Automated Machine Learning (AutoML)-based Surface Registration Methodology for Image-guided Surgical Navigation System

BACKGROUND

While the surface registration technique has the advantage of being relatively safe and the operation time is short, it generally has the disadvantage of low accuracy.

PURPOSE

This research proposes automated machine learning (AutoML)-based surface registration to improve the accuracy of image-guided surgical navigation systems.

METHODS

The state-of-the-art surface registration concept is that first, using a neural network model, a new point-cloud that matches the facial information acquired by a passive probe of an optical tracking system (OTS) is extracted from the facial information obtained by computerized tomography (CT). Target registration error (TRE) representing the accuracy of surface registration is then calculated by applying the iterative closest point (ICP) algorithm to the newly extracted point-cloud and OTS information. In this process, the hyperparameters used in the neural network model and ICP algorithm are automatically optimized using Bayesian Optimization with Expected Improvement to yield improved registration accuracy.

RESULTS

Using the proposed surface registration methodology, the average TRE for the targets located in the sinus space and nasal cavity of the soft phantoms is (0.939 ± 0.375) mm, which shows 57.8 % improvement compared to the average TRE of (2.227 ± 0.193) mm calculated by the conventional surface registration method (p < 0.01). The performance of the proposed methodology is evaluated, and the average TREs computed by the proposed methodology and the conventional method are (0.767 ± 0.132) mm and (2.615 ± 0.378) mm, respectively. Additionally, for one healthy adult, the clinical applicability of the AutoML-based surface registration is also presented.

CONCLUSION

Our findings showed that the registration accuracy could be improved while maintaining the advantages of the surface registration technique. This article is protected by copyright. All rights reserved.

Navigation accuracy after automatic- and hybrid-surface registration in sinus and skull base surgery

Objective

Computer-aided-surgery in ENT surgery is mainly used for sinus surgery but navigation accuracy still reaches its limits for skull base procedures. Knowledge of navigation accuracy in distinct anatomical regions is therefore mandatory. This study examined whether navigation accuracy can be improved in specific anatomical localizations by using hybrid registration technique.

Study design

Experimental phantom study.

Setting

Operating room.

Subjects and methods

The gold standard of screw registration was compared with automatic LED-mask-registration alone, and in combination with additional surface matching. 3D-printer-based skull models with individual fabricated silicone skin were used for the experiments. Overall navigation accuracy considering 26 target fiducials distributed over each skull was measured as well as the accuracy on selected anatomic localizations.

Results

Overall navigation accuracy was <1.0 mm in all cases, showing the significantly lowest values after screw registration (0.66 ± 0.08 mm), followed by hybrid registration (0.83± 0.08 mm), and sole mask registration (0.92 ± 0.13 mm).On selected anatomic localizations screw registration was significantly superior on the sphenoid sinus and on the internal auditory canal. However, mask registration showed significantly better accuracy results on the midface. Navigation accuracy on skull base localizations could be significantly improved by the combination of mask registration and additional surface matching.

Conclusion

Overall navigation accuracy gives no sufficient information regarding navigation accuracy in a distinct anatomic area. The non-invasive LED-mask-registration proved to be an alternative in clinical routine showing best accuracy results on the midface. For challenging skull base procedures a hybrid registration technique is recommendable which improves navigation accuracy significantly in this operating field. Invasive registration procedures are reserved for selected challenging skull base operations where the required high precision warrants the invasiveness.

Prospects and limitations of different registration modalities in electromagnetic ENT navigation

The present study examined electromagnetic tracking technology for ENT navigation. Five different registration modalities were compared and navigation accuracy was assessed. Four skull models were individually fabricated with a three-dimensional printer, based on patients' computer tomography datasets. Individual silicone masks were fitted for skin and soft tissue simulation. Five registration modalities were examined: (1) invasive marker, (2) automatic, (3) surface matching (AccuMatch), (4) anatomic landmarks, and (5) oral splint registration. Overall navigation accuracy and accuracy on selected anatomic locations were assessed by targeting 26 titanium screws previously placed over the skull. Overall navigation accuracy differed significantly between all registration modalities. The target registration error was 0.94 ± 0.06 mm (quadratic mean ± standard deviation) for the invasive marker registration, 1.41 ± 0.04 mm for the automatic registration, 1.59 ± 0.14 mm for the surface matching registration, and 5.15 ± 0.66 mm (four landmarks) and 4.37 ± 0.73 mm (five landmarks) for the anatomic landmark registration. Oral splint registration proved itself to be inapplicable to this navigation system. Invasive marker registration was superior on most selected anatomic locations. However, on the ethmoid and sphenoid sinus the automatic registration process revealed significantly lower target registration error values. Only automatic and surface registration met the accuracy requirements for noninvasive registration. Particularly, the automatic image-to-world registration reaches target registration error values on the anterior skull base which are comparable with the gold standard of invasive screw registration.

Successful resection of anterior and anterolateral lesions at the craniovertebral junction using a simple posterolateral approach

Tumors at the craniovertebral junction (CVJ) often present a challenge due to proximity to vital neurovascular structures. In the last few decades, many authors have proposed complex surgical approaches to access pathologies located anterior or anterolateral to the CVJ with the hopes of reducing morbidity. We propose that the simple posterolateral approach in a semi-sitting position can be used to resect most anterior and anterolateral CVJ tumors safely and effectively. We retrospectively reviewed the clinical series of 10 patients treated by the senior author using the posterolateral suboccipital approach to treat anterior or anterolateral CVJ pathologies. We describe our surgical techniques, outcomes, and present illustrative patients. Gross total resection was achieved in eight patients (80%). Good functional outcome (Glasgow Outcome Scale 4-5) was obtained in all patients. Preoperative symptoms and deficits were improved (78%) or stable (22%) in all patients. There was one (10%) surgical complication that was cerebrospinal fluid leak requiring reoperation. There was no permanent morbidity or mortality in this series. There were two (20%) medical complications including deep vein thrombosis and pulmonary embolus. There were three (30%) transient neurologic complications, dysphagia in two and dysarthria in one, all of which resolved completely in early follow-up. The majority of anterior or anterolateral CVJ lesions can be successfully removed using the simple posterolateral approach.

Effect of registration mode on neuronavigation precision: an exploration of the role of random error

The aim of this paper is to analyze the variations in registration accuracy for computer-assisted surgical navigation using three different modes of registration, in order to explore the behavior of random error, and to highlight the precision of neuronavigation as a concept distinct from accuracy. The operational accuracy of three different registration modes (bone fiducials, scalp adhesive fiducials and an auto-registration mask) was evaluated in a total of 20 fresh cadaveric heads. The precision of the neuronavigation system was then assessed by evaluating the variation in the accuracy measurements associated with each registration mode. The coefficient of variation was employed to quantify the degree of variation in the attained accuracy using the following formula: Coefficient of variation = standard deviation/mean * 100. For external targets, the precision of the neuronavigation system was greatest with mask registration (43.75 and 51.41 for anterior and posterior external targets, respectively) and lowest with bone registration (65.30 and 67.17 for anterior and posterior external targets, respectively). For internal targets, the precision of the neuronavigation system was greatest with bone registration (47.69 and 42.6 for anterior and posterior internal targets, respectively) and lowest with mask registration (62.9 and 58.67 for anterior and posterior internal targets, respectively). The precision (reproducibility) of the neuronavigation system is another important quantity besides accuracy that characterizes the performance of the system. Understanding both of these quantities for a given registration mode enhances the use of a neuronavigation system in neurosurgery.