Brain tumor detection and classification using MRI images remain critical in early diagnosis and treatment planning. Accurate identification of tumor regions and classification into benign or malignant categories can significantly improve patient outcomes. However, challenges such as high-dimensional data, varying image quality, and limited computational efficiency continue to hinder clinical deployment. Conventional classification models often struggle with either segmentation accuracy or computational efficiency. Moreover, a standalone deep learning or probabilistic approach may fail to capture both spatial and probabilistic features effectively, impacting overall diagnostic performance. This study presents a hybrid deep learning approach integrating Probabilistic Neural Networks (PNN) and Residual DenseNet classifiers for advanced brain MRI image classification and segmentation. MRI datasets were acquired from open-source repositories, encompassing diverse tumor presentations. The preprocessing stage included image enhancement to improve consistency and clarity. Dimensionality reduction was implemented to streamline the dataset for efficient processing. Feature extraction employed fast discrete curvelet transformation to capture critical texture details. The transformed images were then used to generate probabilistic prototypes based on approximate coefficients, which were input into the PNN for initial feature learning. The final classification was performed using a Residual DenseNet, which effectively distinguishes between benign and malignant tumors, as well as normal and abnormal brain regions. The proposed hybrid model demonstrated improved classification accuracy, efficient segmentation of tumor regions, and significantly reduced computational time compared to standalone methods. The use of curvelet-based features enhanced diagnostic precision, and the combination of probabilistic modeling with deep learning yielded a robust performance across diverse test cases. This framework offers potential for integration into clinical decision support systems for early brain tumor detection and diagnosis.
Vinitha Kanakambaran, Avinash Gour
Mansarovar Global University, India
Brain MRI classification, Probabilistic Neural Network, Residual DenseNet, Curvelet Transformation, Tumor Segmentation
| January | February | March | April | May | June | July | August | September | October | November | December |
| 0 | 0 | 0 | 3 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Published By :
ICTACT
Published In :
ICTACT Journal on Soft Computing
( Volume: 16 , Issue: 1 , Pages: 3782 - 3786 )
Date of Publication :
April 2025
Page Views :
36
Full Text Views :
3
|