science,P. A. College of Engineering and Technology , Pollachi , India
,
Assistant Professor, Department of ECE ,Sri Eshwar College of Engineering , Coimbatore , India
,
Professor, MAEER’S MIT , Mumbai , India
,
Assistant Professor, Department of Computer Science and Applications (MCA), SRM Institute of Science and Technology (FSH) , Chennai , India
Associate Professor, Department of ECE, P.A. College of Engineering and Technology , Pollachi , India
The problem of brain tumors is a range of different subtypes, which have a variety of clinical forms, and the diagnosis and treatment of tumors is a challenging task. This paper introduces a hybrid deep learning system that combines genomic profiling with MRI image analysis to provide an effective brain tumor subtyping. The framework is initiated by the preprocessing of MRI images, which is followed by grayscale conversion and noise reduction as done by Fast Non-Local Means (FNLM) filtering. This will aid in ensuring that important structural data is retained with minimal irrelevant noise. To conduct segmentation, the UNet++ framework is used, which is the best-performing architecture in medical image analysis. UNet++ enhances the conventional UNet by adding embedded skip routes, which allows a more productive information exchange between encoder and decoder networks, improving the accuracy of segmentation. The extraction of features is conducted by a local binary pattern (LBP), Gray-Level Co-occurrence Matrix (GLCM), and Discrete Wavelet Transform (DWT). These methods are able to reproduce both the frequency domain and textural characteristics of the tumor areas. The variables are further narrowed down to the most relevant ones by the Minimum Redundancy Maximum Relevance (mRMR) algorithm, thus only the most relevant features are taken in the classifier. The classification is done by an improved variant of the AlexNet that is optimized with the addition of batch normalization, global average pooling, and local response normalization parameters to minimize overfitting and maximize learning effectiveness. The model postulated in this study has a high performance of 99.79 %accuracy, 96.82 %sensitivity, 98.32 %specificity, and 98.61 %precision. These findings indicate the effectiveness of the hybrid approach, combining handcrafted characteristics and deep learning in early and confident brain tumor subtyping, which has considerable potential to enhance the level of diagnostic accuracy and individual treatment approaches in neuro-oncology.
This is an open access article distributed under the Creative Commons Attribution Non-Commercial License (CC BY-NC) License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
0
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.