Workshop and Challenges in Imaging & Digital Pathology
Challenge 1: Primary Brain Tumor Digital Pathology Nuclear Segmentation
The characteristics of cancer nuclei are central components in many aspects of pathology classification and nuclear features, combined with “omics” have been shown by many research groups to be linked to patient outcome. Although there are many important and predictive features but characteristics of cancer nuclei are overall the most important. Virtually without exception, all commercial systems and academic groups in the digital Pathology area make use of nuclear segmentation algorithms. Given this, ability to segment and then classify nuclei is a key task and a very appropriate challenge topic.
The reference standard will be pathologist generated nuclear segmentation on select regions of TCGA Low Grade Glioma whole slide images for the challenge.
The contestants will be tasked with applying their segmentation algorithms, previously trained on the training data, to segment all nuclei in a tile region. Their results will be submitted online and compared with consensus pathologist segmented sub regions. Winners will be ranked based on their nuclear segmentations best matching the reference standards.
Challenge 2: Imaging and Digital Pathology Primary Brain Tumor Classification
This challenge will help bring expertise in image processing in digital pathology and radiology closer by working on a single task of tumor classification. In the long term this type of collaboration will help reduce discordance and increase diagnostic accuracy of brain tumors. In addition, by computationally encompassing both digital pathology and clinical brain tumor MRI as resources in this Challenge it may generate a result that could contradict present-day classification orthodoxies of tumor stage and aggressiveness that have to date been primarily based on subjective histopathology observations.
Contestants will be allowed to use algorithms of their choice to classify brain tumors into low grade II and low grade III gliomas. They may choose to perform segmentation to extract relevant information from pathology and imaging data. The reference standards will be the result of segmentations performed by expert neuro radiologist and neuropathologist. Winners will be ranked based on the most number of correct tumor classifications derived.
Challenge 3: Guess the Primary based on Brain Mets
The overall objective of this challenge is to study imaging phenotypes of metastatic disease originating from different primary cancers in the body. We challenge the image segmentation community to determine if primary cancers create distinct imaging signatures that can be automatically detected with segmentation algorithms. The impact of successfully predicting the histology of the primary tumor based on features of the brain metastasis would be to provide a biological insight into brain tumor imaging and segmentation that could result in improved clinical treatment decisions.
Through this exercise, we will develop algorithms that will advance segmentation methods for metastatic cancer in the brain. Using one or more of these algorithms, we hope to gain insight into the underlying biology and growth patterns of metastatic lesions and perhaps, their proclivity to seed and grow differently based on the inherent phenotype they carry with them from the primary cancer organ/site. An additional objective will be to develop image processing algorithms that segment, process and analyze MRI sequences for brain metastasis.
Organizers of Workshop and Challenges
- Keyvan Farahani, National Cancer Institute
- John Freymann, Leidos Biomedical Research
- Carl Jaffe, Boston University
- Jayashree Kalpathy-Cramer, MGH Harvard
- Justin Kirby, Leidos Biomedical Research
- Arno Klien, Sage Bionetworks
- Karim Lakhani, Harvard Business School
- Bjoern Menze, TU Munich, INRIA Sophia-Antipolis;
- Tahsin Kurc, Stony Brook Cancer Center
- Russell C. Rockne, City of Hope Cancer Center
- Joel Saltz, Stony Brook Cancer Center
- Andrew Trister, Sage Bionetworks