This post is to announce our first data release and scientific project, NYUMets_Brain v1.0. The dataset was extracted from one of the world’s largest clinical registeries of metastatic brain cancer. We chose to begin our initiative with a focus on metastatic brain cancer for many reasons. Metastatic brain cancer is particularly hard to treat, as the brain’s natural resistance mechanisms and fragility also make it more difficult to treat using standard drugs, radiation techniques, and surgery. On the other hand, the imaging data is particularly easy to standardize and manipulate relative to imaging within other organ systems.

This dataset is most notable for its focus on encouraging scientists and clinicians alike to focus on cancer dynamics - how does metastatic cancer change over time? What can this tell us about such a complex disease? However, the dataset encompasses:

1429 patients with an average of 6 imaging studies per patient over an average time of 17 months 8003 MRI studies containing the following sequences: Expert segmentations: 2367 T1 pre contrast: 5,542 T1 post contrast: 7,245 High-res T1 post contrast: 1,462 T2: 6,920 FLAIR: 5,322 4860 timepoints of clinical follow-up 81,562 medication updates

The concept of cancer dynamics and using data science to better understand this incredibly complex phenomenon are what is most important though, and that is why this dataset is being accompanied by the release of the NYUMets API and an accompantying scientific paper for both structuring the data as well as accessing it computationally and offering some of our initial thoughts towards new ways of thinking about metastatic cancer and leveraging the dataset. We need to fundamentally rethink how we approach cancer analytics if we are to better understand and ultimately treat it. When physicians think about cancer they think about it many different dimensions and modalities: genetics, imaging, clinical, sociological, and so forth. Even more importantly though, physicians don’t just think about cancer as an independent entity, but as a disease belonging to a human being who is living in time with their cancer. Thinking in terms of time is fundamental to how we perceive cancer as people, but it is so often absent from how we model it as computer scientists. NYUMets, starting with metastatic brain cancer v1.0, is our attempt to change this way of thinking.

Link to NYUMets Brain Manuscript

Katherine E. Link
Machine Learning Engineer, NYU OLAB

Participating NYU Langone Teams

NYU OLAB: For more details on the AI experience see the NYU OLAB Website
NYU Department of Neurosurgery
NYU Department of Radiation Oncology
NYU Department of Radiology
NYU Center for Data Science
NYU Langone Health Perlmutter Cancer Center