Title: Near-lossless medical image compression using wavelet subband thresholding and convolutional autoencoder

Authors: Muthalaguraja Venugopal; Kalavathi Palanisamy; Punitha Viswanathan

Addresses: Department of Computer Science and Applications, The Gandhigram Rural Institute (Deemed to be University), Gandhigram, Tamil Nadu, India ' Department of Computer Science and Applications, The Gandhigram Rural Institute (Deemed to be University), Gandhigram, Tamil Nadu, India ' Department of Computer Science and Applications, The Gandhigram Rural Institute (Deemed to be University), Gandhigram, Tamil Nadu, India

Abstract: In the digital era of healthcare, telemedicine services are continuously evolving. Medical images with massive file sizes increase storage and transmission complexity while providing telemedicine services. To address this, image compression becomes obligatory. Learning-based methods show promising results in image compression tasks. However, the problem of maintaining image reconstruction quality still needs to be addressed. This work proposes a wavelet-based convolutional autoencoder for near-lossless medical image compression. Thresholded wavelet subbands of medical images were used to train the compression model. A convolutional encoder-decoder model with a simple encoder network and an extended decoder network is proposed to achieve a near-lossless image compression standard. A combined loss function is employed to improve the model's reconstruction performance. The combined loss function includes mean squared error and structural similarity index metric, focusing on image reconstruction quality. Extensive experiments show the efficiency of the proposed method over the existing image compression techniques.

Keywords: image compression; CNN; wavelet; convolutional autoencoder; CAE; deep learning; medical image compression.

DOI: 10.1504/IJCSE.2025.146086

International Journal of Computational Science and Engineering, 2025 Vol.28 No.3, pp.329 - 345

Received: 06 Nov 2023
Accepted: 31 May 2024
Published online: 06 May 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article