Workshop on Computational Challenges in Digital Pathology

Abstract: Neoadjuvant chemotherapy (NAC) is the systemic treatment of breast cancer patients prior to surgery. The response to NAC can guide treatment decisions post-operatively and it is assessed by examining the excised tissue from the primary tumor site and resected lymph nodes. The residual cancer burden (RCB) index, proposed by Symmans et al, incorporates measurements of the primary tumour with an assessment of metastatic spread in the lymph nodes and this measure is predictive of long-term survival after NAC. In practice, however, the assessment of RCB is too time consuming and most pathologists simply report whether there is a complete response or residual tumour. We have been working closely with breast pathologists to develop AI models that can carry out the RCB assessment automatically. In this talk I will outline some of the challenges we have encountered when working with WSIs which include working with limited data sets, and with sparse or incomplete annotations. I will also present some of approaches we have taken to overcome these challenges.

About Anne Martel: Dr. Anne Martel is a Professor in Medical Biophysics at the University of Toronto, a Senior Scientist at Sunnybrook Research Institute, and a Vector Faculty Affiliate. Her research program, funded by CIHR, NSERC, NIH and the Canadian Cancer Society, is focused on medical image and digital pathology analysis, particularly on applications of machine learning for segmentation, diagnosis, and prediction/prognosis. In 2006 she co-founded Pathcore, a software company developing complete workflow solutions for digital pathology. Dr. Martel is an active member of the medical image analysis community and is a fellow of the MICCAI Society which represents engineers and computer scientists working in this field. She has served as board member of MICCAI and as an Associate Editor of IEEE Transactions on Medical Imaging.

"Guided Representation Learning for the Classification of Hematopoietic Cells" by Philipp Gräbel, Martina Crysandt, Barbara M. Klinkhammer, Peter Boor, Tim H. Brümmendorf and Dorit Merhof

"H&E-adversarial network: a convolutional neural network to learn stain-invariant features through Hematoxylin & Eosin regression" by Niccolò Marini, Manfredo Atzori, Sebastian Otálora, Stephane Marchand-Maillet and Henning Müller

"Self-Supervised Representation Learning using Visual Field Expansion on Digital Pathology" by Joseph Boyd, Mykola Liashuha, Eric Deutsch, Nikos Paragios, Stergios Christodoulidis and Maria Vakalopoulou

"ALBRT: Cellular Composition Prediction in Routine Histology Images" by Muhammad Dawood, Kim Branson, Nasir Rajpoot and Fayyaz Minhas

"Robust Interactive Semantic Segmentation of Pathology Images with Minimal User Input" by Mostafa Jahanifar, Neda Z. Tajeddin, Navid A. Koohbanani and Nasir Rajpoot

"Simultaneous Nuclear Instance and Layer Segmentation in Oral Epithelial Dysplasia" by Adam J. Shephard, Simon D. Graham, Saad Bashir, Mostafa Jahanifar, Hanya Mahmood, Ali Khurram and Nasir Rajpoot

"Improving Self-supervised Learning with Hardness-aware Dynamic Curriculum Learning: An Application to Digital Pathology" by Chetan L. Srinidhi and Anne Martel

"Probeable DARTS with Application to Computational Pathology" by Sheyang Tang, Mahdi S. Hosseini, Lina Chen, Sonal Varma, Corwyn Rowsell, Savvas Damaskinos, Konstantinos N. Plataniotis and Zhou Wang

"Iterative Cross-Scanner Registration for Whole Slide Images" by Luisa Theelke, Frauke Wilm, Christian Marzahl, Christof Bertram , Robert Klopfleisch, Andreas K. Maier, Marc Aubreville and Katharina Breininger

"Joint Semi-supervised and Active Learning for Segmentation of Gigapixel Pathology Images with Cost-Effective Labeling" by Zhengfeng Lai, Chao Wang, Luca C. Oliveira, Brittany N. Dugger, Sen-Ching S. Cheung and Chen-Nee Chuah

"An investigation of attention mechanisms in histopathology whole-slide-image analysis for regression" by Philippe Weitz, Yinxi Wang, Johan Hartman and Mattias Rantalainen

"Multi-Prototype Few-shot Learning in Histopathology" by Jessica Deuschel, Daniel Firmbach, Carol Geppert, Markus Eckstein, Arndt Hartmann, Volker Bruns, Petr Kuritcyn, Jakob Dexl, David Hartmann, Dominik Perrin, Thomas Wittenberg and Michaela Benz

"A Pathology Deep Learning System Capable of Triage of Melanoma Specimens Utilizing Dermatopathologist Consensus as Ground Truth" by Sivaramakrishnan Sankarapandian, Saul Kohn, Vaughn Spurrier, Sean Grullon, Rajath E. Soans, Kameswari D. Ayyagari, Ramachandra V. Chamarthi, Kiran Motaparthi, Jason B. Lee, Wonwoo Shon, Michael Bonham and Julianna D. Ianni1

"A QuadTree Image Representation for Computational Pathology" by Robert Jewsbury, Abhir Bhalerao and Nasir Rajpoot

"Deep Ordinal Focus Assessment for Whole Slide Images" by Tomé Albuquerque, Ana Moreira and Jaime S. Cardoso

"Lizard: A Large-Scale Dataset for Colonic Nuclear Instance Segmentation and Classification" by Simon D. Graham, Mostafa Jahanifar, Ayesha Azam, Mohammed Nimir, Yee-Wah Tsang, Katherine Dodd, Emily Hero, Harvir Sahota, Atisha Tank, Ksenija Benes, Noorul Wahab, Fayyaz Minhas, Shan Raza, Hesham El Daly, Kishore Gopalakrishnan, David Snead and Nasir Rajpoot

"Real-Time Cell Counting in Unlabeled Microscopy Images" by Yuang Zhu, Zhao Chen, Yuxin Zheng, Qinghua Zhang and Xuan Wang

Abstract: Precision medicine is increasingly leveraging data from multiple modalities for creating more comprehensive insights into the diagnostic state of a patient, and more accurate predictions of their expected health trajectory. While genomic data from DNA- and RNA-sequencing yields salient information about a patient’s molecular priors, and clinical data about their historic disease context, imaging has the unique ability to yield critical information about the spatial disease context, e.g. of the tumor microenvironment in oncology. This talk addresses how these different modalities can be used in a multimodal AI precision medicine environment, with example applications ranging across biomarker identification, clinical trial enrollment, risk and response predictions, and operational improvements using digital pathology.

About Martin Stumpe: Dr. Martin Stumpe serves as the Chief of AI at Tempus. Prior to joining Tempus, he founded and led the Cancer Pathology project at Google, which uses artificial intelligence to increase the accuracy in image-based cancer detection and grading. His team also succeeded in integrating these diagnostics into optical microscopy for real-time augmented intelligence. Earlier in his career, Dr. Stumpe was part of the Kepler team at NASA Ames Research Center in Mountain View, California, where he was responsible for the development of algorithms to detect extrasolar planets from stellar light curves. Before that, he was a postdoctoral researcher at Stanford University in the lab of Vijay Pande, focusing on molecular dynamics simulations to study chaperone-induced protein folding. During this time, he also developed and brought to market a particle tracking software which is used primarily in ethological studies. Dr. Stumpe holds a PhD in Computational and Theoretical Physics from the Max-Planck-Institute for Biophysical Chemistry in Goettingen, Germany.