Fast 5DOF needle tracking in iOCT

Abstract

Using solely cross-sectional iOCT information, we are able to successfully and robustly estimate a 5DOF pose of the instrument in less than 5.4 ms on a CPU.

Optische Technologien, digital unterstützt: Operationsmikroskop – quo vadis?

“…A feedback loop and integration of manipulation and visualisation may allow exploration of new treatments here, such as retinal vessel cannulation and controlled subretinal injections of gene therapeutics. [23] In addition to linking the surgical microscope with biometrics and data management systems for OR planning and documentation (OR note) already discussed above, added value will be generated in the future by linking the surgical microscope laterally (with other devices in the OR) as well as vertically (e.g., with the cloud).…”

Robotic Retinal Surgery

“…Perhaps even more promising is the use of high-resolution OCT information at [160,200]. Such combined multi-modal instrument tracking approaches may be the key for precise intraocular tracking of surgical instruments.…”

“…However, these methods require volumetric data sets, and do not work in real time. To address this shortcoming, Weiss et al [200] presented a technique that uses five parallel iOCT B-scans to track the needle, detecting the elliptical section of the needle in each scan. The same group has also presented a similar technique, using a larger number of B-scans (128 along x, and the same number along y), to perform marker-free robot hand-eye calibration, which they have demonstrated in intact porcine eyes ex vivo [220].…”

“…The same group has also presented a similar technique, using a larger number of B-scans (128 along x, and the same number along y), to perform marker-free robot hand-eye calibration, which they have demonstrated in intact porcine eyes ex vivo [220]. Tracking of the needle after it enters tissue remains an open research problem [200], which the group has begun to address by registering the needle to a computer-aided design (CAD) model before the tip enters the retina, and then predicting the tip position and orientation after subretinal entry using the known input commands [221].…”

6DOF Needle Pose Estimation for Robot-Assisted Vitreoretinal Surgery

“…However, they focused on integrating the OCT into the surgical system, rather than estimating the instrument pose. Weiss et al [22] developed an algorithm to track the needle 5DOF pose without needle rotation information. Gessert et al [23] proposed a 3D convolutional neural network (CNN) to segment and estimate the pose of a small marker geometry from OCT volumes directly with a mean error of 14.89 ± 9.3 µm.…”