The task of exchanging medical images with high resolution and low size, to ensure the correct medical diagnosis, is an essential goal for healthcare centers, especially those that adopt the DICOM standard, one of the most widespread and advanced standards in the field of telemedicine.
In this paper, we seek to test a learnable mechanism for DICOM digital medical image compression to provide high resolution image reconstruction, taking advantage of the L3C system architecture used to compress natural images, by modifying and training it to fit this standard, and integrating it with hybrid models in deep learning to help improve accuracy and raising the efficiency of the prediction task during image reconstruction.
A mechanism was built to extract image pixel arrays from DICOM files for training based on a large dataset, modify the recessive networks structure in the EDSR model to include a batch normalization layer and Leaky-Relu activation function for their most useful training benefits, and add a network that combines the ESPCN model and the dual channel pixel attention mechanism to provide a deeper understanding of the relationship between the components of the image and improve the effect of its recovery while reducing the computational complexity, then employ a developed ASPP model that uses dense connections (Dense ASPP) to expand the receptive field and increase the effectiveness of dense prediction.
Research Journal of Aleppo University.
2022.
Students
Teachers
Graduates
Faculties
