Related Experiment Video
Updated: Aug 14, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Evolving artificial neural networks for screening features from mammograms
D B Fogel1, E C Wasson, E M Boughton
1Natural Selection, La Jolla, CA 92037, USA. dfogel@natural-selection.com
Artificial Intelligence in Medicine
|November 20, 1998
Summary
Artificial neural networks (ANNs) show promise in diagnosing breast cancer from mammograms. These computer algorithms can analyze radiographic features, aiding physicians and potentially improving diagnostic accuracy with simpler models.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Mammographic interpretation inconsistencies necessitate advanced diagnostic tools.
- Computerized pattern recognition can aid in assessing radiographic features for breast cancer detection.
Purpose of the Study:
- To evaluate the potential of artificial neural networks (ANNs) for analyzing mammographic features.
- To develop and train ANNs using evolutionary programming for malignancy prediction.
Main Methods:
- Utilized film screen mammograms from 216 suspicious cases, with features quantified by a domain expert.
- Trained various complexity ANNs on 12 radiographic features and patient age.
- Confirmed malignancy status via surgical biopsy (111 malignant, 105 benign).
Main Results:
- The best ANNs achieved a mean area under the ROC curve (AZ) of 0.9196 for suspicious masses at 0.95 sensitivity.
- Achieved a mean specificity of 0.6269 for suspicious masses.
- ANNs with two hidden nodes performed comparably to more complex models, indicating efficiency.
Conclusions:
- Evolved ANNs demonstrate comparable performance to prior studies with significantly reduced complexity.
- The success of small ANNs suggests potential for explainable AI in breast cancer diagnosis.
- This approach may increase physician acceptance and trust in AI-assisted mammography.
