DICOM loading and visualization
Digital Imaging and Communications in Medicine (DICOM) is the most adopted standard for medical imaging. Back in the day, the manufacturers of computed tomography and magnetic resonance devices provided their own solutions for the storage and transmission of medical images. DICOM came as the first effort in the industry to create a standardized protocol that everyone could adopt.
Since the first DICOM standard was released, the specification has been growing to meet the requirements of manufacturers and medical professionals. Today, in 2025, the DICOM spec is very extensive, with over 6000 pages in 22 chapters.
Capture devices in Medical Imaging
Although there are many different types of capture devices, DICOM is the common protocol for all of them.
X-ray Machines
Uses ionizing radiation to capture images of bones and certain tissues. It is widely used for diagnosing fractures, infections, and other conditions.
Computed Tomography (CT)
Use X-rays and computer processing to create cross-sectional images (slices) of the body. CT scans provide detailed images of bones, blood vessels, and soft tissues, making them valuable for diagnosing various conditions, including cancers, cardiovascular diseases, and trauma.
Magnetic Resonance Imaging (MRI)
Use strong magnetic fields and radio waves to generate detailed images of organs and tissues. MRI is particularly useful for imaging the brain, spinal cord, joints, and soft tissues without using ionizing radiation.
Ultrasound Machines
Uses high-frequency sound waves to produce real-time images of internal organs and structures. It is commonly used in obstetrics, cardiology, and abdominal imaging. Ultrasound is non-invasive and does not use ionizing radiation.
Positron Emission Tomography (PET)
Use radioactive tracers to visualize metabolic processes in the body. PET scans are often combined with CT scans (PET-CT) to provide both functional and anatomical information, making them valuable for cancer diagnosis and monitoring.
Single Photon Emission Computed Tomography (SPECT)
Similar to PET, SPECT uses gamma rays and radioactive tracers to create 3D images of functional processes in the body. SPECT is commonly used in cardiology and neurology.
Fluoroscopy Machines
Uses X-rays to provide images in real-time. The main advantage over other capture methods is that we can get moving images and immediate feedback, so it’s especially suited for monitoring the patient during surgeries.
DICOM in the medical and dental industry: Not just an image format
The DICOM spec is not only about the images, there is much more to it. It can represent extensive metadata and communication protocols.
These are some examples of information that can be stored in DICOM, but there are many more:
- Multi-frame images: can store multiple images captured with devices that provide slices of the body, such as CT and MRI devices. It also contains information about the positioning of those images, spacing between slices, and much more.
- Patient information: such as name, ID, birth date, sex, medical history, diseases, etc.
- Study information: such as the date of the study, the hospital or university, doctors, description, etc.
- Acquisition parameters: capture device, model along with its configuration parameters, manufacturer, software versions, protocols, and other technical details.
- Annotations: we can store annotations made by the doctors on the images (arrows, lines, text, handwritings, etc).
Volumetric Rendering
When capture devices generate multiple 2D slices of a body, we can arrange them to build a 3D volume. In this section, we will explain the basics of the volumetric rendering algorithm that we use in Evergine.
Step 1: Generate 3D Texture
Our rendering algorithm needs a 3D texture as input. Thus, the first step will be generating this 3D texture.
The images found in DICOM files can have different arrangements. They can be loose slices positioned in 3D space. They can have different orientations and resolutions. So we must gather all these images, filter them, process them, and sort them in coherent order. This is a very complicated step and, since the DICOM specification is very complex and flexible, it’s not easy to handle every possible case. Still, we have found that our algorithm seems to work with most cases, and being able to generate a single 3D texture is a big advantage for efficient 3D rendering.
Step 2: Iterative Volumetric Ray Tracing
Once we have our 3D texture ready, we will assign this texture to a geometry for drawing, for example, a box, or a cylinder. The geometry shape must be convex and it should, ideally, approximate the shape of the texture’s pixels. Choosing a smaller, better-fitting shape can help improve performance, as it reduces the number of pixels that originate rays.
For example, CT scanners have cylindrical shape, and, therefore, the output DICOM pixels are always distributed inside a cylinder. So it makes sense in those cases to use a cylinder as the geometry of choice.
The algorithm casts rays against the geometry. Since the geometry must be convex, the ray will return two intersection points, where it enters the volume, and where it exits.
Starting from the entering point, we will keep moving along the ray towards the exiting point, sampling the pixels in our 3D texture.
Each ray keeps track of the accumulated density, which will increase depending on the sampled pixels’ density (and the configured density range window). Once the accumulated density reaches a specified threshold, we can consider that the density is high enough that light can’t pass through, so we stop iterating.
All this process is actually leveraged through rasterization. The exiting point is actually computed by writing the back faces of our convex geometry to the depth buffer. Writing to the depth buffer also allows us to consider other objects in the scene. This makes sense as if we find an object before the ray exits the DICOM geometry, we should stop iterating anyway.
After writing the back faces of the DICOM to the depth buffer, we draw the front faces, but this time, with the shader that will do all the heavy workload, and end up writing to the color buffer. This shader will sample the depth buffer to compute the end of the ray (but will not write to the depth buffer).
For the corresponding ray of each pixel, we will do many step iterations. The step size is, in our implementation, equal to the pixel size, so the number of iterations will be in the order of hundreds.
The step size could be increased for low-end devices (with a reduction in quality). However, our tests show that performance is good (above 60 fps) even with integrated GPUs.
Evergine Add-On
We have developed a new DICOM add-on for Evergine. This add-on has been developed to fulfill the requirements of our clients in the medical and dental industry.
It’s completely free to use, and you can check all the features and details in the last Evergine release announcement.
Also, you can read the related documentation if you want to learn how to use it. It’s very easy to make your own DICOM viewer in Evergine!
