Detection of Masses and Microcalcifications in Digital Mammogram Images
Breast cancer is the most common health problem among women today. Early detection of breast cancer can be helpful. However, detection of breast lesions (mass and microcalcification) is a challenging task for radiologists. Computer Aided Detection (CAD) systems are designed and implemented to aid ra...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/21838/1/FPSK%28p%29_2011_6IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Breast cancer is the most common health problem among women today. Early detection of breast cancer can be helpful. However, detection of breast lesions (mass and microcalcification) is a challenging task for radiologists. Computer Aided Detection (CAD) systems are designed and implemented to aid radiologists and detect masses and microcalcifications raising the level of sensitivity of breast cancer detection than human eye detection. In this research, a new system is suggested for detecting lesions. In the preparation phase, 18000 small samples (each of size 8 by 8 pixels) were extracted from different tissue types. Textural features were calculated for each sample. The best features were selected using Weikato Environment for Knowledge Analysis (WEKA) software. Subsequently, 7 selected features were used to extract a decision tree. To reduce false negative cases and remove sharp boundaries, the fuzzy logic theory was used. Input and output membership functions were defined based on the decision tree. In the implementation phase, input images were divided into 8 by 8 pixel tiles. For each tile, all selected features were computed as fuzzy inputs. Based on the fuzzy results, a binary image was produced as an output image. In the output image all lesion areas were shown in white while other areas were shown in black. Sobel filter were employed to detect boundaries. Finally, the boundaries were adjusted on the original image. This is to select and show suspicious locations on original image for radiologists. In the evaluation phase, two experiments were presented; first the suggested system was applied to 322 images obtained from the MIAS data set. Based on the data obtained from the first experiment, results showed that the suggested system has an acceptable sensitivity of 82.56% and a specificity of 88.26%. In the second experiment, the suggested system was applied on 326 local images that were obtained from the national cancer society of Malaysia. A sensitivity of 85.61% and specificity of 90.72% was obtained from that study. |
|---|