Bryar Shareef

Bryar Shareef

Assistant Professor

CV

Education

Work experience

Research experience

During my Ph.D. studies, my research mainly focused on developing new AI/ML approaches for accurate object segmentation and classification and applying them to the early detection of breast cancer using ultrasound images. Specifically, I ….

  1. Developed, implemented, and validated small-object-aware deep networks The newly proposed multi-encoder neural network architectures have filters with different shapes and extract and fuse image features at different scales. The new design does not need extensive pooling operations and can preserve fine details of images, which leads to the accurate detection of small objects.
  2. Developed a deep multi-task network for simultaneous object classification and segmentation The proposed multi-task neural network combines two separate tasks in one shared model. The network learns from both tasks and alleviates the low generalization issue caused by small training datasets. The learned shared features between object segmentation and classification improve the robustness and generalizability of the model.
  3. Benchmarked breast cancer detection using ultrasound images i. Built a large benchmark dataset with more than 3,000 breast ultrasound images. Designed a webpage to share the results, and source code, and prepared detailed documentation. ii. Proposed and implemented two novel deep neural networks for the segmentation of breast ultrasound images. The approaches achieved the lowest false positive rate for both small tumors and tumors of different sizes. iii. Designed a multitask learning approach for breast tumor classification and segmentation. Successfully increased the detection accuracy by 8-12% compared to the current best models.

Awards / Competitions