Related Experiment Videos
Computer-aided classification of mammographic masses and normal tissue: linear discriminant analysis in texture
1Department of Radiology, University of Michigan, Ann Arbor, USA.
Physics in Medicine and Biology
|May 1, 1995
Summary
Texture features from spatial grey level dependence (SGLD) matrices effectively classify breast masses on mammograms. This computer-aided diagnosis approach shows promise for improving detection accuracy.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Biomedical Engineering
Background:
- Mammography is crucial for breast cancer screening.
- Accurate classification of breast masses is essential for early diagnosis.
- Texture analysis offers potential for enhancing mammogram interpretation.
Purpose of the Study:
- To evaluate texture features from SGLD matrices for classifying breast masses.
- To determine the impact of SGLD parameters on classification accuracy.
- To assess the feasibility of a computer-aided diagnosis scheme using these features.
Main Methods:
- Extracted 168 mass ROIs and 504 normal tissue ROIs from mammograms.
- Calculated eight texture features using SGLD matrices.
- Employed stepwise linear discriminant analysis and ROC methodology for evaluation.
- Trained and tested a classifier on independent data sets.
Main Results:
- Five SGLD texture features were identified as important for classification.
- Classification accuracy showed weak dependence on distance (>12 pixels) and bit depth (>7 bits).
- Achieved an average area under the ROC curve (Az) of 0.84 (training) and 0.82 (testing).
Conclusions:
- SGLD texture features are effective for classifying breast masses versus normal tissue.
- Linear discriminant analysis in texture feature space is feasible for computer-aided diagnosis.
- This method can aid in distinguishing true and false mass detections on mammograms.