Study
PLCO
(Learn more about this study)
Project ID
PLCOI-1709
Initial CDAS Request Approval
Oct 16, 2024
Title
Harnessing TME depicted by histological images to improve lung cancer prognosis through a deep learning system
Summary
We want to propose a deep learning-based system to enhance the information of tumor microenvironment (TME) by analyzing histological images to improve the prognostic accuracy of lung cancer. We wanted to design an integrated graph and image deep learning model (IGI-DL) that is able to predict ST expression based on histological images. The system consists of two main parts: the first part is IGI-DL model, which combines convolutional neural network (CNN) and graph neural network (GNN) to process H&E stained histological images and map them to gene expression space; The second part is a super-patch graph based on spatial gene expression prediction, which is used to predict the prognosis of lung cancer. This model provides a method to enhance TME information without the need for spatial transcriptome (ST) data to achieve precise lung cancer prognosis. This result not only improves the accuracy of survival prediction, but also provides a powerful tool for cancer treatment decisions in clinical practice. By exploiting the rich morphological information in histological images, this system is expected to advance precision medicine and provide a new perspective for future AI applications in computational pathology and healthcare more generally.
Aims
1.To provide a new tool: to introduce an integrated graph and image deep learning (IGI-DL) model that is able to predict molecular level information of tumor microenvironment by analyzing histological images without spatial transcriptome (ST) data.
2.Demonstration of model performance: The improvement in prediction accuracy of the IGI-DL model is demonstrated by comparing it with existing methods.
3.To verify the independent prognostic value of the model: By constructing a super-patch graph based on spatial gene expression and combining with clinical features, the effectiveness of the risk score predicted by the system as an independent cancer prognosis indicator was verified.
4.Advancing precision medicine: By providing more accurate prognosis prediction, it helps clinicians to develop more personalized treatment plans for cancer patients, thereby promoting the development of precision medicine.
Collaborators
