X-Ray PoseNet: 6 DoF Pose Estimation for Mobile X-Ray Devices

Bui, Mai; Albarqouni, Shadi; Schrapp, Michael; Navab, Nassir; Ilić, Slobodan

doi:10.1109/wacv.2017.120

Cited by 22 publications

(17 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a consequence, while promising on synthetic results, i3PosNet closely misses the required tracking accuracy for temporal bone surgery on real X-ray data. Solving this issue is a significant CAI challenge and requires large annotated datasets mixed into the training [4] or novel methods for generation [22].…”

Section: Discussionmentioning

confidence: 99%

“…Most published research uses other tracking paradigms, most notably optical tracking [5]. While some deep learning-based approaches directly [4,16] or indirectly [6] extract instrument poses from X-ray images, neither address temporal bone surgery. In this section, we give a brief overview of instrument tracking for temporal bone surgery.…”

Section: Related Workmentioning

confidence: 99%

“…[12]. Recent work introduces deep learning methods for instrument pose estimation using segmentation as intermediate representations [6] and directly estimating the 3D pose [4,16]. While Miao et.al.…”

Section: Related Workmentioning

confidence: 99%

“…[16] employ 974 specialized neural networks, Bui et.al. 's work [4] extends the PoseNet architecture, but does not feature anatomy. Bier et.al.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

i3PosNet: instrument pose estimation from X-ray in temporal bone surgery

et al. 2020

View full text Add to dashboard Cite

Purpose Accurate estimation of the position and orientation (pose) of surgical instruments is crucial for delicate minimally invasive temporal bone surgery. Current techniques lack in accuracy and/or line-of-sight constraints (conventional tracking systems) or expose the patient to prohibitive ionizing radiation (intra-operative CT). A possible solution is to capture the instrument with a c-arm at irregular intervals and recover the pose from the image. Methods i3PosNet infers the position and orientation of instruments from images using a pose estimation network. Said framework considers localized patches and outputs pseudo-landmarks. The pose is reconstructed from pseudo-landmarks by geometric considerations. Results We show i3PosNet reaches errors $$<\,0.05$$<0.05 mm. It outperforms conventional image registration-based approaches reducing average and maximum errors by at least two thirds. i3PosNet trained on synthetic images generalizes to real X-rays without any further adaptation. Conclusion The translation of deep learning-based methods to surgical applications is difficult, because large representative datasets for training and testing are not available. This work empirically shows sub-millimeter pose estimation trained solely based on synthetic training data.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

“…[16] employ 974 specialized neural networks, Bui et.al. 's work [4] extends the PoseNet architecture, but does not feature anatomy. Bier et.al.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

i3PosNet: instrument pose estimation from X-ray in temporal bone surgery

et al. 2020

View full text Add to dashboard Cite

show abstract

“…It was tested extensively using CAD-based phantom simulations. Given the simple symmetry no extension to more demanding deep learning techniques [14,15] should be made, which are more fit for samples with lower symmetry. Using this prior knowledge on the 3D pose, the algebraic reconstruction is improved by including an initial solution that automatically has a correct orientation and that is adapted to the more convenient cylindrical coordinate system.…”

Section: Introductionmentioning

confidence: 99%

Fast Tomographic Inspection of Cylindrical Objects

2020

View full text Add to dashboard Cite

This paper presents a method for improved analysis of objects with an axial symmetry using X-ray Computed Tomography (CT). A cylindrical coordinate system about an axis fixed to the object's center of symmetry forms the most natural base to analyze characteristics of those objects that exhibit axial symmetry, as is often the case for industrial parts. As the sampling grid in the proposed methodology is chosen in reference to the object, down-sampling may be applied in directions where no good resolution is required, to reduce reconstruction time without reducing quality. This is very valuable for in-line applications and fast quality inspection. In combination with algebraic reconstruction algorithms, it permits the use of a pre-computed initial volume perfectly suited to fit a series of scans where same-type objects can have different positions and orientations. Weighted back-projection can also be included when some regions are more likely subject to change, to improve reconstruction stability and defect sensitivity. In this paper, an innovative projection-based pose estimation technique is presented and validated on realistic simulated projections that resemble a setup with high-throughput conditions. Four cylindrical reconstruction strategies are compared to observe the influence of using an initial volume or weighted back-projection on tomographic inspection of medical devices. Finally, the effect of down-sampling the cylindrical reconstruction is analysed on phantoms with varying spatial frequency.

show abstract

Automatic Plane Adjustment of Orthopedic Intraoperative Flat Panel Detector CT-Volumes

Vicario

Kordon

Denzinger

et al. 2020

Medical Image Computing and Computer Assisted Intervention – MICCAI 2020

View full text Add to dashboard Cite

Purpose 3D acquisitions are often acquired to assess the result in orthopedic trauma surgery. With a mobile C-Arm system, these acquisitions can be performed intra-operatively. That reduces the number of required revision surgeries. However, due to the operation room setup, the acquisitions typically cannot be performed such that the acquired volumes are aligned to the anatomical regions. Thus, the multiplanar reconstructed (MPR) planes need to be adjusted manually during the review of the volume. In this paper, we present a detailed study of multi-task learning (MTL) regression networks to estimate the parameters of the MPR planes.Approach First, various mathematical descriptions for rotation, including Euler angle, quaternion, and matrix representation, are revised. Then, three different MTL network architectures based on the PoseNet are compared with a single task learning network.Results Using a matrix description rather than the Euler angle description, the accuracy of the regressed normals improves from 7.7 • to 7.3 • in the mean value for single anatomies. The multi-head approach improves the regression of the plane position from 7.4mm to 6.1mm, while the orientation does not benefit from this approach.Conclusions The results show that a multi-head approach can lead to slightly better results than the individual tasks networks. The most important benefit of the MTL approach is that it is a single network for standard plane regression for all body regions with a reduced number of stored parameters.

show abstract

X-Ray PoseNet: 6 DoF Pose Estimation for Mobile X-Ray Devices

Cited by 22 publications

References 17 publications

i3PosNet: instrument pose estimation from X-ray in temporal bone surgery

i3PosNet: instrument pose estimation from X-ray in temporal bone surgery

Fast Tomographic Inspection of Cylindrical Objects

Automatic Plane Adjustment of Orthopedic Intraoperative Flat Panel Detector CT-Volumes

Contact Info

Product

Resources

About