About Me
Daguang Xu is now a research manager at AI-Infra of NVIDIA. He is leading a research team in healthcare AI, focusing on developing world-class machine learning and deep learning-based methods to solve the challenging problems in medical domain. His current research interest includes but not limited to medical imaging analysis, EHR analysis, computer aided diagnosis, deep learning, pattern recognition and computer vision, etc. He has published 90+ papers on top journals and conferences and has ~50 granted or in-application patents. His team is the main developer of open source software (OSS) MONAI (https://github.com/project-monai) and NVIDIA Flare (https://github.com/nvidia/nvflare).
Daguang Xu received the bachelor degree from the department of Engineering Physics in Tsinghua University, China in 2009. He received the M.S.E and Ph.D. degree from the department of Electrical and Computer Engineers in Johns Hopkins University, USA in 2014, advised by the Jacob Suter Jammer Professor Jin U. Kang. His Ph.D. thesis is on Compressive Sensing for Spectral Domain Optical Coherence Tomography.
After graduation, he joined Siemens Healthineers in Princeton, NJ USA as a research scientist. He developed multiple efficient algorithms in organ segmentation, landmark detection, image classification, tumor detection, annotation and editing, etc. Many of them are translated into clinical products that are used world-wide by clinicians and researchers.
EDUCATION
2009.8 – 2014.12 PhD
Advisor Jacob Suter Jammer Professor Jin U. Kang
2009.8 – 2014.5 Master of Science in Engineering
Advisor Jacob Suter Jammer Professor Jin U. Kang
2005.8 – 2009.7 Bachelor of Engineering
Tsinghua University, China
skills
EXPERIENCE
2018.6 – present
Research Manager, AI-Infra, NVIDIA
Lead a team to develop deep learning based algorithm for medical imaging analysis
Developed Clara Train SDK with transfer learning toolkit and AI-assisted annotation
Developed Clara Federated Learning SDK with privacy preserving and secure communication
Maintain and develop new features of OSS project MONAI (https://github.com/project-monai) and NVIDIA Flare (https://github.com/nvidia/nvflare).
2018.2 – 2018.6
Senior Applied Research Scientist, NVIDIA
Developed deep learning based algorithm for medical imaging analysis
Contributed to the development of a scalable framework based on Tensor flow for better utilization of the GPU cluster
2017.1 – 2018.2
Staff Research Scientist, Siemens Healthineers
Developed two deep learning based algorithms for spine centroid localization and detection in 3D CT image, which improves the state-of-the-art detection rate from 84% to 90%.
Developed an adversarial deep image-to-image neural network for organ segmentation: training with 1000+ annotated 3D CT images (liver,spleen,kidneys); training with 5000+ annotated 3D CT images (lung lobes, airways); training with 1000+ annotated MR images (prostate, liver). The validation on 1000-5000 images shows that these algorithms reduce the average distance error by ~50% comparing to the product in current Siemens CT system. They save the radiologist 70% annotation time.
Developed an anisotropic hybrid neural network for lesion detection in digital breast tomosynthesis. The algorithm is trained using ~3000 3D images and validated on hundreds of images. It achieves sensitivity 80%, specificity 88%, which is comparable to human radiologists.
2014.6 – 2016.12
Senior Research Scientist, Siemens Healthineers
Developed two deep learning based algorithms for automatic detection of kidney centers, which not only reduces the average localization error by 50% but also achieves high robustness to various pathologies.
Developed a product for automatic segmentation and annotation of MR knee image, which is being used by Smith&Nephew LLC now. It helps double their technicians’ efficiency and reduces their annotation cost by more than 50%.
Developed an automatic pipeline for organ segmentation in CT images and integrated it to the Siemens CT scanner system. It can segment liver, spleen, kidneys, lungs, brain and femoral heads within 2min (total time on CPU) with accuracy comparable to human experts.
2013.8 – 2014.2
Research Intern, Siemens Corporate Research
Developed auser-guided shape morphing algorithm for bone segmentation in medical imaging.
Contributed to research projects that develop object detection, tracking and recognition algorithms in 3D image and video analysis.
Developed prototypes(standalone,cloud-based,apps) that implement and show case image and video analysis algorithm.
Worked on bone and vessel segmentation and interactive segmentation editing.
2011.9 – 2014.5
Research Assistant, Photonics and Optoelectronics Lab, Johns Hopkins University
Implemented innovative and robust solutions to research problems in the area of parallel computing and GPU-based acceleration of compressive sensing in optical coherence tomography(OCT).
Conducted an in-depth study of compressive sensing in spectral domain OCT(SD-OCT).
Designed and developed algorithms including compressive sensing with dispersion compensation on non-linear wavenumber sampled SD-OCT, modified compressive sensing OCT with noise reduction, and volumetric (3D) compressive sensing SD-OCT.
2010.1 - 2011.5
Research Assistant, Center For Language and Speech Processing, Johns Hopkins University
Designed and maintained algorithms in statistical machine translation through confusion network based system combination in GALE (Global Autonomous Language Exploitation).
Conducted an in-depth study of Gaussianization with Angular and Radial Normalization.