Related Experiment Video
Updated: Aug 18, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.0K
Mass detection in automated 3-D breast ultrasound using a patch Bi-ConvLSTM network
Amin Malekmohammadi1, Sepideh Barekatrezaei1, Ehsan Kozegar2
1School of Computer Engineering, Iran University of Science and Technology (IUST), Tehran 16846, Iran.
Ultrasonics
|December 9, 2022
Summary
Early breast cancer detection using 3-D Automated Breast Ultrasound (ABUS) is crucial. A new AI model, a convolutional BiLSTM network, accurately classifies ABUS slices for faster mass identification, aiding radiologists and improving patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Oncology
Background:
- Breast cancer mortality is reducible through early detection.
- 3-D Automated Breast Ultrasound (ABUS) offers high sensitivity for screening.
- Manual evaluation of ABUS data is hindered by large slice volumes and mass variability.
Purpose of the Study:
- To develop an automated method for classifying 3-D ABUS slices for breast cancer mass detection.
- To assist radiologists by reducing the time and complexity of manual image evaluation.
- To improve the efficiency and accuracy of breast cancer screening using AI.
Main Methods:
- A convolutional BiLSTM network was designed for slice-based classification.
- The model utilizes a patch-based architecture to generate mass location heatmaps.
- A dataset of 60 ABUS volumes from 43 patients was prepared for training and validation.
Main Results:
- The proposed model achieved high performance in slice classification: 84% precision, 84% recall, 93% accuracy, 84% F1-score, and 97% AUC.
- Free-response Receiver Operating Characteristic (FROC) analysis demonstrated 82% sensitivity with 2 false positives per volume.
- The model effectively identifies the approximate location of masses via heatmaps.
Conclusions:
- The convolutional BiLSTM network shows significant promise for automated breast cancer mass detection in 3-D ABUS images.
- This AI approach can enhance radiologist efficiency and potentially improve early detection rates.
- Further validation on larger datasets is warranted to confirm clinical utility.
