In clinics, long-length medical images such as lower limb or whole spine are required for diagnostic purposes. Because of limited x-ray detector panel size, long-length images should be acquired by stitching of multiple images with minimum overlap area. To find the stitching point, similarity is measured in overlapped area between two serial images by calculating normalized cross correlation (NCC). Typically, medical image stitching technique introduces a lead ruler as a marker for stitching. However, marker-based stitching can result in discordance of anatomical structures near stitching line when there is a patient movement during sequential image acquisition. Therefore, anatomical-structure-based stitching is the ideal method. Nevertheless, scattering of x-rays and noise from detector panel decrease the similarity of overlapped area from two serial images, which can lead to erroneous stitching. In this study, we suggested the introduction of multi-frequency processing (MFP) framework for anatomical-structure-based-stitching to reduce the stitching error.
To reduce the calculation time, fast Fourier transform (FFT) was used. Sample images for NCC calculation were selected on the overlapped region. MFP was applied for these selected samples to get edged enhanced images. Using MFP applied samples, NCC calculation was carried out to determine stitching point. After determinant of stitching point, seam-line is eliminated using triangulation average blending.
NCC-based stitching in MFP framework outperformed the simple NCC-based stitching. Stitching in MFP framework significantly reduces the possibility of erroneous stitching.
Introduction of MFP before NCC calculation greatly increased the performance of stitching. Anatomical-structure-based stitching without a lead ruler can be successfully achieved without visible error.
본 연구에서는 NCC 기반의 의료 영상 정합에서 오류를 효과적으로 감소시키고자 MFT를 도입한다. 의료 영상을 MTF 처리하여 엣지가 부각된 영상을 얻은 후, 이를 바탕으로 NCC 계산을 하여 정합점을 도출한다. 이때, NCC 계산은 전체 영상이 아닌 정합점이 있을 것으로 추정되는 ROI 에 대해 계산하여 가속화한다. 정합점 기준으로 상∙하단 정합 부위에 triangular average 방법으로 자연스러운 연결을 한다. 실험 수행 결과, MTF 처리하지 않은 영상에 비해 MTF 도메인에서의 계산 결과가 더 정확한 정합점 도출을 해냈음을 알 수 있다. 본 연구에서는 NCC 기반의 영상 정합에 MFP를 도입했다. 실험 결과를 통해 원본 영상을 그대로 이용해 정합하는 것 보다 MFP를 도입한 경우가 더 좋은 성능을 보임을 알 수 있다. 이 방법은 NCC 이외의 다른 정합 방식에도 이용될 수 있을 것이다.
