Towards Fully Automatic 2D Us to 3D CT/MR Registration: A Novel Segmentation-Based Strategy

Wei, Wei; Rak, Marko; Alpers, Julian; Hansen, Christian

doi:10.1109/ISBI45749.2020.9098379

by Wei Wei, Marko Rak, Julian Alpers, Christian Hansen

Abstract:

2D-US to 3D-CT/MR registration is a crucial module during minimally invasive ultrasound-guided liver tumor ablations. Many modern registration methods still require manual or semi-automatic slice pose initialization due to insufficient robustness of automatic methods. The state-of-the-art regression networks do not work well for liver 2D US to 3D CT/MR registration because of the tremendous inter-patient variability of the liver anatomy. To address this unsolved problem, we propose a deep learning network pipeline which - instead of a regression - starts with a classification network to recognize the coarse ultrasound transducer pose followed by a segmentation network to detect the target plane of the US image in the CT/MR volume. The rigid registration result is derived using plane regression. In contrast to the state-of-the-art regression networks, we do not estimate registration parameters from multi-modal images directly, but rather focus on segmenting the target slice plane in the volume. The experiments reveal that this novel registration strategy can identify the initial slice phase in a 3D volume more reliably than the standard regression-based techniques. The proposed method was evaluated with 1035 US images from 52 patients. We achieved angle and distance errors of 12.7 ± 6.2° and 4.9 ± 3.1 mm, clearly outperforming state-of-the-art regression strategy which results in 37.0 ± 15.6° angle error and 19.0 ± 11.6 mm distance error.

Reference:

Towards Fully Automatic 2D Us to 3D CT/MR Registration: A Novel Segmentation-Based Strategy (Wei Wei, Marko Rak, Julian Alpers, Christian Hansen), In 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), 2020.

Bibtex Entry:

@inproceedings{wei_towards_2020,
	address = {Iowa City, Iowa, USA},
	title = {Towards {Fully} {Automatic} 2D {Us} to 3D {CT}/{MR} {Registration}: {A} {Novel} {Segmentation}-{Based} {Strategy}},
	isbn = {1945-8452},
	doi = {10.1109/ISBI45749.2020.9098379},
	abstract = {2D-US to 3D-CT/MR registration is a crucial module during minimally invasive ultrasound-guided liver tumor ablations. Many modern registration methods still require manual or semi-automatic slice pose initialization due to insufficient robustness of automatic methods. The state-of-the-art regression networks do not work well for liver 2D US to 3D CT/MR registration because of the tremendous inter-patient variability of the liver anatomy. To address this unsolved problem, we propose a deep learning network pipeline which - instead of a regression - starts with a classification network to recognize the coarse ultrasound transducer pose followed by a segmentation network to detect the target plane of the US image in the CT/MR volume. The rigid registration result is derived using plane regression. In contrast to the state-of-the-art regression networks, we do not estimate registration parameters from multi-modal images directly, but rather focus on segmenting the target slice plane in the volume. The experiments reveal that this novel registration strategy can identify the initial slice phase in a 3D volume more reliably than the standard regression-based techniques. The proposed method was evaluated with 1035 US images from 52 patients. We achieved angle and distance errors of 12.7 ± 6.2° and 4.9 ± 3.1 mm, clearly outperforming state-of-the-art regression strategy which results in 37.0 ± 15.6° angle error and 19.0 ± 11.6 mm distance error.},
	booktitle = {2020 {IEEE} 17th {International} {Symposium} on {Biomedical} {Imaging} ({ISBI})},
	author = {Wei, Wei and Rak, Marko and Alpers, Julian and Hansen, Christian},
	month = apr,
	year = {2020},
	keywords = {biomedical ultrasonics, classification network, coarse ultrasound transducer, crucial module, CT/MR, deep learning network pipeline, Estimation, fully automatic 2D us, image registration, Image segmentation, initial slice phase, insufficient robustness, learning (artificial intelligence), liver, liver 2D US, liver anatomy, medical image processing, minimally invasive ultrasound-guided liver tumor ablations, modern registration methods, multimodal images, plane regression, Registration, registration parameters, regression analysis, rigid registration result, segmentation-based strategy, semiautomatic slice, size 1.8000000000000007 mm to 8.0 mm, size 7.400000000000002 mm to 30.599999999999998 mm, standard regression-based techniques, state-of-the-art regression networks, state-of-the-art regression strategy, target slice plane, Three-dimensional displays, Training, Transducers, tremendous inter-patient variability, tumours, Two dimensional displays, US, US images},
	pages = {433--437}
}