- Special Session 1 -

Generalized Hybrid Tumor Neural Network for Multiclass Brain Region Segmentation: A Deeply Optimized Multimodal MRI Framework Using the EurMor Dataset
● Please submit via: Electronic Submission System and select special session 1

• 
  
• Accurate neuroimaging is vital for effective brain tumor diagnosis and treatment planning. The implemented research presents a robust deep-learning framework for automated brain tumor classification and segmentation using multimodal MRI data. Leveraging the newly developed EurMor-Scan dataset, comprising four modalities. The model segments key tumor subregions known as Edema, Necrotic Core, and Infiltrative Tumor. The architecture integrates the encoder-decoder along with the bottleneck network as a backbone with a Region Proposal Network. This newly developed Brain Region Identification with the Generalized Hybrid Tumor Neural network enables both pixel-wise classification and object localization. Advanced data augmentation techniques are used to make the developed dataset bigger. 37,078 testing images are utilized, securing 9860, 8690, 9639, and 8889 for Edema, Necrotic code, Infiltrative, and non-tumor classes, respectively, to enhance model generalization, while hyperparameter optimization ensures training stability and accuracy. The implemented research evaluates using Average Symmetric Surface Distance and Maximum Symmetric Surface Distance, demonstrating that the model achieves high segmentation accuracy across tumor regions, with average boundary distances above 0.90 for Edema and above 0.80 for other classes. The overall classification accuracy of the non-tumor, Edema, Necrotic code, and Infiltrative tumor classes is 99.6, 96, 88.5, and 83.8% respectively. Through feature map visualizations and class-wise confusion matrices, the analysis is done by showing the high accuracy which is its strength, and capturing heterogeneous tumor structures. This research contributes to the advancement of AI-driven neuroimaging by providing a scalable approach, laying the groundwork for future integration into real-world medical workflows and broader applications in neuro-oncology.
  
  
• ● Related topics | 征稿主题
  Brain Tumor Classification, EurMor-Scan Dataset, Hybrid Tumor Neural Network, Data Augmentation, Average Symmetric Surface Distance (ASSD).
  

● Organizer(s)
Dr. Asad Ullah, Xi'An Eurasia University, China

asadullah@eurasia.edu