Mixture classification model based on clinical markers for breast cancer prognosis

Area of Science:

• Biomedical Informatics
• Computational Biology
• Cancer Research

Background:

• Accurate cancer prognosis is crucial for effective treatment planning.
• Existing prognosis models based on clinical markers have limitations.
• Gene expression data offers potential biomarkers but faces challenges like high dimensionality and noise.

Purpose of the Study:

• To develop an accurate classification model for cancer prognosis utilizing clinical data.
• To leverage information often overlooked by existing high-accuracy prediction methods.
• To address the limitations of gene expression-based biomarkers in clinical applications.

Main Methods:

• A novel mixture classification model is proposed, integrating Rough Set and Support Vector Machine (SVM) classifiers.
• The two-layer MRS model uses Rough Set for initial sample identification and SVM for final classification.
• The MRS model was evaluated on two breast cancer datasets (BRC-1 and BRC-2).

Main Results:

• On the BRC-1 dataset, MRS achieved high accuracy, correctly classifying all poor prognosis cases and most good prognosis cases.
• MRS performance surpassed previous research and 70-gene based biomarkers on the BRC-1 dataset.
• Comparative analysis on the BRC-2 dataset using 5-fold cross-validation showed MRS outperformed other representative methods in prediction accuracy.

Conclusions:

• The proposed mixture classification model effectively integrates diverse methods and overcomes ensemble model limitations.
• MRS maximizes the utility of clinical data for prognosis prediction.
• The implemented MRS classifier provides more accurate breast cancer prognosis than prior methods.