Integrative Analysis of Pathological Images and Multi-Dimensional Genomic Data for Early-Stage Cancer Prognosis

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• Cancer complexity arises from multi-level molecular alterations (genetic, epigenetic, transcriptional).
• Previous image-genomic studies often use single genomic data types, limiting comprehensive analysis.
• Integrating diverse data is crucial for understanding cancer mechanisms and improving prognosis.

Purpose of the Study:

• To develop a novel framework for ordinal multi-modal feature selection (OMMFS).
• To simultaneously identify prognostic features from histopathological images and multi-modal genomic data.
• To improve cancer patient survival outcome prediction, especially for early-stage cancers.

Main Methods:

• Utilized a generalized sparse canonical correlation analysis framework.
• Incorporated ordinal survival information for outcome prediction.
• Applied the OMMFS framework to histopathological images and multi-modal genomic data (mRNA, copy number variation, DNA methylation).

Main Results:

• Selected image and multi-modal genomic markers showed strong correlation with patient survival.
• The OMMFS framework effectively stratified patients with distinct survival outcomes.
• Demonstrated superior performance compared to existing methods on early-stage cancer datasets.

Conclusions:

• The OMMFS framework offers a powerful approach for integrating diverse data types in cancer research.
• Selected features provide valuable insights into cancer prognosis and patient stratification.
• This method holds promise for improving survival prediction in early-stage cancers.