Related Experiment Video
Updated: Jun 27, 2026
07:13
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Retrospective Cohort Study: Extracting Coexisting Background Breast-Lesion Features from Stage I-III Invasive Breast
Ryan Jak Yang Lim1, Phyu Nitar2, Kah Weng Lau3
1Genome Institute of Singapore (GIS), Agency for Science, Technology and Research (A*STAR), Singapore 138632, Singapore.
Cancers
|June 26, 2026
Summary
Background breast features identified by natural language processing (NLP) reflect tumor biology but do not independently predict patient outcomes in invasive breast cancer. This study demonstrates NLP
Area of Science:
- Oncology
- Pathology
- Bioinformatics
Background:
- Background breast features are often documented in pathology reports but their prognostic significance is poorly understood.
- Existing research has not fully characterized the relationship between these features, tumor characteristics, and patient outcomes.
Purpose of the Study:
- To investigate the association between background breast features and tumor characteristics in invasive breast cancer.
- To evaluate the prognostic value of background breast features for patient survival.
- To assess the feasibility of using natural language processing (NLP) for large-scale extraction of these features from pathology reports.
Main Methods:
- Retrospective cohort study of 7603 patients with Stage I-III invasive breast cancer.
- Natural language processing (NLP) applied to over 9754 free-text pathology reports to extract background breast features.
- Unsupervised hierarchical clustering to categorize extracted features; multinomial logistic regression and Cox proportional hazards models used for analysis.
Main Results:
- NLP achieved over 90% accuracy in extracting background breast features.
- Specific features like lobular neoplasia and benign proliferative changes correlated with less aggressive tumor characteristics.
- Early neoplastic and papillary lesions were more common in HER2-enriched and luminal B subtypes; benign proliferative changes showed a trend towards better survival, attenuated by stage and subtype.
Conclusions:
- NLP-enabled extraction of background breast features from pathology reports is feasible and scalable.
- Extracted background features correlate with underlying tumor biology and subtypes.
- These features do not provide independent prognostic information beyond established clinical variables like stage and subtype.
