HiGGsXplore is a computational pathology pipleine that provide insight into patient gene expression state form digitized images of H&E stained tissue section. The framework take a Whole Slide Image (WSI) as input and generate a WSI-Graph. The WSI-Graph is then passed as input to a Graph neural network which predict the expression state of patient using 200 binary variable (Gene Groups status). Each binary variable represent the collective expression of set of genes genes whose gene expression patterns are significantly statistically dependent and covarying across breast cancer patients. The Gene Groups and their status are biologically meaningful, carry histopathological insights and are clinically relevant in terms of association with survival, pharmaco-sensitivity and therapeutic decision-making. For details descroption please refer to the preprint: [TODO]
Live Demo (Webserver)
Workspace directory contain necessary script for constructing graph and training the proposed SlideGraph∞.
Step1: Download TCGA BRCA Diagnostic slides from GCD data portal
Step2: Download tissue segmentation mask from this Link.
Step3: Generate patches of each Whole slide image by running
python patches_extraction.py
Step4: Extract ShuffleNet representation from each of the WSI patch by running
python deep_features.py
Step5: Construct WSI-Graph by running
python graph_construction.py
Step6: Training the Graph Neural Network by running
python main.py
Step1: Download WSIs of patients in CPTAC-BRCA cohort from CANCR IMAGING ARCHIVE.
Step2: Use the same patch-extraction and feature-extraction pipeline and construct graph representation.
Step3: Run the Inference.
python inference.py
The source code of SlideGraph∞ is released under MIT-CC-Non-Commercial license.