TIA Center Demos

Here we provide links to interactive visualizations of models produced by the TIA Center, that have been created as companions to published work to enable a more detailed exploration of model outputs for interested readers. Upon visiting a visualization, additional information/usage guidance can be seen by scrolling down, below the main slide view window.

Iguana: Interpretable gland graphs for colon screening

Iguana presents a graph representations of colon biopsy tissue slides for screening purposes by using interpretable gland-graphs that reflect diagnostic features. This approach increases model explainability and helps improve the confidence of pathologists in automated diagnoses. Using graphs with meticulously-defined input features provides highly interpretable explanations, which is particularly important in medicine to ensure algorithm fairness and identify potential bias in training data.

MesoGraph: Graph networks for mesothelioma subtyping

This paper proposes a model using a dual-task Graph Neural Network (GNN) architecture with ranking loss to accurately diagnose morphological subtypes of malignant mesothelioma. The approach scores tissue down to cellular resolution to allow quantitative profiling of a tumor sample according to the aggregate sarcomatoid association score of all the cells in the sample, and validates the model predictions through an analysis of the typical morphological features of cells according to their predicted score. The proposed approach has predictive performance and ultimately improves treatment decisions for patients with mesothelioma.

Social-IDaRS: Enhanced deep learning model with social analysis of cell networks

Here, a new approach to enhance deep learning-based methods for predicting molecular pathways and mutations in colorectal cancer is proposed. The approach involves incorporating cell interaction information using cell graphs and Social Network Analysis measures, which are computationally efficient and scalable. The approach demonstrated improved performance for several prediction tasks and provides insights into the correlation between cell interactions and molecular pathways/mutations.

HiGGsXplore: Histology gene groups explorer

Identification of gene expression state of a cancer patient from routine pathology imaging and characterization of its phenotypic effects have significant clinical and therapeutic implications. However, prediction of expression of individual genes from whole slide images (WSIs) is challenging due to co-dependent or correlated expression of multiple genes. Here, we use a purely data-driven approach to first identify groups of genes with co-dependent expression and then predict their status from (WSIs) using a bespoke graph neural network. These gene groups allow us to capture the gene expression state of a patient with a small number of binary variables that are biologically meaningful and carry histopathological insights for clinically and therapeutic use cases. Prediction of gene expression state based on these gene groups allows associating histological phenotypes (cellular composition, mitotic counts, grading, etc.) with underlying gene expression patterns and opens avenues for gaining significant biological insights from routine pathology imaging directly.

ODYN: AI-based Prediction of Malignant Transformation in Oral Epithelial Dysplasia

Oral epithelial dysplasia (OED) is a premalignant histopathological diagnosis given to lesions of the oral cavity that have an increased risk of progression to malignancy. We developed a novel transformer-based pipeline, called the Oral Dysplasia Network (ODYN). ODYN can both classify OED and assign a predictive score (ODYN-score) to quantify the risk of malignant transformation, in haematoxylin and eosin (H&E) stained whole slide images (WSIs). Our pipeline outperformed other state-of-the-art methods, and gained comparable results to clinical grading systems, demonstrating the promise of computational pathology for the automatic detection, diagnosis and prognosis of OED.