Life Science Use Cases

A future revolution in Digital Pathology

Deep Convolutional Neural Networks have proven to enable breakthrough performance in both the classification of images (so mimicking pathology diagnostics) and also the semantic segmentation of cell & tissues types by pixel, which is often required at the front-end of digital imaging workflows.

The accuracy of precision and recall (or false positive and false negative) rates achieved by the best deep learning models have been shown to approach the error rates of human trained physicians already in some applications of medical imaging (breast and lung cancer screening) however slide pathology is significantly more complex.  Whether to be used as decision support or in other ways - it is clear that deep learning will allow significant contributions in this field.

Links to external
academic references
(not from Paradigm Shift AI)
Segmentation & Classification of Cells in Biomarker Discovery Research

State-of-the-Art cell imaging uses fluorescent tags to identify target biomolecules such as proteins or antibodies. These fluorescent tagged images typically have many labelled channels and are used in disease mechanism investigation & drug discovery research. 

Often a high accuracy segmentation of cells in the images (typically a reference nuclear stain image) is used as the foundation of a biomarker discovery workflow - this segmentation can be performed with high accuracy at the pixel and cell object level by using recent advances in deep convolutional neural networks, such as U-NET and Mask R-CNN.

Before or after segmentation, machine learning clustering and many other techniques can be performed to classify biomarkers and correlate these to various experimental conditions and outcomes in disease mechanism discovery research.