A novel Gateaux derivatives with efficient DCNN-Resunet method for segmenting multi-class brain tumor

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Accurate brain tumor detection in Magnetic Resonance Images (MRI) is vital for medical diagnosis and treatment planning.
• Variations in tumor shape and size present challenges for precise location identification in MRI datasets.

Purpose of the Study:

• To propose a novel Deep Convolution Neural Network (DCNN) based Residual-Unet (ResUnet) model with Transfer Learning (TL) for precise brain tumor location prediction in MRI.
• To enhance the detection of multi-class brain tumors by improving the identification of boundary edges.

Main Methods:

• A customized Deep Convolution Neural Network (DCNN) model incorporating Residual-Unet (ResUnet) architecture was developed.
• Transfer Learning (TL) was employed for efficient feature extraction and Region of Interest (ROI) selection, accelerating model training.
• Min-max normalization and Gateaux Derivatives (GD) were utilized for enhancing ROI boundary detection and precise multi-class tumor identification.

Main Results:

• The proposed ResUnet model achieved high accuracy (up to 99.78%) on brain tumor and Figshare MRI datasets.
• Evaluation metrics demonstrated strong performance, including Jaccard Coefficient (up to 94.95%) and Dice Factor Coefficient (up to 92.37%).
• The system outperformed existing state-of-the-art segmentation models in MRI brain tumor segmentation tasks.

Conclusions:

• The novel DCNN-based ResUnet model with TL offers a robust and accurate solution for multi-class brain tumor segmentation in MRI.
• The proposed method effectively addresses challenges related to tumor variability, enhancing diagnostic capabilities.
• This approach shows significant potential for improving clinical workflows in neuro-oncology and pathology.