Computer Vision News - December 2016

Solution This challenge aims at improving the detection of cancer metastasis in lymph node images: the goal is to develop algorithms for a (fully) automated analysis of whole-slide images to detect or grade cancer, to predict prognosis or identify metastases. Results Two strategies were retained for evaluating the performance of the algorithms, giving way to two leaderboards: (A) Slide-based Evaluation: how the algorithms discriminate between slides containing metastasis and normal slides. (B) Lesion-based Evaluation: a free-response receiver operating characteristic (FROC) curve, defined as the plot of sensitivity versus the average number of false-positives per image. Both leaderboards were led by Harvard Medical School teams. We would like to mention the team formed by the Harvard Medical School (BIDMC) and the Massachusetts Institute of Technology (CSAIL): the authors explain in this page both their methods and results : their area under the ROC curve surpassed the one given by the pathologist in the study. The second year’s challenge CAMELYON17 will strengthen the challenge by moving from slide-level analysis to patient-level analysis (that means combining the assessment of multiple lymph node slides into one outcome). This is expected to bring the efforts closer to direct usefulness in clinical setting. Compared to CAMELYON16, the dataset will be significantly extended and will contain images from four medical centers and pathology labs. Registrations are already open at the challenge website and the CAMELYON17 workshop will be held at ISBI 2017 in Melbourne, Australia. “Improving the detection of cancer metastasis in lymph node images” Challenge Computer Vision News Challenge 29 Sentinel lymph node: the first lymph node to which cancer cells are most likely to spread from a primary tumor