Classification of mass abnormalities based on denseness and irregularity texture features /
A mammogram image is abundant in textural information. In this research, it is hypothesized that texture analysis, based on radiographic denseness and irregularity of breast cancer, improves the classification of mass abnormalities into malignant or benign diagnostic categories. To test this hypothe...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2000.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=728415691&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | A mammogram image is abundant in textural information. In this research, it is hypothesized that texture analysis, based on radiographic denseness and irregularity of breast cancer, improves the classification of mass abnormalities into malignant or benign diagnostic categories. To test this hypothesis, two textural features, named smoothness/denseness feature and architectural distortion feature, are newly developed and examined. The smoothness/denseness textural feature measures the textural characteristics formed by the radiographic denseness of breast cancer. The architectural distortion textural feature measures the texture characteristics formed by the irregular variations in intensity levels of breast cancer on mammograms. Experiments with 404 mass images indicate that the improvement on the classification performance is statistically significant (p[]0.0001) by using the two textural features when compared with other textural features. Receiver operating characteristic (ROC) analysis is conducted to evaluate the classification performance. The area under ROC curve reached 0.90. A hypothesis test is performed to see if the classification performance is sensitive to the different sizes of marking. The test result indicates that the classification performance is not statistically sensitive to varying sizes of marking at the significance level of 0.19 (p=0.19) when the marking size changes within the standard deviation of 20 in normal distribution. These newly developed textural features are integrated into a computer aided diagnosis system for radiologists to be used as an electronic second opinion system. |
|---|---|
| Item Description: | Vita. "Major Subject: Electrical Engineering". |
| Physical Description: | xiii, 163 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 159-162). |