Related Experiment Videos
False-positive reduction in CAD mass detection using a competitive classification strategy.
1Department of Radiology, College of Medicine, and the H. Lee Moffitt Cancer Center and Research Institute at the University of South Florida, Tampa 33612, USA. lihua@splinter.rad.usf.edu
Medical Physics
|March 13, 2001
Summary
A new computer-aided detection (CAD) method combines "hard" and "soft" classification to significantly reduce false positives in mammograms. This approach improves accuracy for breast cancer detection, especially in dense tissue, with minimal impact on true positives.
Area of Science:
- Medical Imaging
- Computer-Aided Detection (CAD)
- Machine Learning
Background:
- High false-positive (FP) rates in CAD systems are a major challenge, degrading detection performance and increasing patient recall rates.
- Existing CAD systems often struggle with accurately identifying true positive regions while minimizing false alarms.
Purpose of the Study:
- To propose and evaluate a novel classification method for reducing false positives in CAD systems.
- To enhance the accuracy of breast cancer detection by minimizing false alarms without significantly compromising true positive identification.
Main Methods:
- A hybrid classification approach cascading a conventional "hard" decision classifier with a novel "soft" decision classifier.
- The "soft" classifier employs a competitive strategy, selecting only the most likely true masses from initially classified suspicious regions.
- A neural network structure was designed to implement this competitive "soft" classification strategy.
Main Results:
- The proposed "hard"-"soft" classification method demonstrated significant FP reduction compared to "hard" classification alone.
- In a high FP sub-database, FPs were reduced by 56% (from 8.36 to 3.72 per image) with only a 1% loss in true positives (TP).
- Increasing the threshold of the "hard" classifier alone yielded only 27% FP reduction at the cost of a 14% TP loss.
Conclusions:
- The cascaded "hard"-"soft" classification strategy is highly effective in reducing false positives in CAD systems.
- This method offers a substantial improvement over conventional approaches, particularly for challenging cases with dense breast tissue.
- The proposed technique holds significant promise for enhancing the early detection of breast cancer using CAD.