Comparison of wavelet transformations to enhance convolutional neural network performance in brain tumor segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Accurate brain tumor segmentation in MRI is crucial for diagnosis and treatment.
• Deep learning methods, particularly convolutional neural networks (CNNs), show promise for automated segmentation.
• Combining methods, such as integrating wavelet transforms into deep networks, can enhance performance.

Purpose of the Study:

• To investigate the use of wavelet transforms as an auxiliary element in deep learning models for brain tumor segmentation.
• To analyze the requirements and effectiveness of combining wavelet functions with CNNs for MRI analysis.

Main Methods:

• Developmental study involving the application of various wavelet functions.
• Wavelet functions were used for compressing brain MRI images.
• These functions were integrated as auxiliary elements to improve CNN performance in brain tumor segmentation.

Main Results:

• The Daubechies1 wavelet function demonstrated the highest effectiveness in improving CNN performance for MRI segmentation.
• This function successfully balanced network performance with computational demands.
• The study identified optimal wavelet functions for enhancing CNNs in this specific task.

Conclusions:

• The selection of a wavelet function for optimizing CNN performance must align with specific problem requirements.
• Considerations such as computational load, processing time, and the wavelet's ability to enhance CNN output are critical.
• Tailoring wavelet function choice to the task ensures efficient and effective brain tumor segmentation.