The Hong Kong cetacean stranding response program has incorporated postmortem computed tomography, which provides valuable information on the biological health and profile of the deceased animals. This study describes 8 image rendering techniques that are essential for the identification and visualization of postmortem findings in stranded cetaceans, which will help clinicians, veterinarians and stranding response personnel worldwide to fully utilize the radiological modality.
With 6 years of experience in implementing virtopsy routinely into the Hong Kong cetacean stranding response program, standardized virtopsy procedures, postmortem computed tomography (PMCT) acquisition, postprocessing, and evaluation were successfully established. In this pioneer cetacean virtopsy stranding response program, PMCT was performed on 193 stranded cetaceans, providing postmortem findings to aid necropsy and shed light on the biological health and profile of the animals. This study aimed to assess 8 image rendering techniques in PMCT, including multiplanar reconstruction, curved planar reformation, maximum intensity projection, minimum intensity projection, direct volume rendering, segmentation, transfer function, and perspective volume rendering. Illustrated with practical examples, these techniques were able to identify most of the PM findings in stranded cetaceans and served as a tool to investigate their biological health and profile. This study could guide radiologists, clinicians and veterinarians through the often difficult and complicated realm of PMCT image rendering and reviewing.
Virtopsy, also known as postmortem (PM) imaging, is the examination of a carcass with advanced cross-sectional imaging modalities, including postmortem computed tomography (PMCT), postmortem magnetic resonance imaging (PMMRI), and ultrasonography1. In humans, PMCT is useful in investigating traumatic cases of skeletal alterations2,3, foreign bodies, gaseous findings4,5,6, and pathologies of the vascular system7,8,9. Since 2014, virtopsy has been routinely implemented in the Hong Kong cetacean stranding response program1. PMCT and PMMRI are able to depict patho-morphological findings on carcasses that are too decomposed to be evaluated by conventional necropsy. The non-invasive radiological assessment is objective and digitally storable, allowing second opinion or retrospective studies years later1,10,11. Virtopsy has become a valuable alternative technique to provide new insights of PM findings in stranded marine animals12,13,14,15,16. Combined with necropsy, which is the gold standard to explain the pathophysiological reconstruction and cause of death17, the biological health and profile of the animals can be addressed. Virtopsy has been gradually recognized and implemented into stranding response programs worldwide, including but not limited to Costa Rica, Japan, Mainland China, New Zealand, Taiwan, Thailand and USA1.
Image rendering techniques in radiology use computer algorithms to transform numbers into information about the tissue. For example, radiological density is expressed in conventional X-rays and CT. The vast quantity of volumetric data is stored in the Digital Imaging and Communications in Medicine (DICOM) format. CT images can be used to produce isotropic voxel data using two-dimensional (2D) and three-dimensional (3D) image rendering in a postprocessing 3D workstation for high resolution visualization18,19. Quantitative data and results are mapped to transform serially acquired axial images into 3D images with gray-scale or color parameters19,20,21. Choosing an appropriate data visualization method from diverse rendering techniques is an essential technical determinant of the visualization quality, which significantly affects the analysis and interpretation of radiological findings21. This is particularly critical for stranding work that involves personnel without any radiology background, who need to understand the results in different circumstances17. The goal of implementing these image rendering techniques is to enhance the quality on the visualization of anatomical details, relationships and clinical findings, which boosts the diagnostic value of imaging and allows an effective rendition of the defined regions of interest17,19,22,23,24,25.
Although the primary axial CT/MRI images contain most information, they may limit accurate diagnosis or documentation of pathologies as structures cannot be viewed in various orthogonal planes. Image reformation at other anatomically aligned planes permits visualization of structural relationships from another perspective without having to reposition the body26. As medical anatomy and forensic pathology data are predominantly 3D in nature, color-coded PMCT images and 3D reconstructed images are preferred to gray-scale images and 2D slice images in view of improved understandability and suitability for courtroom adjudications27,28. With the advances in PMCT technology, a concern of visualization exploration (i.e., the creation and interpretation of 2D and 3D image) in cetacean PM investigation has been raised12,29. Various volumetric rendering techniques in the radiology workstation allow radiologists, technicians, referring clinicians (e.g., veterinarians and marine mammal scientists), and even laymen (e.g., stranding response personnel, government officers and general public) to visualize and study the regions of interest. Yet, the choice of a suitable technique and confusion of terminology remain a major issue. It is necessary to understand the basic concept, strengths and limitations of the common techniques, since it would significantly influence the diagnostic value and interpretation of radiological findings. Misuse of techniques may generate misleading images (e.g., images that have distortions, rendering errors, reconstruction noises or artefacts) and lead to an incorrect diagnosis30.
The present study aims to assess 8 essential image rendering techniques in PMCT that were used to identify most of the PM findings in stranded cetaceans in Hong Kong waters. Descriptions and practical examples of each technique are provided to guide radiologists, clinicians, and veterinarians worldwide through the often difficult and complicated realm of PMCT image rendering and review for the evaluation of biological health and profile.
NOTE: In the framework of the Hong Kong cetacean virtopsy stranding response program, stranded cetaceans were routinely examined by PMCT. The authors were in charge of virtopsy scanning, data postprocessing (e.g., image reconstruction and rendering), data interpretation, and virtopsy reporting1. This advanced technology emphasizes attentive findings and gives insights on the initial investigation of PM findings prior to conventional necropsy (https://www.facebook.com/aquanimallab).
1. Data preparation
2. Multiplanar reconstruction (MPR)
3. Curved planar reformation (CPR)
4. Maximum intensity projection (MIP)
5. Minimum intensity projection (MinIP)
6. Direct volume rendering (DVR)
NOTE: As 1 of the default display 2×2 interfaces, DVR (upper-right panel) shows the 3D rendered images of the carcass. The default DVR template setting is AAA (abdominal aortic aneurysm; window width: 530, window level: 385), giving a gross skeletal structure of the carcass.
7. Segmentation and Region-of-Interest (ROI) Editing
8. Transfer Functions (TF)
9. Perspective Volume Rendering (PVR)
10. Data evaluation
From January 2014 to May 2020, a total of 193 cetaceans that stranded in Hong Kong waters were examined by PMCT, including 42 Indo-Pacific humpback dolphins (Sousa chinensis), 130 Indo-Pacific finless porpoises (Neophocaena phocaenoides) and 21 other species. A whole-body scan was performed on 136 carcasses while 57 were partial scans on skulls and flippers. Anatomical features and pathologies commonly observed were illustrated with the 8 image rendering techniques for the evaluation of the stranded cetaceans’ biological health and profile.
Figure 1: MPR function displaying a deceased Indo-Pacific humpback dolphin in (A) axial, (B) reconstructed 3D, (C) reconstructed coronal, and (D) reconstructed sagittal views. Area measurements of the atlanto-occipital space are demonstrated in the axial plane. Linear measurements of the ventral tubercle to outer margins of the occipital condyle (coronal), basion-dorsal arch and opisthion-ventral arch (sagittal) for the diagnosis of atlanto-occipital dissociation are demonstrated. Please click here to view a larger version of this figure.
Figure 2: CPR function displaying curved structures in the flipper of a deceased Indo-Pacific finless porpoise in planar view. Please click here to view a larger version of this figure.
Figure 3: MIP function highlighting hyperattenuated pulmonary nodules (intense white dots) in both lungs of a deceased Indo-Pacific finless porpoise. Please click here to view a larger version of this figure.
Figure 4: MinIP function highlighting hyperattenuated gas-filled structures, i.e., tracheobronchial trees in both lungs of a deceased Indo-Pacific finless porpoise. Please click here to view a larger version of this figure.
Figure 5: DVR function displaying different components of a deceased Indo-Pacific finless porpoise. (A) Vasculatures overlaid with the skeletal system are highlighted by AAA. (B) The respiratory system is highlighted by Lung. (C) The skeletal system including the vertebral physeal plates is highlighted by Bone plus Plate. (D) Hyperattenuated ear bones and fish hooks are highlighted by Hardware. Please click here to view a larger version of this figure.
Figure 6: ROI editing function displaying a deceased Indo-Pacific finless porpoise (A) with the CT couch and (B) with the CT couch removed. Please click here to view a larger version of this figure.
Figure 7: TF function displaying different components of a deceased Indo-Pacific finless porpoise. Sand in an air sac is highlighted in cyan. Stomach content is highlighted in green. A parasitic granulomatous mastitis lesion is highlighted in red. Please click here to view a larger version of this figure.
Figure 8: PVR function demonstrating a virtual bronchoscopy of a deceased Indo-Pacific humpback dolphin with the Flythrough function. Please click here to view a larger version of this figure.
Table 1: Keyboard shortcuts of the software for different image postprocessing functions. Please click here to download this table.
For the clear visualization of virtopsy datasets, 8 image rendering techniques, consisting of both 2D and 3D rendering, were routinely applied to each stranded carcass for the PM investigation of their biological health and profile. These rendering techniques included MPR, CPR, MIP, MinIP, DVR, segmentation, TF, and PVR. Diverse rendering techniques are complementarily used together with windowing adjustment. The concepts of each image reformation technique and advantages are also described.
Multiplanar reconstruction (MPR)
MPR is the process of creating non-axial 2D images, including the coronal, sagittal, and any anatomically aligned oblique plane image24,30, which is not acquired directly during the acquisition in an axial plane. This dominant 2D rendering technique is especially helpful in assessing any intact anatomical structure or pathology in the required plane with high quality images31,32. With the help of MRP, cetacean PM investigations of the entire body, orthopedic, and neurological/spine were routinely performed in 3 directions simultaneously, which significantly improved the accuracy of findings (Figure 1). Through comprehensive observation from the 3 planes, the error rate of misidentifying minute pathologies is reduced. In addition, MPR also supports linear and area measurement at the axial, coronal, and sagittal plane. However, it is operator-dependent, and requires sufficient anatomical knowledge to identify both normal structures and pathological conditions, which avoid misinterpretation of the rendered images.
Curved planar reformation (CPR)
CPR is also called curved MPR. Despite being treated as MRP in some peer-viewed literatures, CPR is a distinct 2D rendering technique. Using isotropic imaging that aligns the long axis of the image plane with a selected anatomic structure, 2D images are reformatted with no loss of image quality18,24. This allows the operator to manually define a centerline path for a curved reconstruction within the volumetric dataset. This is particularly crucial when the subject cannot be placed in a true or relatively true anatomical position in reference to the PMCT detectors (i.e., true reconstructed coronal/sagittal/axial image), especially for frozen or mummified carcasses. The alignment of complicated, tortuous or calcified structures is needed to obtain a more symmetrical image for diagnosis. Due to its flexible flattening and distortion characteristics, misinterpretation can be easily induced. The operator must clearly remember the position and shape of the anatomical structures of interest. Flippers are 1 of the most difficult body parts to obtain a true anatomical position as they are curved towards the body flanks, unless resected before the PMCT scan. With the utilization of CPR, most of the anatomical features in the flippers were demonstrated at 1 plane and for skeletal age estimation (Figure 2).
Maximum intensity projection (MIP)
MIP projects only the highest attenuation value in each pixel of the volumetric datasets within the viewer’ sight32 and selects the voxel with the maximum intensity as the value of the corresponding display pixel18. Originally, this technique is recognized to evaluate the osteological material, metallic implants, and contrast-filled structures for CT angiography in clinical radiology antemortem17,33. Due to decomposition of internal structures and organs, and the absence of blood perfusion in stranded carcasses, the adoption of MIP in evaluating the contrast-filled structures for CT angiography become very difficult in virtopsy. However, MIP still takes a dominant character in examining osteological materials, foreign bodies (e.g., food bolus, fish remains, stone, metallic entanglement) and calcifications within soft tissues, as well as highly attenuated, narrow, and blood- or water-filled structures such as the major arteries and veins. Through the adjustment of slab thickness (i.e., image thickness for data reconstruction) subjective to the size of the evaluated target, the visualization of lesions could be emphasized. For instance, using different sliding thin-slab34, the identification of small pulmonary nodules in collapsed lungs of a stranded carcass was intensively improved, as MIP emphasized these minutes of hyperattenuated speckles, which evidenced the presence of lung consolidation and parasitic pneumonia (Figure 3).
Minimum intensity projection (MinIP)
In contrast to MIP, MinIP projects only the lowest attenuation value encountered along a ray pass through a volume toward the viewer’ sight within a volume18,24. Although MinIP is not commonly used in clinical radiology24, this technique still served as an excellent visualization tool on hypoattenuated structures and gas-filled structures, such as the respiratory and gastrointestinal tract. The examination of the morphology and pulmonary parenchymal abnormalities, started from the blowhole down to the tracheobronchial tree, in the stranded cetaceans were significantly enhanced (Figure 4). Similar to MIP, additional control should be taken on the slab thickness, subject to the examined pathologies, to generate a more distinguishable image35, as the slab thickness is critical to determine the distinction of presented structures on the studied structures.
Direct volume rendering (DVR)
DVR is an algorithm that converts an entire 3D image set into 2D images directly without discarding any information18. The final displayed 2D image is created based on its Hounsfield units by assigning each voxel in the image a specific color and opacity value along with other voxels in the same projection ray. As the opposition of creating an intermediate representation (e.g., an extracted surface model by soft-tissue removal tool), the internal and external conditions of a stranded carcass at all depths with the 3D method can be examined at once, without obscuring each other. This 3D rendering technique was a quick, versatile and interactive tool for a whole-body carcass evaluation from any angle. Bony lesions, complex fractures, body fragmentation, and foreign bodies caused by human interaction (e.g., traumatic injuries caused by vessel collision and fisheries) were possible to be identified (Figure 5). The challenge of DVR is that the operator needs to adjust the rendering parameters, i.e., the opacity and brightness, to display the vasculature more accurately21,36.
Segmentation and Region-of-Interest (ROI) Editing
Irrelevant structures, objects (e.g., body bag and CT couch), and artefacts (e.g., metallic zippers) displayed on the DVR model may degrade the image quality and obscure radiological diagnosis. To illustrate certain areas of anatomy or pathology in a better manner, segmentation is used to include or exclude selected volumetric data on either 2D or 3D images18,24. Although automated segmentation programs are available, manual segmentation which requires high tissue recognition and delineation by the operator was performed in most circumstances to assist the identification of radiological findings on the DVR of stranded carcasses. ROI editing was the most common segmentation tool used in the present study, which allowed the operator to include or exclude a region of interest manually by drawing a rectangular, elliptical or other shape to define the precise spatial boundary of the target (Figure 6). Similar to DVR templates provided in the 3D workstation, automated segmentation is based on the rules of connectedness and thresholding, and subjected to clinical radiology, which was mostly unsuitable for this study, except for the automatic body bone removal function.
Transfer Functions (TF)
TF is an algorithm to control the threshold of opacity, brightness, and color of the selected volume18,24. This tool allows the operator to selectively reveal the relevant structures on the DVR model, by selecting the threshold value, range, and shape, to serve different purposes at the defined region. For instance, choosing a lower opacity threshold removes the external low-opacity soft tissues (skin and fat) and obscures the abdominal content, while a high opacity threshold keeps high opaque objects (e.g., bone, calcium, and excreted contrast materials); changing the color, brightness, and contrast scale highlights the region of interest, and makes the appearance of the DVR model to look different. These controls give a better elucidation and quicker differentiation of structures based on their attenuation. However, these are vulnerable to interobserver variability and dependent on operator mastery in optimization of rendering parameters21. With the contribution of segmentation and TF, the relationship of displayed tissues, organs, and foreign bodies in scanned carcasses were well-classified (Figure 7). Fast and clear preliminary findings on stranded cetaceans were demonstrated on the edited DVR model, which gave veterinarians and stranding response personnel an overview on the internal and external condition, as well as the initial PM investigation findings, and facilitated subsequent conventional necropsy.
Perspective Volume Rendering (PVR)
PVR, also called endoluminal imaging or immersive rendering, is mainly applied to air-containing structures such as trachea, colon, esophagus, and arteries. It allows the operator to visualize the internal conditions of the lumen by virtual navigation35. The operator designates the start point, end point, and a centerline path to fly through. By displaying an animation of flying through the structure, the relationships between anatomic structures and endoluminal abnormalities such as polyps or cancerous growths on the walls can be identified as in a non-invasive virtual endoscopy19. The corresponding MPR images displayed alongside allow concurrent reviewing of particular lesions37,38. By extending PVR beyond the lumen, adjacent extraluminal structures can also be visualized24. In the present study, PVR was only applicable on fresh carcasses with uncollapsed structures, which permitted the reconstruction of the endoluminal view (Figure 8).
In the present overview of rendering techniques, only 8 techniques commonly used in the routine virtopsy of stranded cetaceans were described, while others were disputed due to their limited usefulness. The techniques mentioned could also give insight and be applied to other animals in general. In clinical radiology, there are many other rendering techniques and DVR templates, built on threshold-based algorithms with preset values for opacity, brightness, lighting, heat scale, window level and window width, provided in most 3D workstations. Those are designed to emphasize the illustration of different tissue types and body parts for special examinations, for instance, vascular contrast, airways, stomach or thrombus18,24,31. However, in the case of stranded carcasses, there is gas accumulation caused by decomposition with no organ perfusion. Most DVR presets of clinical CT examination, especially CT angiography, require contrast injection and thus could not be applied in the present study. The self-designed DVR templates combined with single or multiple DVR models for cetacean PM investigation could be established after standardization of the threshold-based algorithms in terms of species and their level of decomposition. Nevertheless, based on our experience, the 8 rendering techniques listed were able to identify most of the PM findings in stranded cetaceans, and were sufficient to investigate their biological health and profile.
Preparation and scanning of carcasses is critical for subsequent postprocessing and visualization of virtopsy data. Operation of a CT machine, an ionizing radiological unit, must be performed by a certificated radiological technician or clinician in compliance with the law. Although the scanned subjects were carcasses, the radiation dose should be kept to as low as reasonably achievable. The control of scanning parameters, especially slice thickness, would highly influence the accuracy of the reconstructed coronal and sagittal planes. Moreover, reduction in CT slice thickness permits more precise diagnosis. For instance, acquiring PMCT images at 3 mm thickness may neglect a 1×1×1 mm parasitic granuloma, commonly observed in the mammary glands of stranded cetaceans. To avoid missing any finding and improve the resolution of 2D and 3D rendering, a standardized scanning protocol was used. The slice thickness was controlled at 1 mm, and down to 0.625 mm whenever possible, which is the minimum slice thickness available for the CT machine used.
A proper postprocessing visualization and manipulation of virtopsy datasets requires clear understanding of the principles and pitfalls of the common rendering techniques used for cetacean PM investigation, e.g., the identification of strength and weakness between the techniques21. The choice of rendering techniques depends on the anatomical structures and the underlying pathologies to be illustrated, there is no single technique that can comprehensively recognize all the PM findings. Knowing the pros and cons and choosing the appropriate rendering techniques can boost image quality and interpretability of virtopsy datasets, which aid to obtain a correct diagnosis. Carefully reviewing virtopsy datasets and correlating them with other techniques can avoid potential rendering and segmentation error18. Still, the final judgement and diagnosis should be made by veterinary radiologists or radiological clinicians who are certificated and experienced to report virtopsy findings.
The authors have nothing to disclose.
The authors would like to thank the Agriculture, Fisheries and Conservation Department of the Hong Kong Special Administrative Region Government for the continuous support in this project. Sincere appreciation is also extended to veterinarians, staff, and volunteers from the Aquatic Animal Virtopsy Lab, City University of Hong Kong, Ocean Park Conservation Foundation Hong Kong and Ocean Park Hong Kong for paying great effort on the stranding response in this project. Special gratitude is owed to technicians in CityU Veterinary Medical Centre and Hong Kong Veterinary Imaging Centre for operating the CT and MRI units for the present study. Any opinions, findings, conclusions or recommendations expressed herein do not necessarily reflect the views of the Marine Ecology Enhancement Fund or the Trustee. This project was funded by the Hong Kong Research Grants Council (Grant number: UGC/FDS17/M07/14), and the Marine Ecology Enhancement Fund (grant number: MEEF2017014, MEEF2017014A, MEEF2019010 and MEEF2019010A), Marine Ecology Enhancement Fund, Marine Ecology & Fisheries Enhancement Funds Trustee Limited. Special thanks to Dr. María José Robles Malagamba for English editing of this manuscript.
Aquarius iNtuition workstation | TeraRecon Inc | NA | |
Siemens 64-row multi-slice spiral CT scanner Somatom go.Up | Siemens Healthineers | NA |