Robotic stereotaxic system based on 3D skull reconstruction to improve surgical accuracy and speed

Abstract

Background.

Some experimental approaches in neuroscience research require the precise placement of a recording electrode, pipette or other tool into a specific brain area that can be quite small and/or located deep beneath the surface. This process is typically aided with stereotaxic methods but remains challenging due to a lack of advanced technology to aid the experimenter. Currently, procedures require a significant amount of skill, have a high failure rate, and take up a significant amount of time.

New Method.

We developed a next generation robotic stereotaxic platform for small rodents by combining a three-dimensional (3D) skull profiler sub-system and a full six degree-of-freedom (6DOF) robotic platform. The 3D skull profiler is based on structured illumination in which a series of horizontal and vertical line patterns are projected onto an animal skull. These patterns are captured by two two-dimensional (2D) CCD cameras which reconstruct an accurate 3D skull surface profile based on structured illumination and geometrical triangulation. Using the reconstructed 3D profile, the skull can be repositioned using a 6DOF robotic platform to accurately align a surgical tool.

Results.

The system was evaluated using mechanical measurement techniques, and the accuracy of the platform was demonstrated using agar brain phantoms and animal skulls. Additionally, a small and deep brain nucleus (the medial nucleus of the trapezoid body) were targeted in rodents to confirm the targeting accuracy.

Conclusions.

The new stereotaxic system can accomplish “skull-flat” rapidly and precisely and with minimal user intervention, and thus reduces the failure rate of such experiments.

Introduction

Stereotaxic (or stereotactic) surgery for small animals is an indispensable tool for many types of neuroscience studies (Ferry et al., 2014) and is routinely performed in neuroscience laboratories as part of a variety of experimental procedures, including the creation of site-targeted lesions, injection of anatomical tracers, implantation of electrophysiology electrodes, and insertion of optical fibers or micro-dialysis probes (Athos and Storm, 2001; Fornari et al., 2012; Osten et al., 2007). However, current stereotaxic systems for small animals are largely manually-driven and do not take advantage of modern electronic, mechanical and computer technologies (Athos and Storm, 2001; Carter and Shieh, 2015; Charles and Watson, 2013). For instance, stereotaxic platforms commonly used in today’s laboratories typically have a manually adjustable platform to secure an anesthetized small rodent using a pair of ear-bars or a set of bite-bars. Mechanical arms constructed with translational and/or rotational micrometers hold surgical tools, such as an injection pipette or electrode, and allow for insertion of this tool at various angles and depths into the rodent’s brain. The challenge of this mechanically based design is that precise alignment between the animal’s skull and the surgical tool is difficult to achieve due to the limited accuracy and mechanical stability of the involved manipulators, as well as the “eye-balling” nature of the alignment procedures. The success rates of using mechanical stereotaxic systems varies with the skill and experience of the person using it, as well as with the location and size of the target area. It can be as low as 30% when small and deep brain areas need to be precisely targeted. In addition, the time required to align the (typically anesthetized) animal can be significant with manual adjustments and manual measurements, additionally affecting the mortality and success rate of the procedure. Another challenge is that stereotaxic atlases used to estimate the 3D coordinates of the brain area with respect to anatomical landmarks on the skull are only available for a limited number of species and age groups (Charles and Watson, 2013), contributing to errors in determining the insertion angles and distances. These limitations and shortcomings of the current generation of stereotaxic systems slow down the pace of discovery. Additionally, current devices require a significant amount of skill and experience, thus reducing the reproducibility of the underlying research. Therefore, better solutions and designs to provide faster, more accurate and more automated stereotaxic surgeries for small rodents are needed.

Computer vision and robotics have made a significant impact on modern surgical practice for human patients. Several computer vision and robotic systems have already been used in hospital operating rooms to improve treatment outcomes and reduce the risks of surgical procedures (Lanfranco et al., 2004). In order to increase dexterity and accuracy, some surgical robots were designed to perform motion tasks with multiple degrees of freedom (DOFs) for minimally invasive robotic surgeries (MIRS) (Hagn et al., 2010; Hannaford et al., 2013), ranging from medium-sized laparoscopy and endoscopy to small-sized tissue grafting (Shaikh, 2010; Shang et al., 2017). Such advances also have the potential to reduce surgical time and increase surgical accuracy.

For animal surgeries, there were some efforts to develop improved stereotaxic systems for small rodents to reduce human error, improve surgical accuracy and save time in recent years. Pak et al., developed an automated craniotomy system which controls a motorized drill based on impedance measurements to open a cranial window as small as several millimeters with high reproducibility (Pak et al., 2015). Although this system is robotically and electronically controlled, it is not a full stereotaxic system, lacking the capabilities to perform stereotaxic positioning and automatic insertion of surgical tools. The system was only designed to retrofit to existing manual stereotaxic systems and to open holes on the skull automatically with high spatial precision. Neurostar, on the other hand, developed a robotic stereotaxic system in which the three translational actuators were motorized, and an electronic brain atlas was added to the control computer to guide the experimenter during stereotaxic surgeries (neurostar.de). However, this system only provides three motorized translational actuators, and rotational positioning remains to be manually controlled. This makes it impossible for the system to automatically align the animal to the “skull-flat” position, which is critical for precise targeting. The same system also uses a pair of 2D CCD cameras to identify the single Bregma landmark on the skull and guide the surgical tool, but the design lacks the needed hardware, such as structured illumination light sources, for a full 3D skull profile reconstruction. Another robotic stereotaxic surgical system was developed by Brainsight (rogue-research.com). Their Vet robot uses a full 6 DOF robotic arm to guide a surgical tool to a brain region. The system is also integrated with MRI or CT images for automatic stereotaxic guidance. The Vet robot is equipped with two CCD cameras for 3D positioning registration and a 200 μm positioning accuracy was demonstrated on their system. Compared to the Neurostar system, the Brainsight system’s full 6 DOF robotic arm allows surgical insertions at all angles. However, the stereo camera system still lacks the capability to reconstruct the skull profile in 3D and only a single target point can be identified. Another disadvantage of using a full 6 DOF robotic arm is the high cost of ownership, which makes the system difficult to adapt for many neuroscience laboratories. Recently, L. Ghanbari et al created a Craniobot which integrates a low-force contact sensor to a computer numerical controlled (CNC) milling machine to allow high precision contact-based mapping of a mouse surface skull profile (Ghanbari et al., 2019). Using the mapped skull profile and the contact force measured by the sensor in real-time, the animal’s skull can be thinned with great precision by a fine drill bit controlled by the CNC machine in a feed-back manner. However, the Craniobot relies on micro-CT 3D images as skull positional reference and the system can benefit significantly by incorporating a 3D optical profiling system for automatic skull positional alignment.

In this report, we describe a robotic stereotaxic system for small rodents using a 3D skull surface profiler to precisely control a 6DOF robotic platform to rapidly and accurately target deep brain nuclei with minimal user interventions. The system accomplishes “skull-flat” rapidly and with high accuracy, thereby potentially improving the success rate of brain surgeries. The 3D skull surface profiler uses structured illumination and geometrical triangulation to map the skull surface of a small rodent with high spatial (sub-millimeter) precision. In addition, a full 6DOF robotic platform will provide a large translational and rotational range of motions to position the animal for precise stereotaxic procedures. In this paper, the design and construction of the system is discussed in detail. We present evaluations of the system’s accuracy using mechanical measurements, agar brain phantoms, rodent skulls and live animals into which fluorescent dyes were injected into deep brain nuclei to demonstrate targeting accuracy and speed.

Methods

The goal of this project was to design and construct a stereotaxic device that can perform accurate stereotaxic surgery on small animals such as rodents automatically and with limited human intervention, thus leading to increased targeting accuracy and higher reproducibility especially for small and deep brain nuclei. More specifically, we intended to address two areas of inaccuracy that many existing devices have: 1) Inaccuracies due to manual alignment of the animal. We replaced the manual alignment with 3D visual surface reconstruction to accurately and rapidly achieve the “skull-flat” position, a commonly used standard orientation in stereotaxic surgery. 2) Inaccuracies with the manual movement of the various axes to adjust, rotate, and tilt the animal into position. We replaced manual manipulators with a motorized Steward platform.

First area of concern - measurements of skull-flat: We developed a computer vision system that performs 3D scans of an animal’s skull (Ponce and Forsyth, 2012; Snyder and Qi, 2017). The computer vision system is a 3D skull profiler that can scan the rodent’s skull using a video projector to project line patterns which are then imaged by two regular CCD cameras carefully attached on either side of the projector (Geng, 2011). The acquired images can be used to calculate a 3D skull profile with sub-millimeter spatial resolution based on the techniques of structured illumination and geometrical triangulation (Hartley and Sturm, 1997).

Second area of concern - precision of movement: We developed a stereotaxic platform that is based on a hexapod (Stewart) design (Dasgupta and Mruthyunjaya, 2000; Nanua et al., 1990; Stewart, 1965; Szufnarowski, 2013). A Stewart platform consists of two plates that are connected to each other via six motorized axes. Lengthening or shorting these six axes in a synergistic way will allow the top plate to rotate, tilt, or move against the stationary bottom plate with six DOF (X, Y, Z, roll, pitch, yaw). Based on the spatial information reconstructed from the 3D skull profile, the animal skull can be moved to a desired location according to the specific brain area calculations derived from brain atlas coordinates.

Figure 1 shows both the schematic diagram and the actual photographic image of the automated robotic stereotaxic system illustrating the major components required for the system.

Figure 1.

Robotic stereotaxic system based on 3D structured illumination and geometrical triangulation. A. Schematic diagram of the stereotaxic system illustrating the crucial components of the system. Two cameras (1) mounted onto the two sides of the center axis were equipped with zoom lenses to focus on the rodent’s skull (7). A video projector (3) was used to project structured light patterns onto the skull surface. A 75 mm bi-convex lens with a small slanted angle was used to focus the projection plane of the structured image onto the skull. A 6 degree-of-freedom (3 translational and 3 rotational) robotic platform (5) was used to secure the rodent and allowed positioning the animal’s head with sub-millimeter spatial resolution. A surgical device (6) such as a nanoliter injector can be placed along the center axis to perform stereotaxic surgical procedures. B. Photographic image of the actual automated stereotaxic system.

Optical structured illumination for precision 3D skull mapping

A video LED projector (S1, ASUS, Taipei, Taiwan) was mounted slightly offset from the center longitudinal axis of the setup. Commercial projectors such as this one are designed to project vertical and horizontal lines in a tilted upward angle (Keystone correction), facilitating off-center mounting. Since commercial video projectors were designed to project enlarged images at normal viewing distances, a 75 mm bi-convex lens (LB1901-A, Thorlabs, Newton, NJ, USA) was placed in front of the video projector to focus the projected lines with a much finer spatial resolution onto the rodent’s skull to achieve high 3D reconstruction. Two 2D CCD cameras (BCE-B013-U, Mightex, Pleasanton, CA, USA) were attached on either side of the projector to capture images of the skull illuminated with the projected line patterns. Two 10x zoom lenses (MLH Macro 10X, Computar, Las Vegas, NV, USA) were mounted onto the two CCD cameras to reduce field of view onto the rodent’s skull and maximize the image resolution for the 3D reconstruction. The two CCD cameras have a pixel density of 1280 × 1024 covering a field-of-view sized approximately 50 × 40 mm which translates to a lateral spatial resolution of 30 × 35 mm for the 3D reconstructed profile.

A series of horizontal and vertical black-and-white lines with increasing spatial frequencies (structured illumination) were projected onto the rodent’s skull, and the two CCD cameras were used to capture 2D images of the skull covered with projected lines from both sides. Using these captured 2D images with structure-illuminated patterns, a unique binary spatial code can be created for each of the image pixels for position identification for the 3D reconstruction. The binary spatial code is a one-dimensional binary linear array in which the array index is associated with the sequential order of the captured images. A binary one (or zero) is assigned to the bin of the array if a bright (or dark) pixel is captured for the image in sequence. Based on the unique spatial codes created for all pixels, the 3D locations of the points on the rodent’s skull surface can be estimated based on geometrical triangulation. This geometrical triangulation process can be understood as a lateral displacement of the projected lines when these straight lines are projected onto a non-flat surface. Moreover, the degree of lateral displacement is linearly proportional to the vertical corrugation of the displacing object (in this case the rodent skull). Mathematically, a surface point P in the 3D space can be estimated by

$$P=\frac{1}{2}\left(C^L+C^R+{\lambda }^L\cdot v^L+{\lambda }^R\cdot v^R\right)$$

where v^L and v^R are the viewing vectors pointing towards the point P for the left and right cameras; C^L and C^R are the 3D coordinates of the center points of the left and right cameras; and λ^L and λ^R are parameters can be calculated based on geometric triangulation. A detailed discussion for the equations to calculate λ^L and λ^R, and the 3D structure illumination method can be found in the supplementary information. A custom Python program was written to process the captured 2D images and reconstruct the 3D skull profile. A total of 42 vertical and horizontal line patterns were projected onto the rodent skull and captured by each of the CCD cameras with a frame rate of 25 fps. All the images were captured in standard laboratory ceiling lighting conditions with no additional lighting provided, but also no light shielding employed. The exposure time for each image was 20 ms with an average of 8 frames to improve image contrast and to remove optical noise induced by the projector and room lighting. The overall image capture time was approximately 86 seconds. An Intel Core i5-4430 CPU @ 3.00GHz with 16.0 GB RAM commercial desktop computer was used for the reconstruction process along with a custom Python program. The processing time to reconstruct the 3D profile was approximately 9 seconds.

Full 6 degree-of-freedom robotic platform

The robotic platform provides a full 6 DOF - 3 translational and 3 rotational - movements to allow precise positioning of the rodent’s skull. The platform is based on the Stewart design in which six motorized arms are attached to a moveable top plate and a stationary bottom plate (Stewart, 1965). The movements of the 6 arms are coordinated to give the top plate a full 6 DOFs. This is different from other conventional Stewart designs in which 6 linear actuators are used. The robotic platform used in this project was built with 6 digital rotational servo motors (AX-12A, Robotis, South Korea), and each was connected to a pair of short and long arms which formed a semi-triangle with a rotational pivot. The short and long arms act as the two known edges of a semi-triangle and through rotating the servo angle, the hypotenuse edge of the semi-triangle can be shortened or lengthened to provide the desired linear extension for coordinated platform motion. Details of the control algorithms for the platform can be found in the supplementary information. This design allows the platform to have precise sub-millimeter translational and sub-degree rotational positioning accuracies and at the same time reduces the cost of ownership by using low cost rotational servo motors. In the current system, the top and bottom plates were initially designed with Solidworks mechanical design software and were 3D printed by a 3D plastic printer (Objet30, Stratasys, Rehovot, Israel) with a printing spatial resolution of 28 μm. A heating pad was embedded into the top plate and was controlled by a Proportional-Integral-Derivative (PID) controller to maintain 37°C body temperature for the rodent. Multiple accessories, such as ear and bite bars, were also mounted onto the top plate for securing the animal on the platform (Charles and Watson, 2013).

Robotic stereotaxic skull alignment and positioning

Robotic stereotaxic surgery can be achieved by coordinating the 3D reconstructed skull profile obtained from the computer vision system with the movements of the 6 DOF stereotaxic platform through the custom Python control program. A small rodent was securely positioned onto the top plate of the robotic platform either using a pair of ear bars or a metal head post fixed to the rodent’s skull. The animal’s body temperature was maintained through a PID-controlled heating pad. After the head was shaved and the skull was exposed, a series of structured illuminated images were taken by the two CCD cameras to create the 3D skull profile through the software routine. Users were prompted to select the Bregma and Lambda landmark positions by clicking on the 3D reconstructed skull profile which was rendered in a 3D viewing window on the computer screen. Based on the 3D coordinates of Bregma and Lambda, the mid-point between these landmark points was automatically calculated. In addition, two points that are 4 millimeters lateral in either direction from the mid-point and were perpendicular to the line connecting Bregma and Lambda points were estimated. Thus, these five points (Lambda, Bregma, mid-point, two side points) formed two perpendicular lines that define a 3D surface normal plane for the rodent’s skull in space. Using this surface normal plane, the robotic plane was translated and rotated to achieve the “skull-flat” position, defined as the Bregma and Lambda landmarks being at the same height level in the robotic plane. At this point, the users entered the desired translational and rotational displacements referenced to the landmarks of the skull to initiate movements of the robotic platform moving to any desired location.

Animal protocol and procedures for stereotaxic evaluation

All experimental procedures complied with all applicable laws and NIH guidelines and were approved by the University of Colorado IACUC. All experiments were conducted in adult Mongolian gerbils (Meriones unguiculatus). This rodent species was chosen because gerbils are readily available and typical sizes of adult animals are almost exactly between typical mouse and rat sizes. Gerbils were first anesthetized with a mixture of ketamine-xylazine (60mg/kg-5mg/kg), and a maintenance dose of (25mg/kg-5mg/kg) was given every 30 minutes following complete anesthesia to maintain the anesthetized state. Once the animal was properly anesthetized, the fur over the skull was shaved off and the underlying skin sanitized with a disinfectant. Skin and muscle overlying the skull were removed and a craniotomy was made in the skull at 4 mm posterior of lambda and 0.8 mm and 0.6 mm lateral using a dental drill. The coordinates used here allowed us access the auditory brainstem for dye injections into the Medial Nucleus of the Trapezoid Body (MNTB). This nucleus was chosen because it is very deep (almost at the ventral side of the brain) and small (about 0.5 mm in diameter) and thus very challenging to target. The rationale was that if that brain area could be targeted successfully, a user should be able to target any other brain area with a similar or better success rate. Dye injections were made into MNTB using a Nanoliter injector (World Precisions Instruments, Sarasota, FL). 32.2nL of dye was injected at a depth of 7.2 mm and 7.5mm once every 30 seconds and 8 injections were made at each location. After all injections were concluded, the animal was given an overdose of pentobarbital (0.03 mL/g) and perfused transcardially with phosphate buffer solution (PBS), followed by 4% paraformaldehyde (PFA). Once perfusion was complete, the brain was fixed in 4% PFA overnight, then removed and placed into 4% agar. The brainstem and cerebellum were sliced coronally in 100 μm sections using a vibratome (Leica VT 1000s, Nussloch, Germany). The sections were then stained with a 1:100 concentration of Neurotrace Nissl stain (ThermoFisher, Waltham MA, 640/660 deep red fluorescent Nissl stain) (Albrecht et al., 2014). The slices were mounted on slides with Fluoromount-G (Diagnostic BioSystems, Pleasanton CA) and imaged with an Olympus FV1000 (Tokyo, Japan) confocal microscope using the laser lines of 405nm 555nm, and 647nm.

Results

3D skull profile reconstruction using structured illumination

The skull of a rodent was reconstructed in the 3D space through the technique of structured illumination. Figure 2A shows example images of the skull illuminated with vertical and horizontal line patterns taken by the left and right 2D CCD cameras. A total of 42 photos were taken by each camera with spatial frequencies varying from 0.025 to 25.6 lines/millimeter for 3D reconstruction. According to the rule of optical projection, vertical displacement on the rodent’s skull in turn creates lateral displacements in the acquired 2D images. These lateral displacements can be used to estimate the vertical displacement of each point on the skull surface using geometrical triangulation. To correctly calibrate the camera systems and obtain sub-millimeter spatial reconstruction, 3D printed calibration targets and optical calibration targets with known dimensions were first imaged to estimate the correct parameters for the focal lengths of the CCD cameras and the relative positional parameters for the cameras and the projectors. Through this calibration process, it was determined that the 3D reconstructed skull profile can achieve a spatial resolution of 98.5±4.5 μm.

Figure 2:

3D skull profile reconstruction using structured illumination. A. A Mongolian gerbil skull was illuminated by example vertical and horizontal lines observed by the left (top) and the right (bottom) cameras. B. Reconstructed calibration standards using structured illumination (top left) on 3D printed calibration standards in pyramidal shapes (top right). Reconstructed optical target (bottom left) and the calibrated optical target (bottom right) for calibrating the reconstruction routine to obtain sub-millimeter spatial resolution. C. A regular 2D image taken by the left black and white CCD camera (top) and the reconstructed 3D skull profile (bottom) of a Mongolian gerbil skull. The reconstructed 3D points of the skull profile were colored with greyscale intensity to visualize stereotaxic landmarks (Bregma and Lambda). D. Reconstructed 3D skull profile of an anesthetized Mongolian gerbil in two different view angles clearly showing the exposed skull and skin (top). The Bregma (top red dot) and Lambda (bottom red dot) landmarks were used to identify and to estimate the center intersection point (green dot) and two side points (left blue and right cyan dots) 4 millimeters perpendicular to the center connecting line (bottom). Note: all scalebars are 1 mm in length.

The 3D profiler was used to reconstruct the profile of an in-vitro Mongolian gerbil skull. The 2D normal CCD image and the reconstructed 3D profile of the gerbil skull are shown in Fig. 2C. The reconstructed 3D skull profile was superimposed with greyscale intensity obtained from a normal 2D skull images for better visualization. The stereotaxic landmarks and bone sutures can be clearly identified from the reconstructed 3D skull profile. The reconstructed skull profile has very few missing 3D points on the skull surface. The structures on either side of the skull surface were not reconstructed, and this is because only one camera can see a given side of the skull, making geometrical reconstruction not possible for these points. However, the inability to reconstruct the side surfaces was not important for our purposes since the normal plane for the rodent’s skull can be accurately determined using the top skull surface alone.

After confirming the functions and accuracies of the computer vision 3D profiler with an isolated animal skull, an anesthetized Mongolian gerbil was then used for additional system evaluation. The gerbil was secured to the robotic platform using a headpost attached to the dorso-frontal cranium with dental cement, and bolted to the robotic platform (Figure 3H). To this end, the scalp of the animal was surgically opened, exposing the top surface of the skull including Bregma and Lambda, two skull suture points which are commonly used in stereotaxic surgeries (red dots in figure 2D). The surface of the skull was scanned by the 3D profiler without the use of the image enhancer, fiducial markings or other treatment methods. The reconstructed 3D surface profile is shown in Fig. 2D in two different viewing angles (top figure). The skull surface was reconstructed showing details including the stereotaxic landmarks and part of the exposed scalp. Using custom written control software, the reconstructed 3D skull profile was manipulated on the computer screen in real-time using a computer mouse, and the Bregma and Lambda points were confirmed by the user. Based on these two stereotaxic landmarks, the mid-point between them, plus two additional points 4 mm lateral to the midline on either side of the midline were estimated automatically by the software. Based on these points, two perpendicular 3D lines were determined, defining a normal surface plane for the gerbil skull. This normal plane was then be used to move the robotic platform to the “skull-flat” position, in which Bregma and Lambda are at the same height with an error of less than 100 μm, for subsequent alignment and stereotaxic procedures.

Figure 3:

Components and add-ons of the six degree-of-freedom robotic platform. A. A 3D printed head rest that was optionally attached to the platform to better position the animal’s head for stereotaxic surgery. B to D. 3D printed sub-components of a rotatable and extendable manipulator (F) to hold a metal bar were fixed to the skull for securing the animal (H). E. A temperature heating pad controlled by a PID controller was embedded into the top plate of the robotic platform to maintain the body temperature of the rodent. G. 3D printed top and bottom plates with attaching parts and short arms for motors of the robotic platform in which the top plate has several mounted holes to allow add-ons to be installed.

Full six degree-of-freedom robotic stereotaxic platform developed using 3D prototyping technology

The full six DOF robotic stereotaxic platform was manufactured using rapid prototyping technology. The majority of the components were created using 3D printing technology and were designed using Solidworks 3D CAD design software (Dassault Systèmes, Vélizy-Villacoublay, France). The components were then printed using a high-resolution 3D printing system (Objet30, Stratasys, Rehovot, Israel) with a spatial resolution of 28 μm. Rapid prototyping manufacturing technology allowed the platform to be easily configured to hold different animal species and sizes. Additional accessories were also required for stereotaxic work to be successful with live animals. For example, many animal species and/or surgical procedures require that the animal maintain physiological temperature, so a temperature-controlled heating pad was installed into the plate of the hexapod. Also, the animal’s head needs to be immobilized so it remains stable and in position during the surgeon’s handling and manipulation. Depending on the brain area and investigator’s experimental requirements, this fixation may occur via ear bars, bite bars, a head post, a head rest, or a combination of these methods. Figure 3 shows some custom add-ons that were manufactured with 3D printing prototyping technology. Figure 3A is a custom head rest that is designed for medium sized rodents such as Mongolian gerbils. In addition, many stereotaxic devices use ear bars to secure the rodent on the platform. Since these cannot be used in some types of experiments, for example those involving the auditory pathway, and thus a head post mount was designed to hold a head post which can be secured with dental cement to the skull of the rodent. Figures 3B – D show the details of the sub-components of the manipulator which subsequently were assembled to attach to the top plate, as shown in Figure 3H. Figure 3E shows the heating pad controlled by a PID controller with a thermostat placed underneath the animal to ensure that a constant temperature of 37°C is maintained throughout the entire stereotaxic surgery. The heating pad was embedded into the top plate of the platform seamlessly to provide maximum compatibility. Figure 3G shows the top and bottom plates of the robotic stereotaxic platform with constructive parts, also prototyped using 3D printing technology. The technology allowed for custom mounting holes and placement grooves for the top plate to easily mount a variety of components to secure the animal or install other life-support related installments.

The robotic platform was first characterized using mechanical calibration techniques. Considering the radius of a small rodent’s skull, which is in the order of 20 mm or smaller, and the typical dimensions of target brain regions inside the rodent’s brain which can be as small as 0.2 mm, translational and rotational accuracies of 200 μm and 0.5 degrees, respectively are required to accurately target any given brain region. In addition, the translational and rotational motions of the robotic platform must be linear to ensure smooth positioning for stereotaxic surgeries. Three mechanical dial gauges (Model 25–611, L. S. Starrett, Athol, MA, USA) and a digital 9 DOF inertial measurement unit (IMU) chips (MPU9150A, InvenSense, San Jose, CA, USA) were used to measure both the translational displacements and the rotational angles of the robotic platform against the desired translational distances or rotational angles issued by the control computer. The three translational gauges were mounted against the platform perpendicular to one another for all three axes, and the IMU chip was directly mounted at the center of the top plate to measure the rotational angles. Figure 4 shows some representative calibration curves (the z linear axis and the pitch angle) to demonstrate the linearity in both translational and rotational motions. Fig. 4C compares the desired and actual movements in all three translational and rotational axes. The overall error was less than 4.4 %. Using calibration tools, the robotic platform was determined to have a full translational distance of ± 15 mm with an accuracy of ± 0.25 mm, and a full rotational angle of ± 20° with an angular accuracy of better than ±0.1°, as summarized in Table 1.

Figure 4:

Positioning accuracy of the robotic stereotaxic platform A. Desired Z translational positions against the actual Z translational positions by the platform. B. Desired pitch rotational angles against the actual pitch rotational angles by the platform. C. The estimated accuracies in all six degree-of-freedom motions for the robotic platform. D. A cross pattern with an inter-spacing of 1 mm was injected into an agar brain phantom using the stereotaxic platform. E. Longitudinal injections, performed with the stereotaxic platform, with a depth difference of 1 mm were quantified in the agar brain.

Table 1:

Maximum translational and rotational ranges and resolutions of 6 DOF robotic platform

Top platform	Measured ranges and accuracies
	Translational	Rotational
Maximum range	± 15 mm	± 20°
Resolution	± 0.25 mm	± 0.1°

Calibration to line up the 3D camera to the robotic platform

In order to use the measured 3D camera profiler to accurately guide the robotic platform for precise positioning of the animal skull, careful calibration was performed to position the 3D surface profile correctly on top of the robotic platform. The calibration goal was to align the virtual 3D space constructed by the 3D camera profiler to the physical dimensions of the robotic platform in the laboratory frame. A small dimple marking the center of the top plate was added to the Solidworks design and was engraved to the top plate during the 3D printing process. The center dimple was then used to roughly align the camera system in which the image views of the two cameras were centered and focused onto this center dimple. A calibration target with a two-dimensional dot pattern and a dot-to-dot center spacing of 1 mm was used and placed at the platform center to allow the 3D camera profiler to reconstruct its 3D dot profile. Based on the dimensional measurements of the reconstructed dots in the virtual 3D space, the camera profiler was repositioned accordingly until the 3D reconstructed dimensions of dot pattern were measured correctly in the virtual 3D camera space.

Stereotaxic accuracy tested using an agar brain phantom

The robotic platform was first evaluated using a brain phantom made from agar to demonstrate its positioning accuracy. The robotic platform was first programmed to print a “cross” pattern with an inter-spacing of 1 mm into the agar brain phantom, as shown in Fig. 4D. The diameter of the glass micropipette tip used in the injection was estimated to be ~10 μm and there is no observable deviation of the printed pattern on the agar brain phantom. Injection depth accuracy was then tested via vertical injections with a 1 mm depth difference as shown in Fig. 4E, and the injection printing again has no observed deviation.

Stereotaxic in-vivo injections into a deep brain nucleus

To test the system’s performance in actual stereotaxic procedures, an anesthetized gerbil was placed on the top plate of the robotic stereotaxic platform and secured via a head post attached to a customized manipulator on the top plate, as shown in Fig. 5A. The top scalp of the gerbil was incised and skin and muscle overlying the dorsal skull were removed, exposing the skull Structured illumination scanning was used to construct the 3D skull profile on the computer. Bregma and Lambda landmarks were identified and selected on the computer to determine the skull “normal” surface. The “skull-flat” position was achieved by calculating the required translational and rotational displacements of the skull normal surface and translating into a computer command to move the robotic platform with the gerbil to the determined position. From skull-flat, the animal on the top plate were tilted via an additional rotational displacement of −20 degrees pitch to facilitate access to the desired brain area, and additional translational displacements of 4 mm posterior, then 0.8 mm and 0.6 mm laterally to the left or right of Lambda, respectively, were initiated. These displacements were estimated based on a gerbil brain atlas (Radke-Schuller et al, 2016) to target the MNTB in the brainstem. After drill positions were marked and two small holes were drilled manually, the skull was re-scanned by the 3D computer vision skull profiler. The robotic platform then was commanded to compensate for any positional errors induced by the drilling process. Two glass pipettes with tip diameters of ~10 μm were filled with one of the following two fluorescent dyes: cascade-blue (ThermoFisher, Waltham, MA) for the left hemisphere, or micro-ruby (ThermoFisher, Waltham, MA) for the right hemisphere. The electrodes were inserted into to the cranium through one of the two holes and advanced to a depth of 7.5 mm for the left side and 7.2 mm for the right side. Pipette advancement was done by gradually raising the top plate of the robotic platform through a series of commands towards the glass pipette, as shown in Fig. 5B. Note that this procedure is unusual and was used for demonstration purpose with our prototyping setup. The main purpose of these dye injections was to test the accuracy of the platform and therefore we aimed to avoid additional third-party technology that might obscure our measurements. During typical laboratory use of the platform, the investigator would most likely use an additional motorized axis for lowering the electrode into the cranium to achieve better control and higher flexibility. After the injections, the animal was sacrificed, transcardially perfused, and the brain was extracted and sectioned for fluorescence imaging. Fig. 5C shows a coronal section with a yellow and a blue tract, showing the path of the electrode to the target area at the MNTB. The insert in figure 5C show this area marked in a brain atlas sketch. This live stereotaxic surgery confirms that the system can be used with live animals and is suitable to inject dye into the intended brain region precisely with very minimal user input.

Figure 5:

Robotic stereotaxic brain surgery on a Mongolian gerbil. A. In situ image of the anesthetized gerbil on the platform. B. Close-up view of the stereotaxic injection procedure with the glass micropipette inserted into the animal’s brain. C. Fluorescence imaging of a brain section showing the successful double-injection of the two MNTB nuclei. The exposed skull was first 3D scanned and automatically positioned for stereotaxic injection into the ventral end of the left medial nucleus of the trapezoid body (MNTB) (blue tracer, left, 0.8 mm lateral of midline, 7.5 mm below brain surface) and the dorsal end of the right MNTB (yellow tracer, right, 0.6 mm lateral of midline, 7.2 mm below brain surface), deep within the brain. MNTB was chosen as a sample target because it is located almost at the ventral end of the brain and is only about 0.5 mm large in gerbils, requiring a targeting accuracy of better than 1 degree. Blue and yellow = tracer injections; magenta – fluorescent Nissl. Superimposed onto the image are the outlines of major brain areas which are present in this section. The outlines and atlas information were adapted from the gerbil brain atlas (Radtke-Schuller et al., 2016). The target area of the injection sites plus surrounding brain areas are labeled, all other labels have been omitted for clarity. For a complete set of all labeled brain areas see plate 45 of the Schuller et al., 2016. MNTB = medial nucleus of the trapezoid body; SPN = superior olivary nucleus; MSO = medial superior olive; VNTB = ventral nucleus of the trapezoid body; tz = trapezoid body; py = pyramidal tract.

A video showing the entire stereotaxic alignment process including image taking, 3D skull profile reconstruction, picking the skull landmarks and computer-guided platform moving to achieve skull-flat position is provided in the supplementary information. Note that the video was 5x time-accelerated for a total video length of 63 seconds, indicating that the entire process was completed in less than 5:30 minutes.

Discussion

In this report, we demonstrate that a novel robotic stereotaxic system was realized by combining a 3D computer vision optical skull profiler based on the techniques of structured illumination and geometrical triangulation with a 6DOF robotic platform based on the Stewart design. Through this combination, the skull profile of a small rodent can be reconstructed in the virtual space with a high degree of accuracy. The 6DOF robotic platform can then be instructed by the positioning estimations to move the rodent’s skull to the correct position for stereotaxic surgeries. To the best of our knowledge, this is the first time a 3D optical profile is being used to capture the 3D point cloud of a rodent’s skull for stereotaxic purposes. While other existing robotic stereotaxic systems use single or multiple CCD cameras, these cameras are mostly used to determine the coordinate of a single or multiple fiducial spheres or landmarks. The advantage of using the approach described here is that no preparation or device mounting is required to provide a full 3D reconstruction of the skull profile. The Stewart based platform design described here provides full 6 DOF motions in all translational and rotational axes with a large range of movements (±15 mm and ±20°) and high precision (±0.25 mm and ±0.1°), which is sufficient to cover all translational and rotational needs for small rodent stereotaxic procedures. In addition, the ranges of platform movements are related to the variation lengths of the six arms, thus the platform can be designed in terms of platform size and arm lengths to accommodate other animal dimensions.

Another strength is that the system is constructed with relatively low-cost components, making it highly affordable and suitable for wide use. The 3D optical skull profiler is constructed using two low-cost CCD cameras with two 10x tele lenses and a commercial computer projector with a low-cost 75 mm bi-convex lens. This design does not require special time-of-flight 3D cameras or other special optical components for the 3D reconstruction (Cui et al., 2013; Izadi et al., 2011). The robotic platform was built using 6 low-cost digital servos and the other components were 3D printed using rapid prototyping technologies. Originally, six linear translational actuators were used to realize the Stewart platform design, but these required good spatial resolution which drastically increased the building costs of the platform. In contrast, we found digital rotational servos to be more economical, and found they can be used to achieve precise movement outcomes through trigonometric relation as illustrated in the supplementary information. Therefore, the digital servos were chosen to build the rotational platform, allowing the device to remain accessible while maintaining high spatial positioning accuracy. Additionally, the present system was designed to move the animal together with the robotic platform while most other robotic stereotaxic systems keep the animal stationary. This allows the robotic platform to achieve accurate stereotaxic results, but with a significant cost reduction.

An increasing number of experimental approaches in neuroscience require the precise placement of a recording electrode, injection pipette or some other tools into a specific brain area that can be quite small and/or located deep beneath the surface. Reaching these brain areas with traditional methods and devices can be challenging for several reasons. First, some brain nuclei are less than 0.5 mm in diameter and may be as deep as 7–10 millimeters in typical rodent species, requiring a target accuracy of better than 1 degree. Second, investigators try to minimize the size of the opening holes in an animal’s skull through which a tool is advanced to minimize surgical trauma from the intervention. Third, jaws, face, or ears additionally limit the locations on the head where craniotomies can be performed, such these practical reasons, most craniotomies use a dorsal approach. These limitations make the targeting of deep and small brain areas (midbrain, brain stem, thalamus, subregions of hippocampus, and many other types of non-surface structures) especially challenging, and the challenges are further increased by a lack of sophisticated technology. As a result, many in-vivo manipulations either require a significant amount of experience by the experimenter, and/or have a significant failure rate. Failed experiments are costly in terms of wasted investigator time, research animals and materials. However, even in cases where a lab has the expertise to target the desired brain area with a relatively high success rate (for example, because they employ a very skilled student/postdoc/lab tech), this is problematic since the success depends on that person and their “magic touch” – effectively reducing the reproducibility of these experiments for all other labs that don’t have access to that person. Even the same lab may have trouble reproducing their own experiments once that person with the “magic touch” leaves. The idea behind the device described here is to eliminate qualitative aspects of stereotaxic procedures, such as the skill and experience of the operator as much as possible, and to replace these qualities with precisely measured and precisely repeatable automated procedures. Neuronavigational features can be added to the device in the future, which would increase the automatism and the repeatability of stereotaxic work with this device further.

We propose that this platform with its current dimensions can be used with a number of animal models that are typically used in laboratories, such as mice, rats, gerbils, guinea pigs, and other similar sized species. Of these, the most commonly used models are mice and rats, and therefore we expect that most users would use this platform with one of these two species. We therefore chose to test the platform with an animal species that “splits the difference” between rats and mice, namely Mongolian gerbils (Meriones unguiculatus). Adult gerbils weigh about 80 – 150 grams and are thus an ideal medium between the smaller mouse model (adult mice weigh about 18–30 grams) and the larger rat model (adult rats weigh 250–500 grams). In other words, by showing that the platform is compatible with gerbils, it is easy to conceptualize that it would work equally well a slightly smaller species (mouse), or a slightly larger species (rat). Should a user wish to use this platform with a significantly larger species, there would be two major considerations for adapting the system – the optical magnification of the 3D profiler and the movement ranges and resolutions of the robotic platform. For larger animals, including dogs, cats, rabbits or chinchillas, a smaller optical magnification is required to cover a larger skull surface, which is easy to achieve by using lower magnification optics. The translational and rotational ranges of the robotic platform are required to be increased which can be achieved by using longer arms for larger reach. For animals even smaller than mice, the appropriate resolution of the 3D optical profiler can be achieved even using even higher magnification optics.

The current system has a translational and rotational resolutions of ±0.25mm and ±0.1°, which are marginally adequate for targeting small nuclei deep within the brain. To further improve resolution, we are working on redesigning the robotic platform using newer digital servo motors, which can provide a 4 × higher resolution, to increase targeting resolution. In addition to increasing the spatial resolutions, the system can be further enhanced in the following areas: Designing a larger open area between the animal and the cameras to allow investigators to mount additional stereotaxic tools, such as drills and injection devices, for various surgical procedures. In this regard, magnification lenses with longer object distances would be suitable to create a larger surgical space. The optical projector and the 2 cameras could also be mounted on high precision swivel bases to allow the optical systems to be moved out of the way when not used for skull scanning. In addition, the current software was written with the Python programming language, and its mathematic calculations are slowed by the interpretative nature of the programming language. In addition, only a single core CPU was only used for the 3D reconstruction, and improvements in computational speed can be achieved using multiple CPU cores or GPU acceleration (Du et al., 2012). The current software requires users to visually identify the Bregma and Lambda points on the 3D reconstructed skull profile and this identification process can further be automated using image processing techniques in the future. Holes for injections were still manually drilled, and additional drilling add-ons could be installed to allow automatic drilling. Installing an impedance-based sensor on the drill attachment and platform like the design of Pak et al. would allow for precise drilling of the skull without damaging brain tissue (Pak et al., 2015).

Conclusion

A new type of stereotaxic system for small animal brain surgeries has been developed by combining a 3D computer vision sub-system and a 6DOF robotic platform. High resolution 3D reconstruction of an animal skull has been demonstrated. No special 3D camera or hardware were required, but rather a series of images were captured by two regular computer cameras mounted on either side of the skull. A series of structured patterns were projected onto the animal skull using a video projector to increase optical contrast and allow high resolution 3D capture on a relatively featureless skull structure. Six low-cost digital servo motors were used and controlled the 6 extendable arms to allow placement of the top platform in all 3 translational and 3 rotational positions with few limitations. Both the 3D camera sub-system and the robotic platform were characterized using agar brain phantoms and rodent skulls. We found the device to have sub-millimeter and sub-degree spatial resolution suitable to precisely target a small brain region within a rodent’s head. A simulated stereotaxic surgery using an anesthetized Mongolian gerbil confirmed that the MNTB, a small and deep brain nucleus located in the ventral brain stem can be accurately targeted using the system, demonstrating targeting accuracy of better than 1 degree. We anticipate that this new system will help the advancement of neuroscience research through an increased ease of use and success rate for stereotaxic surgeries, and reduced surgical time which will additionally help the success of the experimental manipulation.