3D Printing for Genetic Mouse Models of Congenital Heart Disease

Malek Khalifa1, Darshit Thakrar2, Cecilia WY Lo3 and Yijen L Wu3,4 1Department of Chemical Engineering and Biomedical Engineering, Carnegie Mellon University, Pittsburgh, USA 2Department of Radiology, Children’s Hospital of Pittsburgh of UPMC, Pittsburgh, USA 3Department of Developmental Biology, University of Pittsburgh, Pittsburgh, USA 4Rangos Research Center Animal Imaging Core, Children’s Hospital of Pittsburgh of UPMC, Pittsburgh, USA *Corresponding author: Yijen L Wu, Ph.D., Department of Developmental Biology, Rangos Research Center 8121, 530 45th street, 15201, Pittsburgh, PA, USA DOI: 10.26717.BJSTR.2019.14.002507


Introduction
Three-dimensional (3D) printing is an additive manufacturing process to turn digital files into graspable 3-dimentional physical objects [1]. 3D printing from medical imaging data has been gaining wider clinical applications in surgical planning, clinical training, and imaging research [1][2][3]. With the advancement of medical imaging, the 3D computed tomography (CT) and magnetic resonance imaging (MRI) can be obtained in high resolution within a reasonable breath-hold time window [2,4]. However, the visualization of the 3D medical imaging source data is still largely limited to 2-dimentional (2D) multiplanar display, which makes diagnosis and surgical planning challenging. 3D printing can overcome this barrier by turning the multiplanar imaging data into anatomically accurate 3D models, to train new surgeons in controlled environment without risking patient safety, to aid radiologists for better diagnosis [5], and to facilitate patient-specific pre-surgical planning [2,4,6]. The rapid prototyping of the heart by 3D printing from medical imaging data is particularly transformative for managing the congenital heart disease (CHD), because the presentation of the complex CHD is highly variable among individuals [3,[7][8][9][10].
Wide spectrum of anomalies in great vessels, valves, and intracardiac chambers can co-exist or be missing in multiple locations in a single CHD patient, which can make surgical planning very challenging. The individualized structurally accurate physical model of the heart and vessels for each CHD patient can allow individualized comprehensive diagnosis, optimizing planning for surgical intervention, and allowing rehearsals of the surgical procedures to improve precision; consequently, to improve patient outcome. Small animal models, particularly mouse models, of human diseases are indispensable tools for biomedical research.
Genetically engineered transgenic mice and mutant mouse models generated by forward genetic screening with N-ethyl-N-nitrosourea (ENU) chemical mutagenesis are essential in understanding human similar structure and function of visceral organs. More importantly, many genes, gene modifiers, as well as biological processes and molecular pathways are conserved between mice and humans.
Detailed phenotypic characterization by imaging [11][12][13][14] is crucial in the development of animal models, mechanistic studies into disease pathogenesis and testing the efficacy of experimental therapeutics. With the advancement of modern transgenic technology, including CRISPR-Cas9 gene targeting, as well as several national and international programs, such as IMPC, KOMP, IPAD-MD, MGD, and EMMA, systematically knocking out every single gene in the mouse genome, generating unprecedented numbers of mutant mice, it is now possible to pursue detail mapping of genetic functions and pathways in CHD.
Detailed characterization of the structural cardiac and vascular presentation of these genetic mutants can help mapping genetic networks governing heart development [15,16] and their effects on topological manifestation of CHD. We have utilized micro-CT, micro-MRI, and episcopic confocal microscopy (ECM) extensively for ultra-high-resolution imaging and phenotyping of CHD mutant mice [11,12,[17][18][19], especially in fetal and neonatal mice with CHD.
However, the micro-CT, micro-MRI, and ECM images of fetal and neonatal mice are drastically different from human medical imaging in spatial resolution, voxel dimensions, voxel numbers, dynamic ranges, imaging contrast and texture, as well as source imaging data format, because of the minuscule sizes of mouse fetuses. As a result, the routine 3D printing tools developed for human medical imaging are not directly applicable for mouse micro-CT, micro-MRI, and ECM. In this work, we introduce a practical work flow for converting micro-imaging data to 3D printed physical models. We will discuss this practical guide in three sections: imaging acquisition, image post-processing for 3D modeling, and 3D printing.

Micro-CT
Micro-CT is a powerful imaging modality that utilizes x-ray attenuation to acquire high-resolution three-dimensional images from radio-opaque tissues, such as bones. Iodine-based contrast agents can enhance X-ray attenuation in soft tissues, enabling micro-CT imaging with high accuracy [12,[19][20][21]. Formalin fixed mouse embryos or newborn pups stained with iodine ( Figure   1A) by simple immersion in 0.025 N potassium tri-iodide (Lugol) solution can provide excellent contrast, enabling micro-CT scans to be used for performing virtual necropsies [22][23][24]. A typical scan takes approximately 25 minutes, during which four to eight fixed embryos (embryonic day 12-16) or 12 fixed newborn mouse pups can be scanned at a resolution of 15μm or 45μm, respectively. The imaging data collected in this manner ( Figure 1B) can provide remarkably accurate visualization of fetal and newborn mouse anatomy [12,18,19,25]. With the ability to simultaneously image 8-12 fetuses/pups per scan, mutant mice can be readily identified from an entire litter with just a single scan, greatly accelerating the diagnosis of CHD and other structural birth defects.

Micro-MRI
MRI is based on the detection of protons utilizing low-energy radio-frequency (RF) waves in magnetic fields to form images without ionizing radiation. MRI has excellent intrinsic soft tissue contrast because the relaxation characteristics of water protons are governed by the micro-environment of each tissue [11]. With the development of small-board high-field magnets and small micro-RF coils, micro-MRI is capable of very high-resolution 3D imaging of mouse embryos with excellent contrast [26][27][28][29], thus termed "MR Microscopy" [30]. Hence, micro-MRI is a suitable imaging modality for "digital pathology" or "virtual histology" [11,28,31,32] for phenotyping and characterizing mouse embryos without tissue sectioning as conventional histology. Formalin fixed mouse embryos or newborn pups can be soaked in 0.05 % gadolinium (Gd) solution for one to a few days for contrast enhancement prior to imaging. Figure 2 shows multiple mouse embryos harvested on the embryonic day E16.5 imaged by micro-MRI with 33 m X 46 m X 46 m spatial resolution. Congenital structural birth defects can be identified ( Figures 2C-2E).

ECM
Histopathological imaging with episcopic confocal microscopy (ECM) is the ultimate gold standard for validating the diagnosis of birth defects in mouse embryos and neonatal pups. With ECM imaging [12,17,19], the paraffin embedded specimen is sectioned using a sliding microtome with a fixed photoposition, and the block face is serially imaged after each cut using a laser scanning confocal macroscope mounted over the microtome (Figure 3). In this manner, the 2D serial image stack collected of the specimen is in registration, allowing rapid 3D reconstructions to be generated and it is possible to digitally re-slice the serial 2D image stack in any imaging plane. In this manner, it is possible to make a complete assessment of any structural cardiac anomaly, regardless of the embedding plane of the specimen. This ECM histopathology has all the conveniences of micro-CT or micro-MRI data, but with the much higher resolution (5 m) achieved by histology.
On the other hand, micro-CT and micro-MRI allow simultaneous whole-body imaging for multiple mouse embryos or neonatal mouse pups, whereas the ECM can only section and image much smaller sample sizes, only 1 organ/tissue from 1 single embryo at a time.

Image Post-processing for 3D Modeling
Three-dimensional multiplanar images are reconstructed to form 3D lattice, followed by 3D sculpting, surface lasso, contouring, meshing to generate 3D models for 3D printing and quantification. printing. Here we outline the work flow for 3D imaging processing of 3D modeling for micro-CT, micro-MRI, and ECM for 3D printing into three steps: (1) segmentation, (2) surface rendering, and (3) optimization ( Figure 4). Each step is essential to creating a final product, being a 3D printed model that accurately displays the phenotype of a given genetic mouse model, that can then be used to gain a deeper understanding of defects present in genetic models.

Segmentation
Segmentation (Figure 4) is to be carried out in the Horos software and is arguably the most important step of the process with regards to creating an accurate model. It is also where the largest deviation from human and preclinical modality conversion methods manifests. Human modalities have a much better contrast present than that of preclinical modalities, hence the segmentation process is more complicated. Instead of just thresholding and using simple tools to select regions, unwanted areas need to be cropped out by the segmentor. Essentially, it is more difficult to signify to the computer which pixels in the image stack are to be converted into a model (tissue) and which pixels are to be left out (grey space and cavities) and so extra measures must be taken. Hence, a process for segmentation and conversion was developed for ECM data. The ECM scan data which is in .tiff format is loaded into the Horos software using the import data function in Horos. This automatically loads the data in three planes. Interpolation, cropping, and thresholding were used to achieve accurate segmentation in this process. This segmentation process can be ordered into three different steps if it is sufficient to proceed. If it looks destroyed and incorrect, segmentation needs to be re-checked. Once this task is complete, the software will output a file in the filetype of .stl which is then compatible with 3D editing software.  model. In addition to printing physical models, the segmentation and rendering models can provide volumetric information. Figure   6 shows 3D rendering and segmentation outcomes from various mutant mouse brains from micro-MRI source data.

3D Printing
The final step in the process is 3D printing. Once a file is sufficiently prepared in Meshmixer and accuracy is ensured, it can be sent to 3D printer for production. The resultant model should be the exact same as what is viewable on the screen and can be used to better understand the phenotype of a given genetic model. This process generally takes 16-24 hours to complete. We utilized a single-color resin printer that utilizes SLA technology to solidify resin into the print. The models were enlarged to satisfactory dimension to visualize the details and were then printed in transparent resin which was then painted to further highlight the anatomic details.
Upon completing several segmentations, mouse heart models were printed depicting hypoplastic left heart syndrome (HLHS) ( Figure   7). These files were also kept in database to use for future phenotyping of deformities. These saved files and printed models provided further insight into the phenotype of each deformity present in the genetic mouse models. Before this method was developed, specifically for ECM scans, there was no way to create a model that would accurately depict deformities in three dimensions. With this new method, it is now possible to go from an image stack to a 3D model that accurately depicts phenotype, providing further insight into deformities shown in genetic mouse models.

Conclusion and Future Work
We have introduced a practical workflow that can be used to generate 3D printed models of mouse embryonic and neonatal hearts from the ultra-high-resolution ECM, micro-CT, and micro-MRI source data files, to facilitate genetic mapping and deep phenotyping of mutant mouse models for CHD. This pipeline in application for other organ and model systems. Time spent is segmentation is the most limiting factor hampering the highthroughput of the prototyping with 3D printing. Automatic or semi-automatic segmentation algorithm would greatly reduce the application barrier to allow rapid and effective phenotyping.
In addition, the development of a "scripting" module can further speed up and automate the workflow.