Znet: Deep Learning Approach for 2D MRI Brain Tumor Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Brain tumor detection and segmentation in MR images are critical for early diagnosis and treatment planning.
• Deep learning (DL) offers significant potential for improving accuracy in medical image analysis.

Purpose of the Study:

• To present a novel deep neural network (DNN) framework, Znet, for segmenting 2D brain tumors in MR images.
• To leverage data augmentation to enhance the performance of the Znet model with limited expert-annotated data.

Main Methods:

• The Znet framework utilizes skip-connection and encoder-decoder architectures.
• Data augmentation strategies were employed to generate a large dataset of synthetic tumor cases from a smaller set of expert-delineated tumors.
• The model was trained and evaluated on MR images, focusing on low-grade glioma (LGG) segmentation.

Main Results:

• The Znet model achieved a high mean Dice similarity coefficient of 0.92 on the independent testing dataset.
• Other evaluation metrics, including pixel accuracy (0.996) and F1 score (0.81), also indicated strong performance.
• Visualizations confirmed the model's capability for accurate tumor localization and auto-segmentation.

Conclusions:

• Deep learning methods, specifically the Znet framework, are effective for brain tumor detection and segmentation in MR images.
• Metrics like Dice and IoU are more reliable than pixel accuracy for evaluating semantic segmentation in imbalanced datasets.
• The Znet approach shows promise for clinical deployment as an automated tumor segmentation tool.