Related Experiment Videos
Computerized classification of malignant and benign microcalcifications on mammograms: texture analysis using an
H P Chan1, B Sahiner, N Petrick
1Department of Radiology, University of Michigan, Ann Arbor 48109-0326, USA. chanhp@umich.edu
Physics in Medicine and Biology
|March 1, 1997
Summary
Texture analysis of mammograms can differentiate malignant from benign microcalcifications. This artificial neural network (ANN) approach achieved 100% sensitivity, potentially reducing unnecessary biopsies.
Area of Science:
- Radiology
- Medical Imaging
- Computer-Aided Diagnosis
Background:
- Mammography is crucial for breast cancer screening.
- Microcalcifications are key indicators, but distinguishing benign from malignant can be challenging.
- Computerized analysis may improve diagnostic accuracy.
Purpose of the Study:
- To assess the feasibility of using texture features from mammograms to predict the pathology of microcalcifications.
- To develop and evaluate an artificial neural network (ANN) classifier for this task.
Main Methods:
- Texture features were extracted from regions of interest containing microcalcifications using spatial grey level dependence matrices.
- A stepwise feature selection technique identified optimal texture features.
- A backpropagation ANN was trained and tested using a leave-one-case-out method.
- Performance was evaluated using receiver operating characteristic (ROC) analysis.
Main Results:
- A subset of six texture features yielded the highest classification accuracy.
- The ANN achieved an area under the ROC curve of 0.88.
- The system demonstrated 100% sensitivity for malignant cases and 39% specificity for benign cases, correctly identifying 11 of 28 benign cases.
Conclusions:
- Computerized texture analysis can extract valuable information from mammograms not apparent by visual inspection.
- This approach shows potential for assisting mammographic interpretation, reducing benign biopsies, and improving the positive predictive value of mammography.