DOC PREVIEW
UW-Madison ECE 533 - Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors

This preview shows page 1-2-3-4 out of 13 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 13 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 13 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 13 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 13 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 13 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors Matt McCormick Adam Slater ECE 533 Final Project December 19, 2005Introduction Background For the past twenty years, new medical imaging techniques have been under development that image the solid mechanical properties of tissues using pre-existing imaging modalities. An imaging technique involves exposing the object to a form of energy and creating an image from how the object interacts with the input energy. For instance, with a commonplace photographic camera, the object is exposed to visible light energy, and the picture is related to the light reflecting properties of the object. To make images of solid mechanical properties, we expose the object to the energy of the underlying modality. But we also expose the object to an additional form of energy, mechanical stress. Stress is the force applied per unit area. The object properties are revealed by the deformation the object displays in response to the stress. Currently, the most popular underlying imaging modality used is ultrasound. While a variety of stress field schemes have been applied to create contrast, the images used for this project were created by exposing the tissue to quasi-static compression. This is implemented by compressing the tissue with ultrasound transducer. 'Quasi-static' denotes that the compression was applied very slowly. The stress causes a change from the pre-compression image to the post-compression image. The local displacement of the object is determined by finding the lag of a region that corresponds to the maximum normalized cross correlation of that region. The gradient of the displacement is then taken as an estimation of local deformation; this makes a strain image. An elastic modulus image displays a spatial map of tissue solid mechanical properties. A strain image is a result of the energy used to excite the object's response (the stress field) and the object's properties (the elastic modulus field). If we assume the stress field is uniform during quasi-static compression, the strain image is a rough estimate of the elastic modulus image. The strain image reveals the same mechanical properties as would be elicited during manual palpation. For example, a lump felt in a breast self-examination would show up as an area of contrast in a strain image. The mechanical properties of tissues are highly correlated with disease states, such as breast cancer. Problem Statement Although tissue solid mechanical property images have been tested in research labs for many years, they have yet to see use in a clinical setting. Researchers at UW-Madison have made great strides toward the actualization of these imaging methods in a clinical setting. In the process of advancing strain imaging techniques and preparing them for clinical use, they have collected in vivo breast tumor data. The purpose of this project is to segment a breast tumor from the surrounding tissue in a strain image. The segmentation can be used by our clients to validate the effectiveness of strain imaging in locating and distinguishing benign and malignant breast tumors. Once the tumor region and background region are defined, quantitative measures such as contrast or area can be calculated.Approach For this project, we were given the constraints that: a tumor would be present in each image, there would be only one tumor in each image, tumors have a fairly rounded shape and few concavities, tumors do not contain “holes,” (non-tumor regions surrounded by tumor regions) and each tumor would be fairly close to the center of its image. While our original approach involved the use of level-set methods, we found that, with these constraints, a very reasonable segmentation could be obtained using basic morphological transformations. Therefore, our approach to the problem is a series of image transforms which we determined by trial and error. Figure 1 shows the stages of our approach. Figure 1: Stages of Algorithm Threshold Remove Holes Find the Tumor Gaussian Filter Open Remove Other Regions RestoreStep 1: Gaussian Filter The first stage of our approach is the application of a simple Gaussian filter. This smooths out the high-frequency components of the image, making some of the noisier areas more uniform and easier to segment. Figure 3 illustrates this transformation. Figure 2: Original image used in example segmentation Figure 3: Gaussian filtered imageStep 2: Threshold The second stage is to apply a threshold, determined via trial and error, to the image in order to convert it into a binary image. The results of this operation can be seen in Figure 4. Step 3: Remove Holes In stage 3, a border of zeroes is added to the image. Starting at a point on this border, all connected components in the image with a value of zero are extracted. If all other pixels are set to one, this results in the removal of all holes from the image. An inherent assumption in this operation is that the image does not have a border of ones after step 2, but this seems to be a reasonable constraint, as none of our test images exhibited this property. This operation is demonstrated in Figure 5. Figure 4: Thresholded imageStep 4: Open After step 3, an opening operation is performed on the tumor region using a structural element whose size was determined by trial and error. This operation serves to remove any small peninsulas coming out of the tumor and to strip off any non-tumor areas which were mistakenly kept past step 5. This is shown in Figure 8. Figure 5: Image with holes removed Figure 5: Opened imageStep 5: Find the Tumor Our program allows the user to manually input the coordinates of a point within the tumor; however, it also includes an optional automatic tumor-finding algorithm based on the constraints of our problem. Since the tumor regions to be extracted will be close to the center and fairly rounded, we developed an algorithm to attempt to find a pixel which best exhibits these characteristics. First, our algorithm determines the distance from each pixel to the nearest zero pixel. In the case of pixels whose value is zero, this distance is zero. Next, our algorithm calculates the distance from each pixel to the center of the image. It then finds the pixel which maximizes the distance from a zero pixel while minimizing the distance from the image center. This pixel is given as its


View Full Document

UW-Madison ECE 533 - Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors

Documents in this Course
Load more
Download Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Automatic Medical Ultrasound Strain Image Segmentation for Breast Tumors 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?