Large Scale Whole Slide Image Analysis » Computational Microscopy Imaging Laboratory » College of Medicine

Screenshot of a histology image analysis software interface. The interface is divided into four main sections: 'Input Files' on the left, 'Image Viewing' in the center, 'Select Features' to the right of 'Image Viewing', and 'Analysis Toolbox' on the far right. The 'Image Viewing' section displays a microscopic image of tissue, with accompanying controls for annotation and labeling. The 'Input Files' section shows a file path and options for loading data. The 'Select Features' section contains sliders and options for configuring feature extraction. The 'Analysis Toolbox' section provides tools for statistical analysis and visualization. Below the sections are brief descriptive labels.

HistoLens

We developed HistoLens, a desktop software for quantitative analysis of annotated structures in histological Whole Slide Images (WSIs). It enables dynamic labeling, interactive figure generation, feature visualization, classification training, and lesion annotation for downstream analysis.

Diagram illustrating a CNN (Convolutional Neural Network) parallel processing system. A large cloud shape labeled 'CMIL, Computer Vision & Machine Learning' is connected by lines to two CNN blocks, one at the top and one at the bottom. The top block is labeled 'CNN' and the bottom 'High performance cloud computing'. A 'Parallel processing' label connects the two CNN blocks. Each CNN block also includes icons representing computer processing units.

Cloud Based Digital Pathology

We created a cloud based framework for visualization of large digital Whole Slide Images (WSIs), as well as detection, segmentation, and quantification of structures for WSIs via press of a button.

Diagram illustrating a Convolutional Neural Network (CNN) architecture. On the left, a CPU chip is shown with an arrow pointing to it labeled "Train using selected patches". On the right, a GPU with cooling fans is shown with an arrow pointing away from it. Connecting the CPU and GPU is a large rectangle containing the text "CNN". An arrow points from the top of the rectangle towards a series of connected nodes, representing the CNN’s output. An arrow on the left side of the rectangle points towards the CPU. The diagram illustrates the interaction between CPU training and CNN processing through the GPU.

Histo-Fetch: On-the-Fly Processing of Gigapixel WSIs

We created a pipeline to efficiently extract random patches and labels from digital pathology WSIs for input to a neural network on-the-fly.

Flowchart illustrating a process: WSI down sampling, augmentation, network training, AI prediction, and hand correction feeding back to WSI donwsampling and chopping.

Human Artificial Intelligence Loop (HAIL)

We have developed an AI tool to annotate large digital pathology WSIs and display AI predictions in a commonly used WSI viewer.