Deep learning model integrating features and novel classifiers fusion for brain tumor segmentation.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology

Background:

• Accurate segmentation and classification of brain tumors in medical images remain challenging.
• Conventional neural networks often struggle with the complexity of tumor delineation.

Purpose of the Study:

• To develop and evaluate deep learning models for precise brain tumor segmentation.
• To improve tumor delineation accuracy by combining Convolutional Neural Networks (ConvNet) and Long Short-Term Memory (LSTM) networks.

Main Methods:

• Utilized the MICCAI BRATS 2015 dataset comprising multi-modal MRI scans (T1, T2, T1c, FLAIR).
• Applied various preprocessing techniques including noise removal, histogram equalization, and edge enhancement.
• Trained individual ConvNet and LSTM models, then combined them into an ensemble for improved performance.
• Addressed class imbalance using class weighting in the proposed models.

Main Results:

• Individual ConvNet model achieved 75% accuracy.
• Individual LSTM network achieved 80% accuracy.
• The ensemble fusion of ConvNet and LSTM models yielded a final accuracy of 82.29%.

Conclusions:

• Deep learning models, particularly ensemble approaches, show significant promise for accurate brain tumor segmentation.
• The combination of ConvNet and LSTM networks offers superior performance compared to individual models.
• Further research in deep learning can enhance the precision of medical image analysis for oncology.