A BERT-GNN Approach for Metastatic Breast Cancer Prediction Using Histopathology Reports

Abdullah Basaad ¹, Shadi Basurra ¹, Edlira Vakaj ¹

⁰School of Computing and Digital Technology, Birmingham City University, Birmingham B4 7XG, UK.

Diagnostics (basel, Switzerland) +

|

July 13, 2024

View abstract on PubMed

Summary

This study introduces a novel BERT-GNN model for non-invasive metastatic breast cancer (MBC) prediction using histopathology reports. The model achieves high accuracy in identifying MBC patients, offering a promising tool for improved cancer diagnostics.

Area Of Science

Oncology
Bioinformatics
Artificial Intelligence

Background

Metastatic breast cancer (MBC) is a significant cause of cancer mortality in women.
Accurate and early identification of MBC is crucial for effective treatment and patient outcomes.
Current diagnostic methods may have limitations in detecting cancer metastases non-invasively.

Purpose Of The Study

To develop and evaluate an innovative non-invasive classification model for predicting metastatic breast cancer (MBC).
To leverage the power of large language models (LLMs) and graph neural networks (GNNs) for MBC patient identification.
To explore the utility of histopathology reports in predicting MBC using advanced AI techniques.

Main Methods

A novel BERT-GNN approach (BG-MBC) was developed, integrating graph information derived from BERT embeddings of histopathology reports.
Nodes were constructed from patient medical records, and BERT embeddings vectorized word representations to capture semantic information.
Feature selection methods including univariate selection, extra trees classifier, and Shapley values identified the top 30 crucial features from 676 embeddings.

Main Results

The BG-MBC model demonstrated outstanding predictive performance with a detection rate of 0.98 and an Area Under the Curve (AUC) of 0.98.
The model effectively utilized attention scores from the LLM to capture pertinent features from histopathology reports for classification.
The study identified the most impactful features contributing to MBC prediction.

Conclusions

The developed BERT-GNN model (BG-MBC) shows significant promise as a non-invasive tool for accurate metastatic breast cancer prediction.
Integrating LLMs and GNNs with histopathology report analysis offers a powerful approach for improving cancer diagnostics.
Further research is warranted to validate these findings and explore clinical applications of this innovative model.

Keywords:

BERT GNN LLM MBC XAI extra trees classifier node classification random forest classifier univariate selection