A Benchmark for Breast Ultrasound Image Classification

Classification of breast ultrasound (BUS) images is an essential yet challenging task in computer-aided diagnosis systems. Recently, deep learning-based approaches for BUS image classification have demonstrated state-of-the-art performance; however, it is difficult to reproduce their results and identify the most useful strategies due to the lack of public datasets and method implementations, and inconsistencies in the reported evaluation metrics. Therefore, there is a pressing need to develop a benchmark, to objectively compare current approaches and gain insights on techniques that improve the generalization of BUS image classification. In this work, we build a benchmark for BUS image classification that consists of a large public dataset with 3,641 B-mode BUS images, provide open-source code of state-of-the-art approaches, and identify the best strategies for deep learning-based BUS classification.

Read the benchmark paper (TBA)

To annotate your images, visit our free tool:

Image annotation tool

If you use any of the results of this project, please cite the following papers:

@article{BusB, 
	title = {A benchmark for breast ultrasound image classification},
	author = {Bryar Shareef, Min Xian, Aleksandar Vakanski, Jianrui Ding, Chunping Ning, Heng-Da Cheng},
	journal = { arXiv },
	Year = {2022}
}
	@article{ESTAN, 
	title = {ESTAN: Enhanced small tumor-aware network for breast ultrasound image segmentation},
	author = {Bryar Shareef, Aleksandar Vakanski,  Min Xian},
	journal = { arXiv },
	Year = {2020}
}
	@article{STAN, 
	title = {STAN: Small tumor-aware network for breast ultrasound image segmentation},
	author = {Bryar Shareef, Aleksandar Vakanski,  Min Xian},
	conference = { IEEE 17th International Symposium on Biomedical Imaging (ISBI) },
	Year = {2020}
}

Note:

We are actively working on this project, if you would like to contribute, please contact us. The contributions can be new BUS images, algorithm implementations, your results, or any other knowledge that can enrich the benchmark.

Contact us:

Prof. Min Xian
Email: mxian@uidaho.edu
MIDA Lab
Department of Computer Science
University of Idaho