A state-of-the-art technique to perform cloud-based semantic segmentation using deep learning 3D U-Net architecture

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Glioma is an aggressive brain tumor requiring accurate segmentation for diagnosis and treatment.
• Semantic segmentation using deep learning, particularly 3D U-Net architectures, is crucial for processing brain tumor imaging data.
• Existing methods for glioma segmentation vary in accuracy and efficiency.

Purpose of the Study:

• To present a novel cloud-based 3D U-Net method for precise brain tumor segmentation.
• To evaluate the performance of the proposed method using the BRATS dataset.
• To compare the developed method against existing state-of-the-art techniques in glioma segmentation.

Main Methods:

• Implementation of a unique cloud-based 3D U-Net architecture for semantic segmentation.
• Training the model using the Adam optimization solver with optimized hyperparameters.
• Utilizing the BraTS (Brain Tumor Segmentation) dataset for model training and validation.

Main Results:

• Achieved an average Dice score of 95%, indicating high accuracy in brain tumor segmentation.
• The proposed cloud-based 3D U-Net method demonstrates superior performance compared to existing architectures.
• The study provides a comprehensive literature review of recent brain tumor segmentation methodologies.

Conclusions:

• The developed cloud-based 3D U-Net offers a highly accurate and efficient solution for glioma segmentation.
• This method represents a significant advancement in cloud-based deep learning applications for brain tumor analysis.
• The high Dice score achieved positions this method as a leading approach for clinical applications in neuro-oncology.