Related Experiment Videos
Texture analysis of breast tumors on sonograms
D R Chen1, R F Chang, Y L Huang
1Department of General Surgery, China Medical College & Hospital, Taichung, Taiwan. dlchen88@ms13.hinet.net
Seminars in Ultrasound, CT, and MR
|October 3, 2000
Summary
Artificial neural networks can distinguish benign and malignant breast tumors using ultrasound texture features. This AI approach shows high accuracy, outperforming radiologists in classifying breast pathology from sonograms.
Area of Science:
- Medical imaging
- Artificial intelligence
- Oncology
Background:
- Accurate breast pathology classification is crucial for patient outcomes.
- Sonography is a common imaging modality for breast lesion assessment.
- Distinguishing benign from malignant breast tumors remains a clinical challenge.
Purpose of the Study:
- To assess the feasibility of using artificial neural networks (ANNs) with sonogram texture features for breast pathology prediction.
- To compare the diagnostic performance of the ANN model against experienced radiologists.
Main Methods:
- A dataset of 1,020 sonogram images from 255 patients was analyzed.
- An ANN model utilized 24 autocorrelation texture features to classify tumors as benign or malignant.
- Three radiologists independently classified the same images without prior knowledge.
Main Results:
- The ANN achieved a receiver operating characteristic (ROC) area index of 0.9840 ± 0.0072.
- The ANN correctly identified 35/36 malignancies and 211/219 benign tumors.
- Radiologists identified an average of 19/36 malignancies, with several indeterminate classifications.
Conclusions:
- Texture features from digital ultrasonic images, analyzed by ANNs, can effectively distinguish between benign and malignant breast tumors.
- The proposed ANN system demonstrates superior diagnostic performance compared to human radiologists in this study.
- Interpixel correlation analysis in sonograms holds significant potential for improving breast cancer diagnosis.