Can video super resolution methods via deep learning enhance resolution of consecutive ultrasound images?
Biomedical Signal Processing and Control, cilt.126, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 126
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.bspc.2026.110951
- Dergi Adı: Biomedical Signal Processing and Control
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, EMBASE
- Anahtar Kelimeler: Deep neural Network, Ultrasound Imaging, Video super resolution
- Gazi Üniversitesi Adresli: Evet
Özet
This study examines the utilization of deep learning models in video super-resolution (VSR) to enhance the ultrasound (US) image resolution. VSR models, which leverage information from adjacent frames, could be an alternative to single-image super-resolution methods in US imaging, because there is a temporal correlation between successive US frames. To this end, the performance of six distinct VSR deep neural network (DNN) models was evaluated. These models can be categorized into those employing motion estimation and compensation and those without alignment. The lack of a suitable dataset for VSR in ultrasound imaging led us to gather an in vivo dataset from the volunteer's thyroid gland. Five performance metrics were used to evaluate the model performance. The results demonstrate that the Super-Resolution Optical Flow for Video Super-Resolution (SOF-VSR) model outperforms other models assessed in this study and is capable of maintaining both pixel-level accuracies of 43.4042 and 31.7061, respectively, for mean square error (MSE) and peak signal to noise ratio (PSNR) values and structural consistency with 0.8242, 0.9239, and 0.9464 values for structural similarity index method (SSIM), feature similarity index method (FSIM), and multi-scale structural similarity index (MS-SSIM), respectively. The results were followed by those of the RRN and DSMC models. Although the DSMC model outperforms pixel-level accuracy, RRN prevails from a structural consistency point of view. In addition, the current findings are benchmarked against previous studies on single-image super-resolution and reveal that VSR models significantly enhance the quality of US images.