A navigation system for flexible endoscopes using abdominal 3D ultrasound

Feature matching for texture-less endoscopy images via superpixel vector field consistency

Feature matching is an important technology to obtain the surface morphology of soft tissues in intraoperative endoscopy images. The extraction of features from clinical endoscopy images is a difficult problem, especially for texture-less images. The reduction of surface details makes the problem more challenging. We proposed an adaptive gradient-preserving method to improve the visual feature of texture-less images. For feature matching, we first constructed a spatial motion field by using the superpixel blocks and estimated its information entropy matching with the motion consistency algorithm to obtain the initial outlier feature screening. Second, we extended the superpixel spatial motion field to the vector field and constrained it with the vector feature to optimize the confidence of the initial matching set. Evaluations were implemented on public and undisclosed datasets. Our method increased by an order of magnitude in the three feature point extraction methods than the original image. In the public dataset, the accuracy and F1-score increased to 92.6% and 91.5%. The matching score was improved by 1.92%. In the undisclosed dataset, the reconstructed surface integrity of the proposed method was improved from 30% to 85%. Furthermore, we also presented the surface reconstruction result of differently sized images to validate the robustness of our method, which showed high-quality feature matching results. Overall, the experiment results proved the effectiveness of the proposed matching method. This demonstrates its capability to extract sufficient visual feature points and generate reliable feature matches for 3D reconstruction and meaningful applications in clinical.

Evaluation of 3D ultrasound for image guidance

PURPOSE

In this paper we compared two different 3D ultrasound (US) modes (3D free-hand mode and 3D wobbler mode) to see which is more suitable to perform the 3D-US/3D-US registration for clinical guidance applications. The typical errors with respect to their impact on the final localization error were evaluated step by step.

METHODS

Multi-point target and Hand-eye calibration methods were used for 3D US calibration together with a newly designed multi-cone phantom. Pointer based and image based methods were used for 2D US calibration. The calibration target error was computed by using a different multi-cone phantom. An egg-shaped phantom was used as ground truth to compare distortions for both 3D modes along with the measurements of the volume. Finally, we compared 3D ultrasound images acquired by 3D wobbler mode and 3D free-hand mode with respect to their 3D-US/3D-US registration accuracy using both, phantom and patient data. A theoretical step by step error analysis was performed and compared to empirical data.

RESULTS

Target registration errors based on the calibration with the 3D Multi-point and 2D pointer/image method have been found to be comparable (∼1mm). They both outperformed the 3D Hand-eye method (error >2mm). Volume measurements with the 3D free-hand mode were closest to the ground truth (around 6% error compared to 9% with the 3D wobbler mode). Additional scans on phantoms showed a 3D-US/3D-US registration error below 1 mm for both, the 3D free-hand mode and the 3D wobbler mode, respectively. Results with patient data showed greater error with the 3D free-hand mode (6.50mm - 13.37mm) than with the 3D wobbler mode (2.99 ± 1.54 mm). All the measured errors were found to be in accordance to their theoretical upper bounds.

CONCLUSION

While both 3D volume methods showed comparable results with respect to 3D-US/3D-US registration for phantom images, for patient data registrations the 3D wobbler mode is superior to the 3D free-hand mode. The effect of all error sources could be estimated by theoretical derivations.