Advanced Medical Image Reconstruction Using the Dual Encoder Split Path Autoencoder (DESPAE) Architecture

Image reconstruction is critical in medical imaging, where accurate data restoration is essential for precise analysis and diagnosis. This research proposes an innovative architectural framework for medical image reconstruction, termed the Dual Encoder Split Path Autoencoder (DESPAE). DESPAE uses autoencoder networks and segmentation strategies to enhance reconstruction quality by dividing input medical images into two splits, each processed by dedicated encoder networks. This division captures diverse, intrinsic characteristics specific to different regions of medical images, which are often complex and heterogeneous. The resulting separate latent spaces are seamlessly integrated into a unified representation, enabling efficient and high-fidelity reconstruction by the decoder network. Experimental evaluations on the NIH Chest X-ray dataset demonstrate the promising performance of DESPAE, validated by Peak Signal-to-Noise Ratio (PSNR) metric. The model achieved an impressive PSNR of 47.29 dB after 100 epochs, significantly improving from an initial 37.7 dB. Moreover, DESPAE outperformed prominent autoencoder architectures, including the Deep Autoencoder with multiple Backpropagation (DA-MBP), Deep Autoencoder with Restricted Boltzmann Machine (DA-RBM), Deep Convolutional Autoencoder with Restricted Boltzmann Machine (DCA-RBM), and Deep Autoencoder with DPM (DA-DPM). This study advances medical image reconstruction methodologies by offering a novel solution that enhances reconstruction quality, thereby contributing to the ongoing development of effective techniques in the field of image processing.