State of Robotic Mastoidectomy: Literature Review

Training and validation of a deep learning U-net architecture general model for automated segmentation of inner ear from CT

BACKGROUND

The intricate three-dimensional anatomy of the inner ear presents significant challenges in diagnostic procedures and critical surgical interventions. Recent advancements in deep learning (DL), particularly convolutional neural networks (CNN), have shown promise for segmenting specific structures in medical imaging. This study aimed to train and externally validate an open-source U-net DL general model for automated segmentation of the inner ear from computed tomography (CT) scans, using quantitative and qualitative assessments.

METHODS

In this multicenter study, we retrospectively collected a dataset of 271 CT scans to train an open-source U-net CNN model. An external set of 70 CT scans was used to evaluate the performance of the trained model. The model's efficacy was quantitatively assessed using the Dice similarity coefficient (DSC) and qualitatively assessed using a 4-level Likert score. For comparative analysis, manual segmentation served as the reference standard, with assessments made on both training and validation datasets, as well as stratified analysis of normal and pathological subgroups.

RESULTS

The optimized model yielded a mean DSC of 0.83 and achieved a Likert score of 1 in 42% of the cases, in conjunction with a significantly reduced processing time. Nevertheless, 27% of the patients received an indeterminate Likert score of 4. Overall, the mean DSCs were notably higher in the validation dataset than in the training dataset.

CONCLUSION

This study supports the external validation of an open-source U-net model for the automated segmentation of the inner ear from CT scans.

RELEVANCE STATEMENT

This study optimized and assessed an open-source general deep learning model for automated segmentation of the inner ear using temporal CT scans, offering perspectives for application in clinical routine. The model weights, study datasets, and baseline model are worldwide accessible.

KEY POINTS

A general open-source deep learning model was trained for CT automated inner ear segmentation. The Dice similarity coefficient was 0.83 and a Likert score of 1 was attributed to 42% of automated segmentations. The influence of scanning protocols on the model performances remains to be assessed.

Feasibility of Robotic Transorbital Surgery

BACKGROUND AND OBJECTIVES

The transorbital approach (TOA) facilitates access to pathologies lateral to the optic nerve, a region that is difficult to access with an endonasal approach. In this study, we sought to investigate the feasibility of robotic-assisted surgery in lateral TOA.

METHODS

Six colored-silicon-injected human postmortem heads were prepared for dissection. The DaVinci Xi model was used with a 0-degree camera, 8 mm in diameter. A black diamond microforceps with an 8-mm diameter and 10-mm jaw length was used. The entry point of V1 (superior orbital fissure), V3 (foramen ovale), and posterior root of the trigeminal ganglion were chosen as the surgical targets. The length from the entry opening to each target point was measured. The angles formed between pairs of target points were measured to obtain the horizontal angle (root of the trigeminal ganglion-entry-V1) and the vertical angle (root of the trigeminal ganglion-entry-V3).

RESULTS

Dissection was performed on 12 sides (6 specimens). The median distance from the entry point was 55 mm (range 50-58 mm) to the entry point of V1 (superior orbital fissure), 65 mm (range 57-70 mm) to the entry point of V3 (foramen ovale), and 76 mm (range 70-87 mm) to the root of the trigeminal ganglion. Meanwhile, the median of surgical angle between the entry point and the target was 19.1° (range 11.8-30.4°) on the horizontal angle and 16.5° (range 6.2-21.6°) on the vertical angle.

CONCLUSION

This study found that application of lateral TOA in robotic-assisted surgery is premature because of the large size of the tool. However, although the entrance in lateral TOA is narrow, the internal surgical space is wide; this offers potential for design of appropriate surgical tools to allow increase tool usage.

Pilot Study of Optical Topographic Imaging Based Neuronavigation for Mastoidectomy

BACKGROUND

Mastoidectomy involves drilling the temporal bone while avoiding the facial nerve, semicircular canals, sigmoid sinus, and tegmen. Optical topographic imaging (OTI) is a novel registration technique that allows rapid registration with minimal navigational error. To date, no studies have examined the use of OTI in skull-base procedures.

METHODS

In this cadaveric study, 8 mastoidectomies were performed in 2 groups-4 free-hand (FH) and 4 OTI-assisted mastoidectomies. Registration accuracy for OTI navigation was quantified with root mean square (RMS) and target registration error (TRE). Procedural time, percent of mastoid resected, and the proximity of the mastoidectomy cavity to critical structures were determined.

RESULTS

The average RMS and TRE associated with OTI-based registration were 1.44 mm (±0.83 mm) and 2.17 mm (±0.89 mm), respectively. The volume removed, expressed as a percentage of the total mastoid volume, was 37.5% (±10.2%) versus 31.2% (±2.3%), P = 0.31, for FH and OTI-assisted mastoidectomy. There were no statistically significant differences between FH and OTI-assisted mastoidectomies with respect to proximity to critical structures or procedural time.

CONCLUSIONS

This work is the first examining the application of OTI neuronavigation in lateral skull-base procedures. This pilot study revealed the RMS and TRE for OTI-based navigation in the lateral skull base are 1.44 mm (±0.83 mm) and 2.17 mm (±0.89 mm), respectively. This pilot study demonstrates that an OTI-based system is sufficiently accurate and may address barriers to widespread adoption of navigation for lateral skull-base procedures.

Image-Guided Mastoidectomy with a Cooperatively Controlled ENT Microsurgery Robot

Mastoidectomy is a common surgical procedure within otology. Despite being inherently well suited for implementation of robotic assistance, there are no commercially available robotic systems that have demonstrated utility in aiding with this procedure. This article describes a robotic technique for image-guided mastoidectomy with an experimental cooperatively controlled robotic system developed for use within otolaryngology-head and neck surgery. It has the ability to facilitate enhanced operative precision with dampening of tremor in simulated surgical tasks. Its kinematic design is such that the location of the attached surgical instrument is known with a high degree of fidelity at all times. This facilitates image registration and subsequent definition of virtual fixtures, which demarcate surgical workspace boundaries and prevent motion into undesired areas. In this preliminary feasibility study, we demonstrate the clinical utility of this system to facilitate performance of a cortical mastoidectomy by a novice surgeon in 5 identical temporal bone models with a mean time of 221 ± 35 seconds.

Robotic laser osteotomy through penscriptive structured light visual servoing

PURPOSE

Planning osteotomies is a task that surgeons do as part of standard surgical workflow. This task, however, becomes more difficult and less intuitive when a robot is tasked with performing the osteotomy. In this study, we aim to provide a new method for surgeons to allow for highly intuitive trajectory planning, similar to the way an attending surgeon would instruct a junior.

METHODS

Planning an osteotomy, especially during a craniotomy, is performed intraoperatively using a sterile surgical pen or pencil directly on the exposed bone surface. This paper presents a new method for generating osteotomy trajectories for a multi-DOF robotic manipulator using the same method and relaying the penscribed cut path to the manipulator as a three-dimensional trajectory. The penscribed cut path is acquired using structured light imaging, and detection, segmentation, optimization and orientation generation of the Cartesian trajectory are done autonomously after minimal user input.

RESULTS

A 7-DOF manipulator (KUKA IIWA) is able to follow fully penscribed trajectories with sub-millimeter accuracy in the target plane and perpendicular to it (0.46 mm and 0.36 mm absolute mean error, respectively).

CONCLUSIONS

The robot is able to precisely follow cut paths drawn by the surgeon directly onto the exposed boney surface of the skull. We demonstrate through this study that current surgical workflow does not have to be drastically modified to introduce robotic technology in the operating room. We show that it is possible to guide a robot to perform an osteotomy in much the same way a senior surgeon would show a trainee by using a simple surgical pen or pencil.