Developing a new case based computer-aided detection scheme and an adaptive cueing method to improve performance in...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computer-aided detection (CAD) schemes are crucial for improving screening mammogram interpretation.
• Enhancing the performance of CAD systems is essential for early cancer detection and reducing false positives.

Purpose of the Study:

• To evaluate a novel method for improving the performance of computer-aided detection (CAD) schemes in screening mammograms.
• To develop and integrate new feature extraction and scoring techniques into existing CAD frameworks.

Main Methods:

• A new case-based CAD scheme was developed using global mammographic density, texture, spiculation, and structural similarity features.
• Features were selected using a modified sequential floating forward selection algorithm and classified using a scoring fusion artificial neural network.
• A novel adaptive cueing method combined case-based risk scores with conventional lesion-based CAD scores.
• Methods were evaluated using ten-fold cross-validation on 924 screening mammograms (476 cancer, 448 recalled/negative).

Main Results:

• The area under the receiver operating characteristic curve (AUC) achieved was 0.793 ± 0.015.
• The odds ratio increased monotonically from 1 to 37.21 with increasing CAD-generated case-based detection scores.
• The adaptive cueing method improved region-based sensitivity by 2.4% and case-based sensitivity by 0.8% at a false positive rate of 0.71 per image.

Conclusions:

• Supplementary information derived from global mammographic density features can significantly improve CAD performance.
• The developed case-based CAD scheme and adaptive cueing method enhance the detection of suspicious mammographic lesions.
• This approach holds promise for improving the accuracy and efficiency of mammography-based cancer screening.