Selective Brain MR Image Compression Through Wavelet Optimization and Enhanced Convolutional Neural Network Model

Tele-healthcare systems must store and transmit the digital data created by the various imaging modalities. For efficient handling of these data, compression techniques that provide a greater compression ratio with notable image quality are needed. This research proposes a selective image compression technique for brain MR images leveraging the energy compaction property of the wavelet coefficients and the feature extraction and learning capabilities of CNN. The procured images undergo a hybrid filter to eradicate the possible noise in the image. The Fuzzy C-Means technique optimized using the Greywolf optimization algorithm is used to segment the clinically relevant region from the rest of the MR image. Clinically significant areas are compressed using optimized zerotree wavelet transform to ensure the reconstructed image quality. The background information is compressed using enhanced CNN models to achieve a greater compression ratio. The algorithm put forward is implemented using MATLAB 2021a, and the algorithm completion is evaluated using the BRATS 2018 brain MR image dataset. The evaluation metrics show a commendable performance over the analyzed compression techniques with a PSNR of 42.11dB, CR-29.3, and MSE of 14.37.