Analytics

Persistent homology is a study for the topological features of an object. We use different spatial resolution which is filtration levels to extract the features of images. It makes us analyze the images from a different viewpoint.

Self-Updating Process (SUP) is a clustering analysis method with robustness and noise isolation. We provide a high quality and high performance SUP software with GPU acceleration ready for any scale of data.

We aim to test the hypothesis that re-casting multi-label classification as an objection detection problem will significantly improve performance with medical imaging and improve data efficiency.

Radiomics is a method to describe the characteristics of tissues from medical images. These quantitative radiomic features include the information which is more than we can see by our eyes. The process of radiomic approach are image acquisition, region of interest (ROI) segmentation, radiomic feature extraction, and model building.

After radiation therapy, postoperative follow-up is a very important and indispensable part. By classifying MR images, we can filter out patients who need to follow-up treatment.

Deep learning is a powerful tool for medical image segmentation. However the training of deep learning model needs large amount of data and takes long time for training. With transfer learning, you can start with a pre-trained weight, which can save your time on training and model fine-tuning.

Head and Neck Tumor Response Prediction

We aim to develop radiographic image-derived deep neural network (DNN) based model for the prediction of tumor prognoses in patients with head and neck cancer. With this approach, we could help doctors to choose the proper treatment approaches by using our accurate and efficient predictive DNN model.