The application of augmented reality–based navigation for accurate target acquisition of deep brain sites: advances in neurosurgical guidance

OBJECTIVE The objective of this study is to quantify the navigational accuracy of an advanced augmented reality (AR)–based guidance system for neurological surgery, biopsy, and/or other minimally invasive neurological surgical procedures. METHODS Five burr holes were drilled through a plastic cranium, and 5 optical fiducials (AprilTags) printed with CT-visible ink were placed on the frontal, temporal, and parietal bones of a human skull model. Three 0.5-mm-diameter targets were mounted in the interior of the skull on nylon posts near the level of the tentorium cerebelli and the pituitary fossa. The skull was filled with ballistic gelatin to simulate brain tissue. A CT scan was taken and virtual needle tracts were annotated on the preoperative 3D workstation for the combination of 3 targets and 5 access holes (15 target tracts). The resulting annotated study was uploaded to and launched by VisAR software operating on the HoloLens 2 holographic visor by viewing an encrypted, printed QR code assigned to the study by the preoperative workstation. The DICOM images were converted to 3D holograms and registered to the skull by alignment of the holographic fiducials with the AprilTags attached to the skull. Five volunteers, familiar with the VisAR, used the software/visor combination to navigate an 18-gauge needle/trocar through the series of burr holes to the target, resulting in 70 data points (15 for 4 users and 10 for 1 user). After each attempt the needle was left in the skull, supported by the ballistic gelatin, and a high-resolution CT was taken. Radial error and angle of error were determined using vector coordinates. Summary statistics were calculated individually and collectively. RESULTS The combined angle of error of was 2.30° ± 1.28°. The mean radial error for users was 3.62 ± 1.71 mm. The mean target depth was 85.41 mm. CONCLUSIONS The mean radial error and angle of error with the associated variance measures demonstrates that VisAR navigation may have utility for guiding a small needle to neural lesions, or targets within an accuracy of 3.62 mm. These values are sufficiently accurate for the navigation of many neurological procedures such as ventriculostomy.

Quiet eye studies in sport within the motor accuracy and motor error paradigms

This paper reveals new insights that comes from comparing quiet eye (QE) studies within the motor accuracy and motor error paradigms. Motor accuracy is defined by the rules of the sport (e.g,. hits versus misses), while motor error is defined by a behavioral measure, such as how far a ball or other object lands from the target (e.g. radial error). The QE motor accuracy paradigm treats accuracy as an independent variable and determines the QE duration during an equal (or near-equal) number of hits and misses per condition per participant, while the motor error QE paradigm combines hits and misses into one data set and determines the correlation between the QE and motor error, which is used as a proxy for accuracy. QE studies within the motor accuracy paradigm consistently find a longer QE duration is a characteristic of skill, and/or interaction of skill by accuracy. In contrast, QE motor error studies do not analyze or report the relationship between the QE duration and accuracy (although often claimed), and rarely find a significant correlation between the QE duration and error. Evidence is provided showing the absence of significant results in QE motor error studies is due to the low number of accurate trials found in motor error studies due to the inherent complexity of all sport skills. Novices in targeting skills make fewer than 20% of their shots and experts less than 40% (with some exceptions) creating imbalanced data sets that make it difficult, if not impossible, to find significant QE results (or any other neural, perceptual or cognitive variable) related to motor accuracy in sport.

Expecting to teach a novel golf putting task did not enhance retention performance: A replication attempt of Daou and colleagues

While research has identified several practice variables that purportedly enhance motor learning, recent replication failures highlight the importance of conducting high-powered, pre-registered replications. The "expecting to teach" phenomenon was first reported in the motor learning literature by Daou and colleagues and suggested learners benefit from practicing with the understanding they will later need to teach the skill. The extant data have been mixed but generally positive. While expecting to teach has been shown to enhance motor learning of a golf putt, the mechanisms linked with this benefit are yet to be determined. As such, this study sought to replicate the expecting to teach effect and to extend those findings by exploring participants’ thought processes. Participants (N = 76) were randomly assigned to one of two groups in which they were told that they were learning a golf putt in order to 1) be tested on the skill or 2) teach the skill to another individual. On Day 1, participants completed pre-test putts, a pre-acquisition intrinsic motivation inventory (IMI), a 2-minute study of an instructional booklet, 50 practice putts and a post-acquisition IMI. During practice, participants were also afforded opportunities to continue studying the booklet and to complete additional putts. Participants returned 24 hours later to complete a retention, a transfer (50-cm longer golf-putt), and a free recall test, as well as a post-study survey to reveal thoughts they engaged in after practice but before (or during) the retention test. Similar to Daou et al., no significant differences were found with study time, number of acquisition putts, or motivation. However, golf-putting performance during retention resulted in no differences for radial error, g = -0.13 (95% C.I. [-0.55, 0.29]), between the two groups and no differences were shown for the recall test. The present study fails to replicate the benefits reported in the original experiments.

Asleep deep brain stimulation with intraoperative magnetic resonance guidance: a single-institution experience

OBJECTIVE Deep brain stimulation (DBS) is traditionally performed on an awake patient with intraoperative recordings and test stimulation. DBS performed under general anesthesia with intraoperative MRI (iMRI) has demonstrated high target accuracy, reduced operative time, direct confirmation of target placement, and the ability to place electrodes without cessation of medications. The authors describe their initial experience with using iMRI to perform asleep DBS and discuss the procedural and radiological outcomes of this procedure. METHODS All DBS electrodes were implanted under general anesthesia by a single surgeon by using a neuronavigation system with 3-T iMRI guidance. Clinical outcomes, operative duration, complications, and accuracy were retrospectively analyzed. RESULTS In total, 103 patients treated from 2015 to 2019 were included, and all but 1 patient underwent bilateral implantation. Indications included Parkinson’s disease (PD) (65% of patients), essential tremor (ET) (29%), dystonia (5%), and refractory epilepsy (1%). Targets included the globus pallidus pars internus (12.62% of patients), subthalamic nucleus (56.31%), ventral intermedius nucleus of the thalamus (30%), and anterior nucleus of the thalamus (1%). Technically accurate lead placement (radial error ≤ 1 mm) was obtained for 98% of leads, with a mean (95% CI) radial error of 0.50 (0.46–0.54) mm; all leads were placed with a single pass. Predicted radial error was an excellent predictor of real radial error, underestimating real error by only a mean (95% CI) of 0.16 (0.12–0.20) mm. Accuracy remained high irrespective of surgeon experience, but procedure time decreased significantly with increasing institutional and surgeon experience (p = 0.007), with a mean procedure duration of 3.65 hours. Complications included 1 case of intracranial hemorrhage (asymptomatic) and 1 case of venous infarction (symptomatic), and 2 patients had infection at the internal pulse generator site. The mean ± SD voltage was 2.92 ± 0.83 V bilaterally at 1-year follow-up. Analysis of long-term clinical efficacy demonstrated consistent postoperative improvement in clinical symptoms, as well as decreased drug doses across all indications and follow-up time points, including mean decrease in levodopa-equivalent daily dose by 53.57% (p < 0.0001) in PD patients and mean decrease in primidone dose by 61.33% (p < 0.032) in ET patients at 1-year follow-up. CONCLUSIONS A total of 205 leads were placed in 103 patients by a single surgeon under iMRI guidance with few operative complications. Operative time trended downward with increasing institutional experience, and technical accuracy of radiographic lead placement was consistently high. Asleep DBS implantation with iMRI appears to be a safe and effective alternative to standard awake procedures.

Expecting to teach a novel golf putting task did not enhance retention performance: A replication attempt of Daou and colleagues (2016;2016)

While research has identified several practice variables that purportedly enhance motor learning, recent replication failures highlight the importance of conducting high-powered, pre-registered replications. The “expecting to teach” phenomenon was first reported in the motor learning literature by Daou and colleagues and suggested learners benefit from practicing with the understanding they will later need to teach the skill. The extant data have been mixed but generally positive. While expecting to teach has been shown to enhance motor learning of a golf putt, the mechanisms linked with this benefit are yet to be determined. As such, this study sought to replicate the expecting to teach effect and to extend those findings by exploring participants’ thought processes. Participants (N = 76) were randomly assigned to one of two groups in which they were told that they were learning a golf putt in order to 1) be tested on the skill or 2) to teach the skill to another individual. On Day 1, participants completed pre-test putts, a pre-acquisition intrinsic motivation inventory (IMI), a 2-minute study of an instructional booklet, 50 practice putts and a post-acquisition IMI. During practice, participants were also afforded opportunities to continue studying the booklet and to complete additional putts. Participants returned 24-hours later to complete a retention, a transfer (50 cm longer golf-putt), and a free recall test, as well as a post-study survey to reveal thoughts they engaged in after practice but before (or during) the retention test. Similar to Daou et al., no significant differences were found with study time, number of acquisition putts, or motivation. However, golf-putting performance during retention resulted in no differences for radial error, g = −.13 (95%CI [−.55, .29]), between the two groups and no differences were shown for the recall test. The present study fails to replicate the benefits reported in the original experiments.

The Poised Cannula Technique Reduces the Stereotactic Error of the Fiducial-Less Frameless DBS Procedure

<b><i>Background:</i></b> In this study, we describe a technique of optimizing the accuracy of frameless deep brain stimulation (DBS) lead placement through the use of a cannula poised at the entry to predict the location of the fully inserted device. This allows real-time correction of error prior to violation of the deep gray matter. <b><i>Methods:</i></b> We prospectively gathered data on radial error during the operative placements of 40 leads in 28 patients using frameless fiducial-less DBS surgery. Once the Nexframe had been aligned to target, a cannula was inserted through the center channel of the BenGun until it traversed the pial surface and a low-dose O-arm spin was obtained. Using 2 points along the length of the imaged cannula, a trajectory line was projected to target depth. If lead location could be improved, the cannula was inserted through an alternate track in the BenGun down to target depth. After intraoperative microelectrode recording and clinical assessment, another O-arm spin was obtained to compare the location of the inserted lead with the location predicted by the poised cannula. <b><i>Results:</i></b> The poised cannula projection and the actual implant had a mean radial discrepancy of 0.75 ± 0.64 mm. The poised cannula projection identified potentially clinically significant errors (avg 2.07 ± 0.73 mm) in 33% of cases, which were reduced to a radial error of 1.33 ± 0.66 mm (<i>p</i> = 0.02) after correction using an alternative BenGun track. The final target to implant error for all 40 leads was 1.20 ± 0.52 mm with only 2.5% of errors being &#x3e;2.5 mm. <b><i>Conclusion:</i></b> The poised cannula technique results in a reduction of large errors (&#x3e;2.5 mm), resulting in a decline in these errors to 2.5% of implants as compared to 17% in our previous publication using the fiducial-less method and 4% using fiducial-based methods of DBS lead placement.

IMU Data and GPS Position Information Direct Fusion Based on LSTM

In recent years, the application of deep learning to the inertial navigation field has brought new vitality to inertial navigation technology. In this study, we propose a method using long short-term memory (LSTM) to estimate position information based on inertial measurement unit (IMU) data and Global Positioning System (GPS) position information. Simulations and experiments show the practicability of the proposed method in both static and dynamic cases. In static cases, vehicle stop data are simulated or recorded. In dynamic cases, uniform rectilinear motion data are simulated or recorded. The value range of LSTM hyperparameters is explored through both static and dynamic simulations. The simulations and experiments results are compared with the strapdown inertial navigation system (SINS)/GPS integrated navigation system based on kalman filter (KF). In a simulation, the LSTM method’s computed position error Standard Deviation (STD) was 52.38% of what the SINS computed. The biggest simulation radial error estimated by the LSTM method was 0.57 m. In experiments, the LSTM method computed a position error STD of 23.08% using only SINSs. The biggest experimental radial error the LSTM method estimated was 1.31 m. The position estimated by the LSTM fusion method has no cumulative divergence error compared to SINS (computed). All in all, the trained LSTM is a dependable fusion method for combining IMU data and GPS position information to estimate position.

Targeting accuracy of robot-assisted deep brain stimulation surgery in childhood-onset dystonia: a single-center prospective cohort analysis of 45 consecutive cases

OBJECTIVE Deep brain stimulation (DBS) is an established treatment for pediatric dystonia. The accuracy of electrode implantation is multifactorial and remains a challenge in this age group, mainly due to smaller anatomical targets in very young patients compared to adults, and also due to anatomical abnormalities frequently associated with some etiologies of dystonia. Data on the accuracy of robot-assisted DBS surgery in children are limited. The aim of the current paper was to assess the accuracy of robot-assisted implantation of DBS leads in a series of patients with childhood-onset dystonia. METHODS Forty-five children with dystonia undergoing implantation of DBS leads under general anesthesia between 2017 and 2019 were included. Robot-assisted stereotactic implantation of the DBS leads was performed. The final position of the electrodes was verified with an intraoperative 3D scanner (O-arm). Coordinates of the planned electrode target and actual electrode position were obtained and compared, looking at the radial error, depth error, absolute error, and directional error, as well as the euclidean distance. Functional assessment data prospectively collected by a multidisciplinary pediatric complex motor disorders team were analyzed with regard to motor skills, individualized goal achievement, and patients’ and caregivers’ expectations. RESULTS A total of 90 DBS electrodes were implanted and 48.5% of the patients were female. The mean age was 11.0 ± 0.6 years (range 3–18 years). All patients received bilateral DBS electrodes into the globus pallidus internus. The median absolute errors in x-, y-, and z-axes were 0.85 mm (range 0.00–3.25 mm), 0.75 mm (range 0.05–2.45 mm), and 0.75 mm (range 0.00–3.50 mm), respectively. The median euclidean distance from the target to the actual electrode position was 1.69 ± 0.92 mm, and the median radial error was 1.21 ± 0.79. The robot-assisted technique was easily integrated into the authors’ surgical practice, improving accuracy and efficiency, and reducing surgical time significantly along the learning curve. No major perioperative complications occurred. CONCLUSIONS Robot-assisted stereotactic implantation of DBS electrodes in the pediatric age group is a safe and accurate surgical method. Greater accuracy was present in this cohort in comparison to previous studies in which conventional stereotactic frame-based techniques were used. Robotic DBS surgery and neuroradiological advances may result in further improvement in surgical targeting and, consequently, in better clinical outcome in the pediatric population.

Automatic Cephalometric Landmark Identification System Based on the Multi-Stage Convolutional Neural Networks with CBCT Combination Images

This study was designed to develop and verify a fully automated cephalometry landmark identification system, based on multi-stage convolutional neural networks (CNNs) architecture, using a combination dataset. In this research, we trained and tested multi-stage CNNs with 430 lateral and 430 MIP lateral cephalograms synthesized by cone-beam computed tomography (CBCT) to make a combination dataset. Fifteen landmarks were manually and respectively identified by experienced examiner, at the preprocessing phase. The intra-examiner reliability was high (ICC = 0.99) in manual identification. The results of prediction of the system for average mean radial error (MRE) and standard deviation (SD) were 1.03 mm and 1.29 mm, respectively. In conclusion, different types of image data might be the one of factors that affect the prediction accuracy of a fully-automated landmark identification system, based on multi-stage CNNs.