HistomicsML

HistomicsML is an interactive machine-learning system for analyzing whole-slide imaging digital pathology datasets. The web-server enables interaction with whole-slide imaging datasets and image analysis metadata to train and validate classification rules for identifying microanatomic objects. HistomicsML operates using data generated by image segmentation and feature extraction algorithms applied to whole-slide images that characterize objects like cell nuclei, and can operate interchangeably with any image segmentation and feature extraction algorithms. The web-interface facilitates classifier training and also provides interfaces for generating ground-truth validation data, validating classifiers, loading data, and exporting results.

• Installation
  • Installing HistomicsML via Docker
• Training a classifier
  • Initializing the classifier
  • Instance-based active learning
  • Heatmap-based active learning
  • Reviewing a training set
• Validating a classifier
  • Generating a validation set
  • Reviewing a validation set
  • Validating a classifier
• Exporting results
  • Report with training dataset
• Importing datasets
• Formatting datasets
  • Whole-slide images
  • Slide description
  • Object boundaries
  • Histomic features

Resources

• Paper: [M Nalisnik et al, “Interactive phenotyping of large-scale histology imaging data with HistomicsML”](http://www.biorxiv.org/content/early/2017/05/19/140236).

• Project source: HistomicsML (https://github.com/CancerDataScience/HistomicsML.git).

• Related projects: HistomicsTK (https://github.com/DigitalSlideArchive/HistomicsTK) and HistXtract (https://github.com/CancerDataScience/HistXtract).