Many resources are available across NIH that can assist you with machine learning / deep learning solutions. See the list below to learn about current NIH resources and programs that may provide just the information you need. If you don't see what you need -- or to suggest a new user group -- let us know by sending a brief description about your work and your contact information to George Zaki: email@example.com
Nucleus segmentation from fluorescent images is a necessary and primary step for many processing pipelines in quantitative bioimaging applications. For instance, genomic spatial organization using DNA oligos-based fluorescent in situ hybridization (FISH) requires accurate identification of nucleus envelope. Similarly, high content imaging-based phenotype screens using small molecules (e.g., siRNA, chemical libraries) also requires accurate detection of (sub) cellular objects to quantify the effects of small molecules. HiTIF develops and make use of recent advances in deep convolution neural networks (D-CNN) for object detection which have surpassed, both qualitatively and quantitatively, traditional image processing approaches.Contact: firstname.lastname@example.org or email@example.com
The immunotherapy group works on the development of a digital health clinic pioneering new applications for wearable devices, patient reported outcomes and other digital applications to facilitate efficient, effective care at the NIH Clinical Center. The immunotherapy group has a partnership with the FDA and several leading academic and industry organizations.
The Molecular Imaging Branch (MIB):
The Molecular Imaging Branch (MIB) researches and develops artificial intelligence (AI) solutions for clinical imaging. Current work heavily focuses on challenging applications in prostate cancer imaging, including: (1) the detection, grading, and staging of prostate cancer by multiparametric MRI (mpMRI), (2) computer-user interaction studies evaluating the impact of AI-based systems for improving radiologist detection and agreement in mpMRI interpretation, (3) characterization of bone disease on clinical imaging, including classification of benign bone conditions to distinguish from those of metastatic origin. MIB is also actively involved in AI initiatives in other clinical imaging realms, including digital pathology, for automated detection of morphological architectures of prostate cancer, as well as synergistic applications between mpMRI and digital pathology for enhanced detection and characterization of aggressive prostate disease.
MIB AI lead: Dr. Baris Turkbey (website)
MIB Director: Dr. Peter Choyke
Recent highlighted publications:
- Gaur et al. “Can computer-aided diagnosis assist in the identification of prostate cancer on prostate MRI? a multi-center, multi-reader investigation (PMID: 30333911)
- Greer et al “Computer-aided diagnosis prior to conventional interpretation of prostate mpMRI: an international multi-reader study” (PMID: 29651763)
High quality delineation of important features is critical in biomedical image interpretation for accurate diagnosis and disease assessment. Commonly, biomedical image interpretation was performed by human experts but image interpretation by humans is limited by many factors such as fatigue and variations across human interpreters. Deep Learning (DL) techniques such as Deep Convolutional Neural Networks (DCNNs) had been highly successful in image classification and segmentation tasks, potentially promising higher throughput and more consistent results in biomedical image interpretation. The Imaging and Visualization Group (IVG) at Frederick National Laboratory for Cancer Research (FNLCR) collaborates with NCI and FNLCR imaging laboratories to develop Deep Learning based image segmentation workflows aiming at augmenting image understanding and interpretation in cancer and biomedical research. Specifically, IVG has been working on DL based tumor segmentation for preclinical cancer models on MRI images and DL based histo-morphological feature quantification and correlation on digital pathology whole slide images.
High Performance Computing Group:
The National Cancer Institute and the Department of Energy have formed a partnership to accelerate key challenges in cancer research, Joint Design of Advanced Computing Solutions for Cancer (JDACS4C): to provide better understanding of the disease, to make effective use of the ever-growing volumes and diversity of cancer related data to build predictive models, and, ultimately, to provide guidance and support decisions on anticipated effective treatments for individual patients. The CANcer Distribute Learning Environment (CANDLE) is a technology that is used to address these challenges on the NIH HPC cluster Biowulf and the next generation of exascale computing systems.
Using CANDLE on Biowulf tutorial
CANDLE benchmarks on GitHub
Contact: George Zaki (firstname.lastname@example.org)
NIH Clinical Center
The Summers Lab focuses on developing techniques for the automated analysis and diagnosis of radiology images. The Lab is internationally recognized for its research on using deep learning to diagnose a wide variety of challenging diseases, such as detecting pre-cancerous polyps on virtual colonoscopy, enlarged lymph nodes in cancer patients, and a wide variety of spine ailments. More details can be found at the lab website:
National Institute of Mental Health (NIMH)
National Library of Medicine (NLM)
National Center for Biotechnology Information (NCBI)
Natural Language Processing (NLP); Biomedical Text Mining; and Medical Image Analysis
We focus on the development of machine learning methods for text and data mining and its applications in PubMed searches, knowledge discovery and clinical text analysis. Through working with domain experts on the NIH campus, we are also applying deep learning to retinal and radiology images for autonomous disease diagnosis and prognosis.
Contact: Zhiyong Lu (PI) email@example.com