Predicting medical image registration error through independent directions
- PDF / 2,036,007 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 50 Downloads / 191 Views
ORIGINAL PAPER
Predicting medical image registration error through independent directions Gorkem Saygili1,2 Received: 1 June 2020 / Revised: 26 July 2020 / Accepted: 12 September 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Estimating registration error through independent directions, horizontal, vertical, and diagonal, is yet an unaccomplished task. If accurately done, this information can be used as feedback to further improve the registration quality. In this paper, we propose an algorithm for this purpose using a random forest regressor with features extracted using block matching. The proposed algorithm only requires two images as input and make predictions densely without requiring multiple registrations. The results on publicly available datasets show that the displacement of the best match after block matching provides strong cues about the registration error and the proposed algorithm is capable of estimating registration error through independent directions with high accuracy. Keywords Medical image registration · Registration error prediction · Random forests
1 Introduction Image registration is one of the fundamental tasks in medical image processing with the aim of aligning different images. This process is essential for applications that require extracting longitudinal information [1], segmenting anatomical regions [2], using intra-operative image guidance [4], and planning a radiotherapy [5]. Extensive survey about medical image registration algorithms and their applications can be found in [6]. Recently, researchers have proposed several accurate registration algorithms such as the ones based on highly optimized numerical methods [7] and deep neural networks [8,9]. Yet, these algorithms may still fail to achieve perfect alignment, making it important to find these misalignments for either informing the expert or providing feedback to the registration algorithm. Considerable effort has been spent to find a reliable way of estimating the registration error either directly or as a measure of uncertainty. Common approaches use similarity measures [12], residual images [13], Hausdorff distance [14]
B
Gorkem Saygili [email protected]
1
Department of Biomedical Engineering, Faculty of Engineering, Ankara University, Ankara, Turkey
2
Department of Interdisciplinary Neuroscience, Health Science Institute, Ankara University, Ankara, Turkey
and gradient of registration cost scores [15] as indicators of misalignment. Another approach is using bootstrap resampling to generate confidence measures [16]. Kybic proposed a faster algorithm that can achieve better performance than bootstrap resampling [17]. Additionally, triplet registrations that can be used with both rigid [18] and nonrigid transformations [19] can also indicate registration accuracy. Machine learning approaches have been used increasingly to classify and predict registration errors. Sofka et al. [20] identified locations with acceptable accuracy inside a registered image using an Support Vector Machi
Data Loading...
