Purpose: The purpose of this work is to segment multiple sclerosis (MS) lesions in magnetic resonance imaging (MRI) images, in which lesions in different sizes are segmented with appropriate accuracy. Automated segmentation as a powerful tool can assist professionals to increase the accuracy of disease diagnosis and its level of progression.
Methods: We present a deep neural network based on the U-Net architecture in which wavelet transform-based pooling replaces max pooling. In the first part of the network, the wavelet transform is used, and in the second part, it's inverse. In addition to decomposing the input image and reducing its size, the wavelet transform highlights sharp changes in the image and better describes local features. This transform has the multi-resolution characteristic, so its use provides improvement in the detection of lesions of different sizes and segmentation.
Results: The results of this study show that the proposed method has a better Dice similarity coefficient (DSC) value compared to the max pooling and average pooling methods.
Conclusion: The proposed method has better results for segmenting MS lesions of different sizes in MRI images than the max and average pooling methods and other methods studied.
Keywords: Deep learning; Magnetic resonance imaging; Multiple sclerosis; U-Net neural network; Wavelet.
© 2021. CARS.