Revolutionizing Sectioned Images of the Visible Korean: Achieving From 0.05 mm to 0.03 mm Sized Pixel

Abstract

Background

The Visible Korean (VK) project aimed to overcome the limitations of traditional cadaver dissection by producing high-resolution sectioned images. This study compared a newly acquired 2024 dataset with a 2023 dataset to evaluate the feasibility of using sectioned images for clinical anatomical research.

Methods

Two female cadavers were frozen and serially milled at constant intervals. The sectioned surfaces were photographed using a digital single-lens reflex (DSLR) camera of 35 mm charge-coupled device (CCD) sensor for 2023 dataset and a DSLR of 44 mm medium-format CCD sensor for 2024 dataset.

Results

The 2024 dataset had 2.78 times higher pixel density (pixel size, 0.03 mm × 0.03 mm) than the 2023 dataset (pixel size, 0.05 mm × 0.05 mm), enabling clearer visualization of fine structures such as brain nuclei, fasciculi, vessel walls, meninges, and facial retaining ligaments. Pathological findings including cardiomegaly, vascular rupture, atherosclerosis, and endometrial neoplasm were clearly identified. The consistent intervals allowed accurate tracing of structures across multiple planes. The accumulated datasets from twelve cadavers now allow statistically reliable anatomical interpretation.

Conclusion

Sectioned images produced with modern DSLR technology can reproduce anatomical structures at a level comparable to cadaveric dissection. Their high resolution, color fidelity, and digital format allow repeated analysis without damage. These datasets are now suitable not only for education but also for clinical anatomy research, potentially contributing to a paradigm shift from traditional to digital dissection.

Graphical Abstract

INTRODUCTION

In 1996, the Visible Human Project (VHP) was launched in which a fresh-frozen cadavers in the United States. The cadavers were milled serially at 1.0 mm (male) and photographed using a semi-digital camera.1 Since film cameras dominated the era, a semi-digital system combining a Hasselblad 553 ELX (film camera) with a Leaf camera back (digital image processor) was employed to produce true-color (24 bits color) and high resolution (2,048 × 2,048 pixels) sectioned images. These were the first digital images to present the natural coloration of body structures in serial sections1,2 unlike computed tomography and magnetic resonance imaging.

In 2000, we recognized the need for new sectioned images because the image quality of the VHP dataset was insufficient for detailed study of gross and neuroanatomical structures.3 The semi-digital camera used in the VHP was the primary factor limiting image quality. Around that time, full-frame digital cameras were developed, with the Kodak DCS 560 being the first digital single-lens reflex (DSLR) camera equipped with a full-frame 35 mm charge-coupled device (CCD) sensor.

Therefore, we initiated the Visible Korean (VK) Project in 2000 using the Kodak DCS560 equipped with a 35 mm CCD sensor.3 From the first cadaver in 2000 to the ninth cadaver in 2023, we used a total of four DSLRs with 35 mm CCD sensor.4 Because of these cameras, we could produce the sectioned images of much higher quality than the VHP. Based on these images,4 we developed and presented various useful results; Browsing software,5,6 three-dimensional (3D) surface models,7,8,9 3D volume models,10,11 and average phantom models.12,13,14,15 Nevertheless, further improvements in resolution were desired to enable even more detailed anatomical analysis. Finally, in 2024, Fujifilm^TM released a new DSLR equipped with a 44 mm medium-format CCD sensor, providing the potential for superior image quality.

Furthermore, in clinical anatomy research, human anatomical morphology has traditionally been studied using cadaveric dissection. These dissections have revealed not only standard morphology but also variations within normal function, which have been published in major anatomical and clinical journals.16,17 However, cadaver dissection has fundamental limitations, for emerging structures, the surrounding tissues must often be altered or removed, sometimes even eliminating the target itself. Decomposition and the need for cremation prevent repeated examination of the same specimen. To overcome these limitations, image-based dissection serves as the only viable alternative. Because it relies on digital data, no anatomical structure is damaged during analysis. The most critical factor in image-based dissection is image quality, specifically color fidelity and spatial resolution.

This study aimed to compare and evaluate the resolution of the sectioned images produced using a new medium-format digital camera equipped with a 44-mm CCD sensor, against those produced using previous digital cameras with a 35-mm CCD sensor. Through this comparison, we sought to demonstrate the feasibility and practicality of the new sectioned images for studying basic and clinical anatomy.

METHODS

The sectioned images of third female whole body in 2023

From the two female cadavers, first and second sectioned images were released in 20157 and 2023, respectively.8 In 2023, we received a female cadaver (73 years old) who had died of tracheostenosis. The cadaver and gelatin solution with methylene-blue were put into embedding box, and frozen immediately after death. The embedding box was sectioned serially. These methods and techniques for sectioning the cadaver followed those established in our previous studies.3,18 The only difference in this study was that the sectioning intervals of head to trunk and lower limb were 0.5 mm and 1.0 mm respectively.

The sectioned surfaces of field of view with 434 mm (width) × 289 mm (height) were photographed using a Canon™ EOS 5DsR DSLR of resolution of 8,688 × 5,792 (50,320,896) pixels (full-frame CMOS image sensor, sensor size of 36 mm × 24 mm, ISO 200, shutter speed of 1/250, aperture of F/16, manual focus, released in 2015) with a Canon™ EF Macro IS USM 50 mm lens. So, the sectioned images with a pixel size of 0.05 mm × 0.05 mm were acquired (Table 1).

Table 1. Information on raw datasets of sectioned images from tenth and eleventh cadavers.

Female cadaver	Age, yr	Cause of death	Region	Intervals, mm	Pixel size, mm	Resolution	A file size, MB	No.	Total file size, GB
2023 dataset	73	Tracheostenosis	Head to perineum	0.5	0.05 × 0.05	8,688 × 5,792	287	1,969	553
			Lower limb	1.0			287	786	221
2024 dataset	82	Heart failure	Head to perineum	0.5	0.03 × 0.03	11,648 × 8,736	582	1,836	1,034
			Lower limb	1.0			582	656	373

Color depth of all images is 48 bits color.

The sectioned images of fourth female whole body in 2024

In 2024, we received a female cadaver who was 82 years old and had died of multiple organ failure caused by heart failure. The processes of freezing and sectioning the cadaver were carried out using the same methods as in previous studies.3,18 The sectioning intervals of head to trunk and lower limb were 0.5 mm and 1.0 mm respectively, which were identical to those used in the 2023 experiment.

We prepared DSLR with 44-mm medium-format CCD sensor which was a Fujifilm^TM GFX100S II (medium-format CMOS image sensor, sensor size of 43.8 mm × 32.9 mm, resolution of 11,648 × 8,736 [102,000,000] pixels, released in 2024) with Fujifilm^TM GF 63 mm 2.8R WR lens.

The sectioned surfaces of field of view with 442 mm (width) × 332 mm (height) were photographed using the Fujifilm DSLR of resolution of 11,648 × 8,736 pixels (ISO 100, shutter speed of 1/160, aperture of F/8, manual focus). Consequently, the sectioned images were produced with 0.03 mm × 0.03 mm pixel size were made (Table 1).

Ethics statement

For this study, we permitted this research from the Institutional Review Board (IRB) of Ajou University School of Medicine and the requirement for informed consent was waived (IRB No. AJOUIRB-UD-2024-545).

RESULTS

Feasibility of sectioned images in reproducing anatomical structures comparable to cadaveric dissection: unparalleled resolution of 2024 dataset

The sectioned images had a pixel size 0.05 mm × 0.05 mm in the 2023 dataset and 0.03 mm × 0.03 mm in the 2024 dataset. A pixel area was 0.0025 mm² in 2023 and 0.0009 mm² in 2024, corresponding to approximately 400 pixels/mm² and 1,111 pixels/mm², respectively. Therefore, a resolution of 0.03 mm per pixel provided approximately 2.78 times more pixels per unit area (1 mm²) than 0.05 mm per pixel (Fig. 1, Table 1), enabling the detection of much finer anatomical structures.

Fig. 1. Comparison of anatomical findings of the 2024 and 2023 datasets. All sentences describe only the 2024 dataset. (A) In the cerebrum, the six layers of the lateral geniculate nucleus are clearly identified. In deep brain of cerebrum, (B) lenticulostriate arteries are clearly distinguished and (C) each fibers of subthalamic fasciculus is observed. (D) In the eyeball, the subarachnoid space between the pia mater and arachnoid mater around the optic nerve is visible. (E) In the cheek, the platysma-auricular ligament, classified as a retaining ligament, is clearly observed attaching to the platysma. (F) In the thoracic spinal cord, the gray matter is distinctly separated from the white matter. (G) In the heart, atrioventricular branches emerging from the atrioventricular node are visible. (H) In the abdomen, the mucosa, submucosa, and serosa of the circular folds of the jejunum are clearly distinguished.

The corresponding voxel size was 0.000125 mm³ (pixels size and intervals, 0.05 mm × 0.05 mm × 0.05 mm) in the 2023 dataset and 0.000045 mm³ (pixels size and intervals, 0.03 mm × 0.03 mm × 0.05 mm) in the 2024 dataset. Accordingly, the voxels per unit volume (1 mm³) was 8,000 for the 2023 dataset and approximately 22,222 for the 2024 dataset. This represents a 2.78-fold increase in voxel density, enabling more detailed and accurate 3D reconstructions of anatomical structures (Table 1).

The following descriptions highlight anatomical structures demonstrating the superior resolution of the 2024 dataset compared with the 2023 dataset. The following descriptions refer exclusively to 2024 dataset.

In cerebrum, gray matter (six layers of lateral geniculate nucleus) and white matter of deep brain (metathalamus) were clearly distinguished in 2024 dataset (Fig. 1A). In deep brain of cerebrum, even the walls of blood vessels were clearly distinguished (Fig. 1B). Subthalamic fasciculus between subthalamus and globus pallidus was clearly visualized (Fig. 1C). The meninges surrounding the optic nerve in orbit were visualized (Fig. 1D). The facial retaining ligament around the masseter could be clearly distinguished from the subcutaneous tissue (Fig. 1E). Not only anterior and posterior horns but also lateral horns of the gray matter in the spinal cord were clearly defined (Fig. 1F). Despite being difficult to identify in cadaveric dissection, the atrioventricular node and bundle in the heart were clearly visualized (Fig. 1G). In the jejunum, the circular folds consisting of submucosa and mucosa were clearly distinguished (Fig. 1H).

Feasibility of sectioned images in reproducing anatomical structures comparable to cadaveric dissection: pathological findings and anatomical variations

The sectioned images enabled comprehensive anatomical evaluation of both disease-related morphological changes and structural variations. For example, the 2024 cadaver exhibited cardiomegaly, which resulted in dilation not only of the heart but also of systemic blood vessels (Fig. 2).

Fig. 2. Comparison of pathological findings of the 2024 and 2023 datasets. (A, B) In the 2024 dataset, the internal jugular veins are enlarged and the venous plexus shows rupture with blood leakage. (C) Arteries and veins emerging from the heart are dilated, and the thoracic aorta and esophagus are displaced leftward and posteriorly in the 2024 dataset. These findings are normal in the 2023 dataset. (D) Due to cardiomegaly, both ventricles contact the thoracic wall in the 2024 dataset, unlike the normal positions in the 2023 dataset. (E) The celiac trunk arises from the left side of the aorta, and blood vessels are dilated in the 2024 dataset, whereas normal anatomy is observed in the 2023 dataset. (F) Inferior vena cava and portal vein are dilated in the 2024 dataset but appear normal in the 2023 dataset.

Above the heart, venous dilation was predominant. In the head, the venous plexuses exhibited not only marked expansion but also evidence of rupture and blood leakage. Although the dural venous sinuses appeared relatively normal, the internal jugular vein was enlarged after exiting the skull. Although such findings were expected in the 2024 dataset due to cardiomegaly, mild venous dilation was also observed in the 2023 dataset, despite the absence of cardiomegaly (Fig. 2A and B).

Near the heart, both the aorta and the vena cava were dilated in the 2024 dataset (Fig. 2C). Due to cardiomegaly, the left ventricle extended to contact the lateral thoracic wall, whereas the right ventricle reached the anterior thoracic wall. Consequently, the left lung was completely displaced posteriorly (Fig. 2D). A comparison of the 2024 and 2023 datasets clearly demonstrated the extent of cardiomegaly and vascular dilation (Fig. 2C and D).

Near the heart, the thoracic aorta in the 2024 dataset appeared to be displaced left and posteriorly, likely owing to cardiomegaly. However, in the 2023 dataset, the aorta was positioned posterior to the left. In the 2023 data set, the right anterior ribs were severely fractured, suggesting prior cardiopulmonary resuscitation, which may have caused posterior displacement of the heart and aorta (Fig. 2C and D).

Below the heart, the celiac trunk typically originated from the anterior aspect of the abdominal aorta. However, in the 2024 dataset, it originated from the left side of the aorta (Fig. 2E). This unusual position of the celiac trunk was not likely a result of an acute event but rather a long-standing anatomical state. Given that the individual lived without physiological compromise, this finding can be regarded as a normal variation. In the case of the portal vein, it was large in the 2024 dataset, whereas it was small in the 2023 dataset (Fig. 2F).

Feasibility of sectioned images in reproducing anatomical structures comparable to cadaveric dissection: serial tracing of continuous anatomical structures

Structures that extended longitudinally within a horizontal plane and those that continuously followed complex paths across multiple planes were accurately identified.

In this study, various complex paths of normal anatomical structures and pathological findings were identified (Figs. 1-3). In the 2024 dataset, the cadaver had atherosclerosis, and progression stages of the disease were clearly observed in the arteries. Below the heart, in the thoracic aorta, the entire sequence of atherosclerosis was visualized in textbook-like clarity, from the early fatty streak stage to rupture and ultimately to complicated plaque (Fig. 3A). Above the heart, in the brachiocephalic trunk, advanced atherosclerotic changes were found, including multiple sites of complicated rupture with prominent atheroma and intraplaque hemorrhage (Fig. 3B). When the most severe lesion was magnified in Fig. 3B, classical features of a complicated rupture were clearly identified, including the atherosclerotic plaque, the intraplaque fibrous cap, and the disrupted plaque (Fig. 3C). In addition, the magnification of the uterus revealed a likely endometrial neoplasm (Fig. 3D).

Fig. 3. Serial morphology of pathological findings of the 2024 dataset. (A) In the thoracic aorta, the typical progressive stages of atherosclerosis are clearly observed, resembling textbook. (B) Atherosclerosis in the brachiocephalic trunk shows advanced and severe changes. (C) Characteristic features of atherosclerosis are identified in the brachiocephalic trunk. (D) A likely neoplasm is observed in the uterus.

DISCUSSION

In the VK project, the sectioned images of the first cadaver were created in 2003.3 To expand the dataset and improve image quality as well as statistical reliability, additional cadavers were subsequently included. Therefore, in this study, we produced sectioned image datasets from the third and fourth female cadavers. The overall dataset, including both male and female cadavers, corresponded to the ninth and tenth datasets. This expansion provided numerous academic and technical advantages, as outlines below.

As an academic advantage, the sectioned images are equivalent to an undissected cadaver in terms of anatomical completeness. In real cadaver dissection, each anatomical region is examined by sequentially exposing one structure and removing another. Similarly, in the sectioned images, anatomical structures are distinguished one by one, enabling digital dissectors to comprehend and retain their morphology. Just like real cadaver dissection, digital dissection based on sectioned images allows for thorough exploration of every structure. However, unlike real cadaver dissection, a distinct advantage of the sectioned images is their ability to present all anatomical structures in their original, undisturbed state, without cutting, removal, or distortion, as typically occurs in conventional cadaver dissection. Another distinct advantage is their ability to show the true color of each structure. Moreover, these images preserve all structures permanently, allowing repeated examination without any damage, as the images are stored as digital information (Figs. 1-3). Therefore, the sectioned images are ideally suited for anatomical education and clinical anatomy research, confirming the feasibility of the sectioned images in reproducing anatomical structures comparable to cadaveric dissection.

In clinical anatomy research, normal and abnormal morphologies of the human body have traditionally been studied through cadaver dissection by numerous anatomists and physicians, and the identified structures have been published in various clinical anatomy journals such as Clinical Anatomy (ISSN: 0897-3806) and Annals of Anatomy (ISSN: 0940-9602), as well as in various clinical specialty journals. However, a cadaver dissection has fundamental limitations, as mentioned previously. In some cases, the target structure itself is intentionally or unintentionally removed during dissection. In addition, because cadavers inevitably undergo decomposition, they must be cremated after dissection despite careful preservation. Consequently, the dissected structures can be observed only once, and repeated dissections on the same specimen are impossible. Therefore, in clinical anatomy research, the number of cadaver dissections performed is often greater than the number officially reported or published. In contrast, image-based digital dissection is free from these limitations.

When a structure extends longitudinally within a single horizontal plane, it can be easily traced owing to the high resolution of the horizontal sectioned images. Even when a vessel follows a complex path across multiple planes, it can still be traced because the horizontal sections are acquired at narrowly spaced and consistent intervals. This applied not only to normal anatomical structures but also to pathological findings (Fig. 3).

As an academic advantage, the sectioned image dataset represents 10 human bodies, which correspond to 20 sides when considering bilateral structures.4 With this increased number of datasets, the study has achieved sufficient statistical reliability for clinical anatomical research. In the Clinical Anatomy journal, the most frequently cited studies were based on dissections of seven cadavers,16 while one of the most recent studies used 16 cadavers to validate clinical relevance.17 In general, most clinical anatomical studies establish statistical significance using approximately ten cadavers. In this study, datasets were acquired from two female cadavers, and the most recent dataset was produced from the twelfth cadaver. Therefore, the accumulated sectioned images from twelve cadavers are expected to provide sufficient statistical reliability for both anatomical and pathological findings observed in the datasets.

As a technical advantage, the sectioned images are in a pixel-based bitmap format, which allows for both two-dimensional (2D) and 3D reconstructions. By rearranging the pixels, it is possible to create coronal and sagittal planes from the horizontal images. Furthermore, the stacking of the segmented images enables the generation of 3D surface models. Because both the pixel dimensions and the inter-slice intervals are known, the size of each anatomical structure can be precisely calculated. These bitmap images are highly versatile, because they can be processed on a standard personal computer without specialized hardware. Thanks to this versatility, codes generated using ChatGPT-4o can be implemented in MATLAB R2024 (MathWorks®, Boston, MA, USA) to automatically reconstruct 2D images. Similarly, when the code from ChatGPT is entered into Python in 3D Slicer 5.6.2 (https://www.slicer.org/) or Autodesk Maya, 3D reconstruction can be performed quickly and automatically. With the aid of modern AI-assisted coding and various software, anatomists can freely reconstruct bitmap images without relying on professional programmers. Therefore, the technical potential of the sectioned images lies in their universal applicability, which empowers anatomists to independently perform two- and 3D reconstructions with ease.

Furthermore, as DSLR technology continues to advance, the resolution of sectioned images is expected to improve beyond current expectations. We believe that future sectioned images may achieve a level of resolution comparable to that of optical microscopy, as current images with a pixel size of 0.04 mm × 0.04 mm (i.e., 40 µm × 40 µm)19 already approach the resolving power of a light microscope.

Until now, sectioned images have been produced in the United States, Korea, and China.1,4 Based on these datasets, several 2D educational software programs5 and 3D virtual dissection tools have been developed, primarily for student education.20,21 However, these datasets are not suitable for clinical anatomical research, as their resolution was insufficient to visualize fine anatomical structures and the number of cadavers did not provide statistical significance.4 In contrast, the current study demonstrated that sectioned images could be utilized for clinical anatomy research. The 2024 dataset, with its significantly enhanced resolution, allowed for the visualization of extremely fine structures (Figs. 1-3). With this dataset included, the total number of cadaver datasets has reached ten, providing sufficient data for statistical analysis. The advantages of using sectioned images over actual cadaver dissection for identifying anatomical structures were outlined at the beginning of this discussion. The fact that clinical anatomical studies can now be conducted using sectioned images suggests a potential paradigm shift in the methodology of clinical anatomy research.

Owing to the high resolution and consistent intervals of the sectioned images, both macro- and micro-structures were clearly visualized (Figs. 1-3). Because of these high-resolution sectioned images, it is now possible to generate true-color 3D models that provide a lifelike and anatomically accurate representation of the human body. The sectioned images of true color and high resolution have the potential to revolutionize the research paradigm of clinical anatomy, which has traditionally relied on cadaver dissection to gain clinical insights into anatomical structures. In this study, we demonstrated that clinical anatomy research can be conducted based on the academic and technical advantages of these sectioned images. Furthermore, we aimed to demonstrate that digital dissection offers more advantages than traditional cadaveric dissection. Now, it is time to adopt digital dissection as a new standard in anatomical research.