DEEP LEARNING-BASED RECONSTRUCTION AND ENHANCEMENT ALGORITHMS FOR NOISE AND ARTIFACT REDUCTION IN ECHOCARDIOGRAPHIC IMAGES
DOI:
https://doi.org/10.65164/jpf54e63Keywords:
deep learning, echocardiography, image enhancement, noise reduction, convolutional neural networks, speckle noise, cardiac imaging, image reconstruction.Abstract
This article presents a comprehensive methodology for developing deep learningbased
reconstruction and enhancement algorithms specifically designed for noise suppression and
artifact reduction in echocardiographic images. The proposed framework integrates a modified UNet
architecture with attention mechanisms and generative adversarial networks (GANs) to address
the inherent challenges of speckle noise, acoustic shadowing, and reverberation artifacts in ultrasound
cardiac imaging. A novel loss function combining structural similarity index (SSIM), perceptual loss,
and adversarial loss was developed to preserve myocardial structural details while effectively
removing noise. The dataset comprising 4,800 echocardiographic video sequences from 1,200
patients was utilized for training and validation. The proposed method achieved a peak signal-tonoise
ratio (PSNR) of 38.4 dB and structural similarity index (SSIM) of 0.94, representing significant
improvements over conventional filtering techniques. Clinical validation demonstrated enhanced
endocardial border detection accuracy by 23% and improved ejection fraction calculation precision.
The developed algorithms provide a robust foundation for automated cardiac functional assessment
and computer-aided diagnostic systems in cardiovascular medicine.
References
[1]. Lattanzi, A., et al. (2021). Deep learning for noise reduction in echocardiographic imaging: A
systematic review and meta-analysis. IEEE Transactions on Medical Imaging, 40(8), 2134-
2148.
[2]. Chen, J., et al. (2022). Attention-gated networks for medical image enhancement: Applications
in cardiac ultrasound. Computerized Medical Imaging and Graphics, 95, 102982.
[3]. Zhang, Y., et al. (2023). Generative adversarial networks for speckle noise reduction in
ultrasound images: A comparative study. Medical Image Analysis, 79, 102476.
[4]. Wang, S., et al. (2024). Real-time deep learning processing for smart hospital imaging pipelines:
Architectures and implementations. Journal of Medical Systems, 48(3), 45.
[5]. ISO 13485:2016. Medical devices — Quality management systems — Requirements for
regulatory purposes. Geneva: ISO, 2016.
