HistomicsStream
Contributed fixes and features to HistomicsStream, the whole-slide-image TensorFlow reader described above.
Notable Work
Selected projects, research, and contributions
A closer look at work referenced on the main site — research collaborations, an entrepreneurship award, and open-source contributions.
AI-accelerated digital pathology — Kitware & Northwestern
As part of an NIH National Cancer Institute-funded collaboration (grant 5U01CA220401-04, "Informatics Tools for Quantitative Digital Pathology Profiling and Integrated Prognostic Modeling"), contributed to building the technical infrastructure that makes whole-slide pathology images usable for machine learning at scale.
Whole-slide images (WSIs) used in cancer diagnosis are enormous — a single slide can be 100,000 × 100,000 pixels, or roughly 30 gigabytes uncompressed. Standard machine learning pipelines aren't built to read data at that scale efficiently, which becomes the bottleneck in training and running diagnostic models. The team, including Kitware's Lee A. Newberg and Matt McCormick and Northwestern's Lee Cooper, built HistomicsStream, a Python package that reads WSI data directly into TensorFlow pipelines.
65%
Faster tile throughput vs. standard large-image reads
8,316
Tiles processed per slide in the benchmark test
4
Institutions collaborating (Kitware, Northwestern, Wake Forest, Emory)
Co-authored the public writeup of this work on Kitware's technical blog, alongside Lee A. Newberg, Matt McCormick, Samantha Schmitt, and Lee Cooper.
Impact Award — UNB TME Student Pitch Competition
In 2015, a record-setting year for UNB's J. Herbert Smith Centre for Technology Management & Entrepreneurship Student Pitch Competition — 33 teams, 60+ students, over $8,000 in prizes awarded — KnowYourChild won the Impact Award, pitched together with Muhammad Saad Amjad (PhD, CS) and Rizwas Ali (MCS).
The competition drew sponsorship and judging from Cox & Palmer, NBIF, FCNB, the Pond-Deshpande Centre at UNB, Stantec, Innovatia, Mariner Partners, and J.D. Irving.
Open-source contributions
Code contributed to the Digital Slide Archive ecosystem and original tooling built for GPU-accelerated inference pipelines.
WSI Superpixel Guided Labeling
Contributed to this Girder 3 plugin, used alongside HistomicsUI and HistomicsTK to support active-learning labeling workflows on whole-slide images.
Girder–Triton Interface
An original interface connecting Girder (the data-management layer behind the Digital Slide Archive) with NVIDIA Triton Inference Server, built to support GPU-accelerated model inference for pathology imaging workflows.