Research

Lab of Artificial Intelligence in Biomedical Imaging

Cell Imaging

Automatic segmentation and tracking method to measure cellular dielectrophoretic mobility

The dielectrophoresis (DEP) technique is increasingly being recognized as a potentially valuable tool for non-contact manipulation of numerous cells as well as for biological single-cell analysis with a non-invasive characterization of a cell's electrical properties. Therefore we developed a novel automatic segmentation and tracking method for analyzing the cellular trajectories of an ensemble of cells influenced by a DEP force in the same environment as inside a microfluidic device. This algorithm allows us to generate predictive models for single-cell capture and occupancy capacity when manipulating cells in microfluidic devices using DEP forces.

Mouse brain tumor region detection on microscopic image using deep learning method for surgery

In a brain tumor removal surgery procedure, it is very important to accurately detect the tumor area through microscopic images. Currently, detecting a brain tumor area requires a lot of time and labor, so there are many difficulties in the process of surgery. Therefore, in this study, we intend to automatically detect brain tumor regions by applying deep learning technology to microscopic brain images. It is expected that high diagnostic accuracy and significant time reduction will be achieved in the future through the results derived so far.

Development of segmentation algorithm of conjunctival goblet cell using convolutional neural networks

Among the methods of diagnosing dry eye syndrome, it can be determined based on the condition of goblet cells. The purpose is to train a high-performance deep learning model that automatically and accurately segment goblet cell clusters. In order to apply for deep learning, we had to create Ground truth. So, we used a new method of segmentation using GUI. The user can adjust the parameters which is related to intensity and contrast value about neighborhood pixel of the interested region, and crop only the area that expresses well and crop it to the desired size. We used Resnet + U-Net models to capture the contexts of the input image and localize features. The experimental results indicate that it can be quantitatively helpful for goblet cell segmentation and that it can be noted for diagnosis of dry eye syndrome.