Research Information System
Results in Engineering, vol.27, 2025 (ESCI)
Ultrasound (US) imaging is a widespread modality in medical diagnostics, preferred for its portability, non-invasiveness, and ability to provide real-time images. However, the low resolution and background speckle noise are the main drawbacks of US images. This study aimed to examine the impact of image enhancement techniques on image registration applications. The mono-modal US-US B-mode rigid image registration model was employed as the registration structure. A supervised regression convolutional neural network (CNN) was employed to determine three rigid registration parameters. As rigid registration parameters, translation and rotation parameters were investigated. Anisotropic filtering (AF), median filtering (MF), Wiener filtering (WF), histogram equalization (HE), and contrast-limited adaptive histogram equalization (CLAHE) were examined as techniques for image enhancement. Quantitative methods were employed to conduct registration experiments on four datasets. The results of the study were made both on the direct outputs of the network and on the registration of images according to the parameters obtained from the network. As a result of the study, in all datasets used in the study, the applications made to the image enhancement methods are more successful in estimating the three rigid parameters. The main contribution of this study is to enhance the effectiveness of US image registration by investigating the performance of various image enhancement techniques on rigid image registration. This is achieved by employing a convolutional neural network to accurately predict three rigid transformation parameters, thereby improving the accuracy of the registration process.