ISBI is the International Symposium on Biomedical Imaging, a popular venue for furthering the type of work you're seeing in this chapter. Their annual conferences often feature multiple challenges posed to the academic community. They posed several challenges in 2016.

One popular challenge was the AIDA-E: Analysis of Images to Detect Abnormalities in Endoscopy. The challenge website is http://isbi-aida.grand-challenge.org/.

Another popular challenge was the Cancer Metastasis Detection in Lymph Nodes, which features pathology data. The challenge website is http://camelyon16.grand-challenge.org/.

On the radiology side, a popular challenge in 2016 was the Data Science Bowl challenge on heart disease diagnosis. Titled Transforming How We Diagnose Heart Disease, the challenge sought to segment parts of the cardiac Magnetic Resonance Imaging data to gauge pump volume, which was then used as a proxy for heart health. The challenge website and dataset is http://www.datasciencebowl.com/competitions/transforming-how-we-diagnose-heart-disease/.

Another popular radiology dataset is the Lung Image Database Consortium's computed tomography (CT) data in the LIDC-IDRI image collection. This is a dataset of diagnostic and lung cancer screening thoracic CT scans. Interestingly, instead of image-level classes, this dataset annotates the actual locations of the lesions.

The two radiology competitions are interesting for two more reasons:

• They feature three-dimensional volume data, which is essentially an ordered stack of two-dimensional images that form an actual space.
• They feature segmentation tasks where you want to classify parts of an image or volume into certain classes. This is a familiar classification challenge, except we're trying to also localize the feature on the image. In one case, we seek to localize the feature and point to it (rather than classify the entire image), and in another case, we seek to classify a section as a way to measure the size of a region.

We'll speak more about dealing with volume data later, but for now, you've got some really interesting and varied datasets to work with.