Medical image computing

Medical image computing is a special kind of science that helps doctors and researchers understand pictures made by machines that look inside our bodies. It uses ideas from computer science, math, and medicine to find important details in these pictures. This can help doctors see problems inside someone more clearly or learn new things about how our bodies work.

The main job of medical image computing is to find useful information from these body pictures. It is different from taking the pictures because it looks at the pictures with computers to learn more. There are many ways to do this, like separating different parts of the body in the picture or matching pictures taken at different times. These tools help doctors make better decisions to keep people healthy.

Data forms

Medical image computing works with data that is collected in a regular grid, like pictures in 2D or 3D. Each point in this grid can have different types of numbers, such as small whole numbers or more detailed floating-point numbers. The meaning of these numbers changes depending on how the image was taken. For example, a CT acquisition records how dense materials are, while an MRI acquisition might record details based on T1 or T2 settings.

Some images are taken over time, and others might look different because of the way they were captured, like fan-shaped images from curved-array ultrasound. There are also special shapes used in detailed studies of how tissues and bones move and change.

Segmentation

Segmentation is a way to divide an image into different important parts. In medical images, these parts can show different tissues, organs, or problems in the body. Doing this can be tricky because images sometimes don’t show clear differences or have noise.

There are several ways to do segmentation. One way uses special maps made by doctors called atlas-based segmentation. Another way uses shapes that change to fit the image, called shape-based segmentation. Some methods start with a basic shape and change it using the image data, like the active contour model. Interactive segmentation lets doctors give some hints, and a computer program finishes the job. There are also advanced computer programs, like convolutional neural networks, that can do segmentation very fast and well.

Registration

Image registration is a way to line up pictures so they match correctly. Imagine you have two pictures of the same thing taken at different times or with different cameras. You want them to look the same so you can see changes or combine information from both.

Doctors use image registration for many reasons. They can watch how something changes over time, like how a disease grows. They can also mix information from different types of pictures to get a fuller view. This helps during surgery, where doctors need to match pictures taken before surgery with what they see during the operation.

There are different ways to line up the pictures, and each has its own rules. Some methods are simple and keep the shape the same, while others allow more bending and stretching. Doctors pick the best way based on what they need to see and how clear the pictures are.

Visualization

Visualization helps us see and understand medical images better. It uses special ways to show pictures from inside the body, helping doctors and scientists study them. For example, it can show slices of images in different directions or create 3D pictures from the data.

The picture called "Visualization of Medical Imaging" shows a few types of views: slices of images, different angled views, and a 3D picture. These views help make important details, like a spot in the body, easier to see and study.

Atlases

Medical images can look very different between people because everyone’s body parts are shaped and sized uniquely. To handle this, scientists use special models called atlases. An atlas is like a guide that helps match new images to a common reference.

One simple type of atlas is called a template, which is an average image. But sometimes, more detailed atlases are used that include extra information about what each part of the image might be. To use an atlas, a new image is adjusted, or “mapped,” to match the atlas. This helps in tasks like separating different parts of the body in an image.

The easiest way to use an atlas is to have just one template that everyone’s images are compared to. For example, brain scans are often matched to a standard brain template. However, this can be tricky if the image has big differences, like in the case of a brain with unusual growth.

To fix this problem, some scientists use many templates instead of just one. They might have one template for healthy brains and another for brains with certain conditions. Deciding how many templates to use can be hard, so some methods automatically figure this out by comparing new images to many examples in a training set.

Statistical analysis

Statistical methods connect medical imaging with modern computer vision, machine learning, and pattern recognition. In recent years, many large sets of medical images have become public. This growth in data means we need new ways to find small changes in the images to answer important health questions. These questions can include comparing groups of people, finding signs of disease, and studying how diseases change over time.

Group analysis

Group analysis looks for differences in medical images between groups of people. One group might be healthy, while the other has a disease. We can see changes in the body, like shrinkage in certain parts of the brain, which might be linked to diseases such as Alzheimer's disease. To compare these groups, images are adjusted to match up, and then we look at each tiny part of the images to find differences.

Classification

Classification helps doctors diagnose diseases early and understand how they progress. This is different from group analysis because it looks at each person individually. There are challenges, such as having enough images to train the methods and making sure the results are easy to understand for doctors.

Clustering

Sometimes diseases affect people in very different ways, and it’s not easy to put them into simple categories. Clustering is a way to sort people into smaller groups based on their images, which can help us understand the disease better.

Shape analysis

Shape analysis studies the shapes of body parts seen in medical images. It helps find small changes between healthy and unhealthy shapes. This involves matching up shapes and then measuring changes at matching points.

Longitudinal studies

Longitudinal studies follow the same person over time using repeated images. This helps reduce noise in the data and improves the ability to spot real changes. However, care must be taken to treat all images the same to avoid mistakes in the results. Special statistical tools are used to analyze this kind of data.

Image-based physiological modelling

Medical image computing helps us understand how the body works by using pictures taken from machines. These pictures show both the shape and how parts of the body are working. Recently, scientists have started using these pictures to predict how diseases might change or how treatments might work.

The Virtual Physiological Human is a big project that wants to create computer models of the entire human body. These models will help scientists study how everything in the body works together, from tiny genes to the whole body. Medical imaging is very important for this because it gives detailed pictures and information about people. This helps create models that match each person exactly.

These models can be used in many ways, like helping doctors understand diseases better or planning treatments before doing them. By combining pictures and computer models, doctors can learn more about how the body works and how to treat illnesses.

Mathematical methods in medical imaging

Many smart math methods are used in medical imaging to help doctors see and understand pictures better. These methods include special equations and flows that can make pictures clearer, separate parts of the body, and match pictures together. Because these methods use equations, they can be made faster on special computers. Recently, ideas from controlling machines have also been used to help pick out parts of the body in pictures. To handle noise and changes in pictures, tools like the Kalman filter and particle filter are also used.

Modality-specific computing

Some imaging methods give very special information. These images are different from regular ones, so they create new areas of study in medical image computing. Two important examples are diffusion MRI and functional MRI.

Diffusion MRI is a special kind of magnetic resonance imaging that measures how molecules move. This helps doctors learn about the properties of tissues and the paths that molecules take in the body. Functional MRI measures brain activity by looking at blood flow. It can be used when a person is doing a task or simply resting, helping scientists understand how the brain works.

Software

Software for medical image computing is a mix of different systems that help doctors and scientists work with medical images. These systems handle showing images, letting users interact with them, and storing data. They are built in layers. The bottom layers are like toolkits that give basic computing powers. The top layers are special programs designed to solve specific medical problems or study different parts of the body.

Additional notes

Medical image computing is connected to the area of computer vision. There is an international group called The MICCAI Society that represents this field. They hold an annual conference and workshops. The papers from this conference are published by Springer in the Lecture Notes in Computer Science series. In 2000, N. Ayache and J. Duncan looked at how the field was doing at that time.