Heming Yao

Heming Yao


Areas of Interest

We are developing an automated system to detect and segment hematoma regions in head Computed Tomography (CT) images of patients with acute traumatic brain injury (TBI). Our proposed method has three main states as follows. 1) Pre-processing, 2) Feature extraction and 3) Training a learning model. Statistical, texture and geometrical features are extracted to capture characteristics of hematoma regions. The uncertainty-based active learning strategy is used to select the most representative samples to train an SVM classifier. We are developing a CNN algorithm for left ventricle segmentation, where filters are initialized as Gabor filters. In the procedure of training, we keep the structure of Gabor filter by updating 5 parameters of Gabor filters rather than weights. We are also studying the tensor decomposition to speed up the training and test procedures.


  • B.S., University of Science and Technology of China (Biology)