Computerized lesion detection on breast ultrasound

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Breast ultrasound is crucial for breast cancer detection.
• Automated lesion detection can enhance diagnostic efficiency and accuracy.
• Current methods may require further optimization for improved performance.

Purpose of the Study:

• To evaluate a novel radial gradient index (RGI) filtering technique for automated breast lesion detection.
• To assess the performance of RGI filtering in segmenting and classifying lesions on breast sonograms.
• To determine the potential of this computerized analysis in breast cancer screening programs.

Main Methods:

• Implementation of a radial gradient index (RGI) filtering technique for initial lesion detection.
• Segmentation of lesion candidates using an average radial gradient (ARD) index.
• Classification of lesion candidates using a Bayesian neural network with round robin analysis.
• Performance evaluation using sensitivity, false-positive rates, and Az values.

Main Results:

• Initial RGI filtering achieved 87% sensitivity with 0.76 false-positives per image.
• Lesion segmentation with ARD index resulted in 75% correct detection at 0.4 overlap.
• Bayesian neural network classification yielded an Az value of 0.84.
• Overall performance demonstrated 94% sensitivity at 0.48 false-positives per image.

Conclusions:

• The RGI filtering technique offers a promising approach for automated breast lesion detection in ultrasound.
• Computerized analysis of breast sonograms can significantly improve diagnostic accuracy.
• This technology may facilitate the broader use of sonography in breast cancer screening.