A deep matrix completion method for imputing missing histological data in breast cancer by integrating DCE-MRI...

Area of Science:

• Oncology
• Medical Imaging
• Data Science

Background:

• Histological data are crucial for breast cancer treatment decisions.
• Missing histological data, such as Ki-67 expression, subtype, and grade, hinder accurate analysis.
• Tumor radiomics extracted from DCE-MRI offer a potential solution for data imputation.

Purpose of the Study:

• To develop and evaluate a deep matrix completion (DMC) method for imputing missing breast cancer histological data.
• To integrate tumor radiomics with existing histological information for improved data imputation accuracy.
• To assess the performance of DMC against traditional imputation methods like NMF and MCF.

Main Methods:

• A deep matrix completion (DMC) method was proposed to impute missing histological data.
• Radiomic features (morphologic, statistical, texture) were extracted from dynamic contrast-enhanced MRI (DCE-MRI).
• The performance of DMC was compared with non-negative matrix factorization (NMF) and collaborative filtering (MCF) using AUC.

Main Results:

• The DMC method demonstrated superior performance in imputing missing histological data compared to NMF and MCF.
• DMC using 120 radiomic features achieved an optimal AUC of 0.793, outperforming NMF (0.756) and MCF (0.706).
• DMC consistently outperformed NMF and MCF across various missing data rates and feature numbers.

Conclusions:

• Deep matrix completion effectively imputes missing histological data by integrating radiomics.
• This approach bridges imaging-scale patterns with molecular-scale histological information.
• The integrated radiohistological approach shows promise for enhanced tumor characterization and patient management.