Seminar: NuCLS: A Scalable Crowdsourcing Approach and Dataset
Recorded July 29, 2021.
This is a talk I gave at the Crowd Science Seminar at Toloka (Yandex). This presentation summarized the motivation, data collection strategy and deep learning modeling approaches we used in the NuCLS project. Compared to the FDA talk below, this presentation was more geared towards a computer science audience.
Talk at FDA: NuCLS scalable crowdsourcing, deep learning approach and dataset
Recorded Mar 10, 2021.
High-resolution mapping of cells and tissue structures provides a foundation for computational biomarker discovery and analysis, including TILs assessment. This is a public talk by I gave at the U.S. Food and Drug Administration (FDA), describing the public release of 220,000 annotations of cell nuclei in breast cancers.
Presentation: Developing quality ground truth for machine-learning algorithms in pathology
Uploaded May 20, 2020.
A brief introductory video of some of the tools being developed to provide quality ground truth for machine-learning algorithms in pathology. This work is funded by grant U01CA220401.
Seminar: Unbiased learning from large-scale histology images: the case for crowdsourcing
Co-presented at the Computer Science Department Seminar, Emory University, on Friday, February 8, 2019, 10:30 am EST.
We describe a study (Amgad et al, 2019) that coordinated pathologists, residents, and students to generate tens of thousands of image annotations to provide training and validation data for convolutional models in breast cancer. This dataset is the largest of its kind, and enabled the development of realistic state-of-the-art performance for semantic segmentation of breast cancer images.
Seminar: Predicting cancer outcomes with machine learning
Co-presented at Department of Biomedical Informatics, Emory University Academic Exchange seminar, on Oct 17, 2017.
We describe two studies (Yousefi et al, 2017, and Mobadersany et al, 2018) that rely on deep-learning models to predict cancer outcomes (specifically, overall patient survival) using histology imaging data and/or genomics (gene expression, mutations, chromosomal aberrations, etc). The studies used brain cancer (glioma) and gynecological cancer (mainly breast cancer) data from The Cancer Genome Atlas.