Kuan Huang
Kean University, NJ·Computer Science
Finding breast cancer early saves lives, and ultrasound is one of the most widely used screening tools because the machines are cheap, portable and available almost everywhere. The trade-off is image quality: ultrasound pictures are grainy and low in contrast, which makes tumors hard to distinguish and makes automated computer analysis unreliable.
- Fields
Computer Science
Biomedical Engineering
Medicine & Clinical Research
- Topics
breast cancer
breast ultrasound
medical image analysis
- Grants
- 1 known grant, through Aug 2027
What they work on
- Develop weakly supervised breast cancer detection that locates lesions in ultrasound images without radiologist-drawn bounding boxes, using a transformer-based classification network that weighs relationships across the whole image.
- Build an active learning loop that iteratively selects the most informative breast ultrasound images for annotation based on detection performance, so the model improves while manual labeling effort stays low.
- Test whether the combined weakly supervised and active learning detector remains accurate under the substantial noise and low contrast inherent to breast ultrasound imaging.
More about the research
There is a second obstacle too. The deep learning methods that perform best at medical image analysis normally need large collections of images in which a radiologist has drawn a box around each lesion, and that labeling is slow and expensive. This project targets both problems at once, building breast cancer detection that works with very few manual annotations and tolerates the noise that comes with ultrasound.
Why it matters
Breast cancer mortality drops when tumors are caught early, and ultrasound is often the only imaging available in lower-resource settings. Detection software that learns from minimal expert labeling and copes with noisy ultrasound could make computer-assisted reading practical where large annotated datasets and specialist radiologists are scarce, supporting earlier diagnosis for women in underserved groups.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Ronald Stanley
Missouri University of Science and Technology, MO·Engineering
Build an online platform with the International Skin Imaging Collaborative (ISIC) where experts worldwide annotate, segment and validate critical structures in dermoscopic skin lesion images.
Kang Wang
University of California-San Francisco, CA·Medicine & Clinical Research
Build a large standardized database of abdominal CT images and radiology reports, using large language models to automatically extract findings from reports and select data for AI training.
Christoph Lee
University of Wisconsin-Madison, WI·Medicine & Clinical Research
Validate a 2D mammography AI algorithm from the DREAM Digital Mammography Challenge on UCLA's Athena Breast Health Network registry, and refine it with radiologist supervision plus genetic mutation and tumor subtype data.
The school
Kean University: 1 labs on PI Pages, $9M a year on research, 30 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.