Computers in Medical Imaging
Computers are being used extensively in the field of medicine. One field of medicine that is dominated by computers is medical imaging. Indeed, computers have changed the field of medical imaging as they have made things possible that were not possible before.
Medical imaging can generally be split into image acquisition, image processing and image analysis. Image acquisition is the act of taking a medical image and then storing that image. Medical images are classified by the different image modalities. A partial taxonomy of the image modalities would include: x-rays, computed tomography (CT) scans, magnetic resonance images (MRIs), single positron emission computed tomography (SPECT) scans, position emission tomography (PET), and ultrasound. Image processing is applying a transformation to the original image to produce a new image. For example, this transformation can reduce the noise in the image and enhance the edges. Image analysis is the act of extracting meaningful information from an image. This type of information can aid in providing a diagnosis or can be used for other purposes. For example, determining the radii or the volume of the interior region of vertebrae in a CT scan can be used to determine if the patient has spinal stenosis. Computers are used at each of these three stages.
Computers in Image Acquisition
Computers guide the machines that take medical images. Even x-ray and ultrasound image acquisition involves computers. For example, a CT scan is essentially taking a series of x-ray images at different angles for a slice of the body. A computer is used to move the components, fire the x-rays, and take measurements at the sensors. Finally, a computer is used to run the back projection algorithm to construct the final image. Many image acquisition machines come equipped with computers that automatically process the acquired image to remove artifacts, such as noise. After the image is acquired, it is generally stored on a computer system. X-rays and ultrasound images do not need to be stored digitally.
Computers in Image Processing
Computers are used the most in medical image processing. You can think of image processing on an image I as applying a transformation T to I such that the new image I' = T(I). This transformation can do many things such as reduce the noise in the image, compensate for non-uniformity intensities, apply histogram equalization or enhance the edges in the image. There are equations and algorithms for each of these operations but I will leave those for another day.
Transformations do not have to augment the image. There is a special type of transformation that produces another image called a segmentation.
Medical Image Segmentation
Fig. 1 A series of geometric shapes and the corresponding segmentation
A segmentation is an image that clearly identifies regions of interest in an image. Allow me to illustrate this using an example. Suppose we have the corresponding image (Fig. 1) that is composed of a series of geometric shapes and we wanted to know where the circle is in the image. An equivalent question would be "what pixels in the original image belong to class circle"? The segmentation is essentially an image that answers this question as it defines what pixels in the image belong to the circle. The segmentation would be a black and white image, where a black pixel (intensity value 0) in the segmentation would mean that the corresponding pixel in the original image is not in the circle. A white pixel (intensity value of 1) in the segmentation would mean that the corresponding pixel in the original image is part of the circle. Mathematically:
S = T(I) If (S(x,y) == 0) I(x,y) is not of class circle Else I(x,y) is of class circle
A segmentation that identifies multiple regions in an image is called a multi labeled segmentation and the image intensity values at a pixel are no longer binary.
The Importance of Segmentation
Before a radiologist can perform any measurements or calculations on a medical image, they must first create the segmentation for the desired region. This process is called segmenting the image. The segmentation that the radiologist produces is called the ground-truth segmentation as it is the most accurate segmentation.
Image segmentation is a very tedious and timely process. However, it is an extremely important piece in the diagnosis process. Medical image segmentation research dominates the research done in medical imaging. This is because of the fact that a radiologistŐs time is very, very expensive. Thus, anything that can save a radiologist time is a very desirable. Furthermore, the segmentation problem is very challenging and has yet to be solved (generally).
Medical Image Analysis
Medical image analysis can be thought of as applying a transformation to an image to extract information. Some of the information can be useless, but sometimes it is very important. Often, the information that is extracted is actually a diagnosis. For example, suppose you are given a series of back images of a patient through time along with the corresponding ground-truth segmentations of the moles on the patients back. The medical image analysis step would be tracking how the moles grow with time and then based on the growth of the moles, determining of the moles are benign or malignant. Medical Image Analysis does not have to be involved in diagnosis. However, it is possibly the most rewarding part of medical imaging research, and - in my opinion - the most interesting as you are exposed to a lot of medical knowledge about the condition that you are trying to diagnose. It makes you feel like a less bad-ass version of House.
To Be Continued ...
I think that is enough about medical imaging for one update. My next medical image related post will discuss the different techniques that almost all segmentation algorithms are derived from and how they differ from one another.