Related Experiment Video
Updated: Aug 7, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
A Study on the Effectiveness of Deep Learning-Based Anomaly Detection Methods for Breast Ultrasonography
Changhee Yun1, Bomi Eom1, Sungjun Park2
1National Information Society Agency, Daegu 41068, Republic of Korea.
Sensors (Basel, Switzerland)
|March 11, 2023
Summary
Deep learning models can detect anomalies in breast ultrasound images. The sliced-Wasserstein autoencoder showed superior performance in identifying abnormal regions, though reducing false positives remains a key challenge.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in healthcare
Background:
- Accurate breast cancer diagnosis via ultrasound is challenging due to data limitations and image variability.
- Computer-aided diagnosis (CAD) systems can aid in visualizing abnormalities like tumors and masses.
Purpose of the Study:
- To implement and validate deep learning-based anomaly detection for breast ultrasound images.
- To compare the performance of a sliced-Wasserstein autoencoder against traditional autoencoder and variational autoencoder models for anomaly detection.
Main Methods:
- Implemented unsupervised deep learning models: autoencoder, variational autoencoder, and sliced-Wasserstein autoencoder.
- Trained models on breast ultrasound images to detect anomalous regions.
- Evaluated detection performance using normal region labels.
Main Results:
- The sliced-Wasserstein autoencoder demonstrated superior anomaly detection performance compared to the autoencoder and variational autoencoder.
- Reconstruction-based anomaly detection methods can yield a high number of false positives.
Conclusions:
- The sliced-Wasserstein autoencoder is a promising approach for anomaly detection in breast ultrasound imaging.
- Future research should focus on mitigating false positives generated by reconstruction-based anomaly detection techniques.
