Remember me
Abstract
Introduction : Preserved congenital heart specimens are an important component of training professionals working with children and adults with congenital heart disease. They are curated in few institutions worldwide and not freely accessible. This was a proof-of-concept project to explore the use of advanced cardiac imaging modalities (computed tomography [CT] and magnetic resonance imaging [MRI]) and virtual reality (VR) simulation to assess the feasibility and identify the best method of imaging curated cardiac pathology specimens.
Methods : Seven specimens in glass jars with formalin, with varied anatomic lesions, from a curated collection were imaged using MRI and high-dose CT to compare the fidelity of models created via each modality. Three-dimensional (3D) models were created and loaded into a VR headset and viewed in virtual space. Two independent physicians performed a “virtual dissection” and scored the resultant models.
Results : The highest fidelity and tissue characterization of more delicate structures was achieved with T2 spoiled gradient-echo sequences on MRI (median score of 4 out of 5). CT (median score of 3), while excellent for external anatomy, lost some fidelity with delicate internal anatomy, even at high-radiation doses. No specimens were damaged.
Conclusions : We believe that in vitro heart specimens can be easily scanned with high fidelity at a relatively low cost, without causing damage, using high-dose CT and MRI. The ability to “walk through” different chambers of the heart makes the understanding of anatomy easy and intuitive. VR and 3D printing are technologies that could be easily adapted to digitize preserved heart specimens, making it globally accessible for teaching and training purposes.
Keywords: Digitization, preserved heart imaging, three-dimensional model, virtual reality
How to cite this article: Introduction 
Cardiac morphology is an integral part of training in congenital heart disease (CHD). Understanding complex developmental cardiac defects with their spatial anatomy within the chest cavity presents a challenge to trainees and other health-care personnel working with children and young adults with CHD. Traditionally preserved cardiac specimens are used to study and understand cardiac anatomy. However, many of these are beginning to disintegrate and replacements for them are hard to come by, due to both CHD being diagnosed and operated early, and due to improved survival. The rates of autopsy are also on the decline due to social and legal limitations. Beyond this, there has also been a chronic state of limited access to these specimens, being housed and curated in only a few institutions around the world. In 2015, there was a passionate plea from various morphologists, surgeons, and cardiologists to help preserve this valuable resource for future posterity.[1] Hence, it is imperative to preserve these specimens by digitization while maintaining the highest possible fidelity.
There is an exponential growth of newer digital technologies largely driven by the gaming industry. Some of these technologies such as virtual reality (VR) and augmented reality have found their way into health care, and are being used in the fields such as rehabilitation and psychiatry.[2],[3] Many heart centers have now started to use three-dimensional (3D) heart models, with VR and AR, to assist catheter-based interventions and surgical planning including ventricular assist device fit testing, and education.[4],[5],[6],[7] There is limited literature, however, about the use of advanced imaging and VR to image preserved specimens.
The aim of this proof-of-concept project was to assess the feasibility of and create simple protocols for imaging of preserved in vitro cardiac pathology specimens with the available computed tomography (CT) or cardiac magnetic resonance imaging (cMRI) and then exporting the images to the VR space to create a digital library.
Methods The study was developed as a collaborative effort between the University of Texas Southwestern, Dallas, TX, USA, and Frontier Lifeline Hospital, Chennai, India. The preserved study heart specimens were in glass containers and had clear labels identifying the pathology, with de-identified patient data available for review [Table 1], and were selected from among Dr. K. M. Cherian's personal collection. Written consent for handling and imaging was previously obtained and verified by the surgeon who provided the specimens.
Imaging was performed at a privately owned, advanced imaging center in Chennai, India. A pathology technician ensured careful handling of the specimens.
Pilot scans (both CT and cMRI) of a fresh pig heart and a preserved human heart, both in formalin, were performed to assess the initial feasibility and logistics. As the resultant images for both specimens were poor and had multiple artifacts due to the tissue formalin interface, the formalin from the jars was drained (by drilling two burr holes on the lid) without disturbing the mounted specimen and re-imaged. Following this, the jars were re-filled with fresh formalin.
Following this initial exercise, all seven test specimens were imaged using both modalities of advanced imaging – CT and magnetic resonance imaging (MRI) for comparison of image fidelity.
CT scans were obtained using a GE LightSpeed VCT 64-slice scanner at high-radiation doses – voltage of 150 kV (maximum on the machine), and a current between 580 and 600 mA. The selected and prepared specimen jars (formalin emptied) were lined up in erect position and scanned in two batches.
Cardiac MRI sequences were obtained using a 1.5T Siemens Avanto Scanner and an eight-channel receiver head and body coil. Multiple sequences were obtained which included T1-weighted spin-echo sequences, T2-weighted spin-echo sequences, 3D balanced steady-state free precession sequences, and 3D T1-weighted gradient-echo sequences [listed in [Table 2]]. Isotropic images were acquired at voxel sizes of 0.7 mm. Nonmagnetic material like cotton and plastic was used to stent cavities and vessel lumen open.
Table 2: Various magnetic resonance imaging test sequences with the imaging parameters used with details of the resulting images and recreated modelsThe Digital Imaging and Communications in Medicine (DICOM) images thus obtained were securely stored on encrypted and password-protected cloud storage. 3D datasets were then volume rendered using the Mimics Research 21.0 software (Materialise NV, Belgium). Mimics 3-Matic Research 13.0 software (Materialise NV, Belgium) was used to further process the 3D models to remove artifact and create 3D Stereolithography (STL) and 3D Portable Document Format (PDF) files. STL file formats were loaded using the VR software (Varyfii Imaging LLC, Dallas, Texas) onto an HTC Vive Headset connected to a VR-capable computer that was used to visualize the 3D heart models in VR space. An overview of this process is shown in [Figure 1]. The two models created from CT and cMRI for each specimen were then compared based on an objective scoring scale from 1 to 5 [Table 3] depending on the quality and fidelity of the reconstructed 3D virtual specimen obtained. Models were reviewed by two independent observers (a cardiologist and a cardiac surgeon), and the average scores were taken. The current repository for the DICOM data and virtual models is a secure network research drive on the UTSW intranet with limited access.
Figure 1: Overview of the process of imaging and virtual 3D model generation. 3D: Three-dimensional
Table 3: Objective scoring system for characterization of image quality, and to help compare imaging methods Results Image acquisition technique
Initial tests of T1 spin-echo sequences with specimens in formalin-filled jars produced lower quality image results, with small air bubbles within the specimen causing a large artifact. T1 sequences were repeated with the formalin from the jars drained, which produced better images. These images demonstrated the surface anatomy well, but on virtual dissection, the image quality of internal structures was very poor. T2 spin-echo sequences produced excellent images with a voxel size of 0.7 mm on the test specimens in the formalin-filled jars, while the images obtained with fluid drained were of poorer quality. Higher resolution images (smaller voxel sizes of up to 0.56 mm) were attempted but abandoned during the testing period as they had shown a drop in the image quality. When loaded into the 3D virtual space and dissected, tissue differentiation was noted to be more detailed and of better quality on T2-weighted spin-echo sequences. The best quality 3D models in the virtual space were obtained using a 3D spin-echo sequence of SK/SP variant with PFP scan option at a slice thickness of 0.6 with a repetition time (TR) of 1100 msec, echo time (TE) of 254, with 2 averages, with 100-phase field of view using a body coil, acquisition matrix set to 0/320/318/0, and a flip angle of 150. Sequences obtained with nonmagnetic material (i.e., cotton) used to fill cavities showed difficulty determining the cotton-myocardial border and artifact from air bubbles within the filler material.
CT performed with formalin in the jar produced poor tissue–fluid interface and hence poorer quality models, when loaded into virtual space. The formalin-drained specimens imaged better and produced easier-to-segment images with resultant higher quality 3D models.
Postprocessing
The postprocessing of DICOM data varied from 30 min to 1 h depending on the complexity of the anatomy. Smaller sized heart specimens (e.g., premature neonatal hearts) were more technically challenging to segment due to the poorer resolution when enlarged.
Image and three-dimensional virtual model quality
A comparison of scores per imaging modality, as listed in [Table 4], showed overall better image quality for T2-weighted spin-echo cMRI imaging, as exemplified in [Figure 2] and [Figure 3]. There was excellent inter-observer agreement with disagreement only on the CT for specimen 5 (with a 2-point difference). Intra-observer agreement could not be reliably tested due to the small number and recent graphical memory. Finer structures such as semilunar valves, papillary muscles, and atrioventricular (AV) valvular apparatus were typically only imaged on the T2-weighted spin-echo cMRI sequences [Figure 4] and [Figure 5]. The ability to perform virtual dissections on both the computer-generated models [Figure 6] and in VR [Figure 7] was instrumental in understanding phenotypic characteristics of morphological lesions such as a ventricular septal defect or transposition of the great arteries [Figure 3]c, [Figure 6], and [Figure 8].
Figure 2: (a and b) A specimen of a perimembranous VSD (a) imaged using high-radiation CT shows a well-recreated external anatomy but with loss of fidelity with finer structures such as AV valves, myocardial trabeculations, and papillary muscles (scale displayed is in centimeters). VSD: Ventricular septal defect, CT: Computed tomography
Table 4: Scoring for the three-dimensional model of each specimen imaged by magnetic resonance imaging versus computed tomography
Figure 3: (a and b) Another specimen of a perimembranous VSD imaged while in the container (a) using the MRI T2 spin-echo sequence which shows better fidelity of internal tissues such as the septal and mural tricuspid valve leaflets with the valvular apparatus (b). The perimembranous VSD is noted partially covered by the tricuspid valve (orange arrowhead) (scale displayed is in centimeters). The same 3D model when virtually dissected through the interventricular septum (black arrowhead) shows the membranous continuity of the tricuspid (orange arrowhead) and mitral (green arrowhead) valves (phenotypic feature of a perimembranous VSD) (c). VSD: Ventricular septal defect, MRI: Magnetic resonance imaging, 3D: Three-dimensional
Figure 4: A look down at the ascending aorta at the normal aortic valve shows three well-imaged cusps and sinuses
Figure 5: Another model showing a large ASD viewed for the left atrial side (orange arrowhead) and excellent fidelity of the mitral valvular apparatus including the leaflets, chordae, and papillary muscles (green arrowheads) using a T2 spin-echo cMRI sequence. ASD: Atrial septal defect, cMRI: Cardiac magnetic resonance imaging
Figure 6: (a and b) A cMRI of a preserved dissected normal heart (a) showing the broad-based right atrial appendage (black arrowhead), with a normal posterior and rightward arrangement of the aorta (green arrowhead) and anterior and leftward arrangement of the pulmonary outflow (orange arrowheads). A depiction shows how the same heart can be virtually sliced into many layers for easy visualization of intracardiac structures (b). cMRI: Cardiac magnetic resonance imaging
Figure 7: An example of a heart as viewed in virtual space shows how these hearts can be labeled, manipulated, and virtually dissected in a manner ideal for learning segmental anatomy and to perform a walk “through” the heart
Figure 8: A virtually dissection of a neonatal heart with (S, D, D) transposed great arteries (a), shows the characteristic parallel arrangement (b) of the anterior rightward aorta (orange arrowhead) and the posterior leftward pulmonary arteries (green arrowhead). The RV cavity and papillary muscles are seen (blue arrowhead). The LV cavity was collapsed which prevented visualization of the mitral-pulmonary continuity. RV: Right ventricleThe vessels and chambers were collapsed in some specimens. While the proximal great vessel anatomy was well depicted, the head and neck vessels and branching patterns were not well discerned on the virtual 3D models due to the way the specimens were curated.
Discussion In this study, we attempted a foray into the uncharted territory of in vitro imaging of preserved heart specimens with CHD and exporting digitized data onto a virtual platform. Jutras performed virtual autopsy on preserved heart specimens using cMRI, and was able to identify a detailed study of the specimens with 3D image processing of all data sets.[8] Kiraly et al. have been working on creating a virtual library of their 400 congenital heart specimens using high-resolution micro-CT and MRI.[9] They reported their initial feasibility study using six test specimens that showed promising results but concluded that more refinement was needed to discern the detailed intracardiac anatomy of more delicate structures. Kaneko et al.[10] have used X-ray phase-contrast tomography and Shinohara et al.[11] have used synchrotron radiation-based phase-contrast CT to image normal infantile and fetal heart specimens to acquire high-quality images showing excellent resolution of the myocardium, coronary arteries, and conduction system. However, the above imaging technologies are not easily available. We were able to develop routine cardiac MR and CT sequences using easily accessible imaging technology, while in a developing country setting. Our study also included the use of a VR platform to study these specimens, on which there is very little written in literature. As suggested by the images here, learners can virtually visualize, interact, dissect, and “walk through” the various structures in the heart in order to understand normal and abnormal heart morphology. This is extremely important for early-career trainees who are learning to differentiate the various phenotypes of lesions such as ventricular septal defects and double-outlet right ventricular morphologies, and understand complex relationships of cardiac anatomic structures. This feasibility study hence succeeded reasonably in showing the finer anatomical details, and offers a foundation for future studies as strides are made in imaging and VR technology.
Some of the challenges of in vitro imaging we identified are: (1) there is significant alteration of anatomy due to great vessels being collapsed, atrioventricular, and semilunar valves stuck to the surrounding tissue; (2) loss of differential medium thus causing artifacts; and (3) loss of ventricular fullness and shape. Formalin-preserved specimens are also dehydrated with loss of all water content making tissue differentiation, particularly challenging. Better fidelity of internal cardiac structures and better soft-tissue differentiation was achieved with cMRI, which correlates with in vivo studies. In patients with CHD, CT is restricted in image quality partly due to the physics giving poorer tissue fidelity but also due to inability to use higher doses of radiation with risk of toxicity. We anticipated that the high-dose radiation CT that we used to hence produce images would match in quality and fidelity with cMRI-generated images. This, however, was not the case, and we hypothesize that this did not have to do with the image fidelity itself but rather with the higher soft-tissue resolution and contrast ratios noted on cMRI T2 sequences which made thresholding and segmentation of the heart during postprocessing much easier and resulted in higher definition on 3D models. The CT images were harder to segment into 3D models as we noted that small changes in threshold values caused a major shift in the ability to visualise AV valve apparatus. On resulting virtual 3D models, interior and exterior surface anatomy was excellent, albeit a little “smooth” as the images had to be “over-thresholded” to include more delicate tissues like AV valve apparatus. This caused the 3D models to be smoother, and hence, less detailed, accounting for lower scores for CT-derived heart models. Automatic and artificial intelligence segmentation techniques may help improve this in the future. This would also help reduce the time taken to manually segment these studies, which can be quite taxing and cumbersome when done on a larger scale. CT scanning, however, has many advantages over cMRI with regard to speed of imaging, imaging of multiple specimens at the same time, cost, and technical expertise. These advantages are more stark when imaging large number of specimens.
Digital images and 3D models cannot reproduce the texture, the nostalgic olfactory experience of formalin, or the hands-on pliable experience of handling actual pathological heart specimens. However, with the delicate nature of precious older and rarer specimens, handling and preservation are challenging. This method hence allows not only to create a learning experience but also a way to preserve these valuable specimens for eternity. Future technologies such as 3D photography and digital optical scanning are other technologies that may work better in conjunction with cross-sectional imaging techniques.
While STL file formats could only be run and viewed in specialized software, 3D PDF is supported by virtually any device including Windows, Mac, Linux, Android, and iOS-based smartphones, and are usually <5 Megabytes (Mb) in size. This also allows for sharing of deidentified images via the Internet and secure cloud storage. Images could then be shared with anyone all over the world. This is revolutionary in advancing the field of cardiac morphology, and CHD. Pathological dissection planes and mounting in the glass jars are designed to show the heart lesion (e.g., ventricular septal defect) in the best possible manner. This, however, is not the ideal way to look at the heart in its native form as it lies within the body. We expect that with improvements in technology and software designed specifically for this, learners would have the opportunity to play around with the model as experienced in the VR space [Figure 7].
Acknowledgment
The authors would like to thank the cardiac pathology team and the curators of the cardiac pathology museum at Frontier Lifeline Medical Center, which houses the largest collection of heart specimens in the world.
Financial support and sponsorship
The study was funded by a discretionary research account (Project ID #60000085) in the Cardiovascular and Thoracic Surgery Department at the University of Texas Southwestern.
Conflicts of interest
Animesh Tandon and Tarique Hussain are co-developers and owners of the VR software – Varyfii Imaging LLC, Dallas, Texas. The other authors declare no conflict of interest.
References 
Correspondence Address:
Dr. Ravi V Vegulla
Childrens Medical Center, Heart Center, 1935 Medical District Drive, Dallas, TX 75235
USA
Source of Support: None, Conflict of Interest: None
CheckDOI: 10.4103/apc.apc_176_21
Comments (0)