- The National Institutes of Health’s Clinical Center is making public a massive data set of images to help researchers improve detection of cancer lesions. Named DeepLesion, the collection contains CT images of more than 32,000 lesions from across the body, ranging from lung nodules to liver tumors.
- Most publicly available medical image data sets have less than a thousand lesions. The hope is that release of such a large data set will lead to the development of a universal lesion detector that could serve as an initial screening tool, sending results to specialist systems for different types of lesions, the hospital said.
- Relationships between different lesion types can be studied and whole-body assessments of the cancer burden made. Data from more than 4,400 anonymous NIH patients are measured and marked with an electronic bookmark tool.
The NIH Clinical Center is the nation’s largest hospital devoted to clinical research, focusing on first-in-human trials and treatment of rare and difficult-to-treat diseases. Patients consent to participate in research studies and are treated without charge.
James Gilman, a cardiologist and U.S. Army veteran, became the center’s first chief executive officer in January 2017 with a mandate to develop new hospital standards for patient safety and care. His appointment followed the revelation that a task force appointed by NIH Director Francis Collins had found a series of gaps in safety oversight at the 2,000-bed hospital.
The center said it plans to add additional images to the DeepLesion data set to improve its detection capabilities, and expects greater reliability with the eventual addition of 3D information. It may be possible to extend DeepLesion to other image modalities such as MRI, as well as combine data from multiple hospitals.
The NIH Clinical Center boasts a long list of medical milestones, including development of chemotherapy for cancer, first use of an immunotoxin to treat a malignancy, the first gene therapy and the first treatment of AIDS (with AZT).