Tumor Connectomics: Mapping the intra-tumoral complex interaction network

TL;DR

This paper introduces a tumor connectomics framework that models intra-tumor interactions using graph theory on MRI data, predicting treatment response in breast cancer with high accuracy.

Contribution

The study develops a novel connectomics framework to analyze intra-tumor networks and demonstrates its effectiveness in predicting chemotherapy response.

Findings

Degree centrality predicts treatment response with 83% AUC.

The connectomics approach visualizes tumor network changes over time.

Graph metrics serve as biomarkers for therapy outcome.

Abstract

Tumors are extremely heterogeneous and comprise of a number of intratumor microenvironments or sub-regions. These tumor microenvironments may interact with eac based on complex high-level relationships, which could provide important insight into the organizational structure of the tumor network. To that end, we developed a tumor connectomics framework (TCF) to understand and model the complex functional and morphological interactions within the tumor. Then, we demonstrate the TCF's potential in predicting treatment response in breast cancer patients being treated with neoadjuvant chemotherapy. The TCF was implemented on a breast cancer patient cohort of thirty-four patients with dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI) undergoing neodjuvant chemotherapy treatment. The intra-tumor network connections (tumor connectome) before and after treatment were modeled using advanced graph theoretic centrality, path length and clustering metrics from the DCE-MRI. The percentage change of the graph metrics between two time-points (Baseline and 1st cycle) was computed to predict the patient's final response to treatment. The TCF visualized the inter-voxel network connections across multiple time-points and was able to evaluate specific changes in the tumor connectome with treatment. Degree centrality was identified as the most significant predictor of treatment response with an AUC of 0.83 for classifying responders from non-responders. In conclusion, the TCF graph metrics produced excellent biomarkers for prediction of breast cancer treatment response with improved visualization and interpretability of changes both locally and globally in the tumor.