Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interpretable Solutions for Breast Cancer Diagnosis with Grammatical Evolution and Data Augmentation

Yumnah Hasan, Allan de Lima, Fatemeh Amerehi, Darian Reyes Fernandez de Bulnes, Patrick Healy, Conor Ryan

TL;DR

This work tackles the dual challenges of imbalanced data and interpretability in breast cancer diagnosis by pairing Grammatical Evolution (GE) with the STEM data-augmentation technique. Using Haralick texture features extracted from DDSM and Wisconsin Breast Cancer datasets, the authors show that GE models achieve superior AUC while remaining inherently interpretable, outperforming an ensemble of eight traditional classifiers. Bayesian signed-rank analysis further establishes STEM as the top augmentation strategy across multiple data configurations, highlighting the method's robustness. The approach holds promise for clinically actionable decision support by delivering transparent, high-performing models and guiding future improvements with feature-level insights and cross-dataset validation.

Abstract

Medical imaging diagnosis increasingly relies on Machine Learning (ML) models. This is a task that is often hampered by severely imbalanced datasets, where positive cases can be quite rare. Their use is further compromised by their limited interpretability, which is becoming increasingly important. While post-hoc interpretability techniques such as SHAP and LIME have been used with some success on so-called black box models, the use of inherently understandable models makes such endeavors more fruitful. This paper addresses these issues by demonstrating how a relatively new synthetic data generation technique, STEM, can be used to produce data to train models produced by Grammatical Evolution (GE) that are inherently understandable. STEM is a recently introduced combination of the Synthetic Minority Oversampling Technique (SMOTE), Edited Nearest Neighbour (ENN), and Mixup; it has previously been successfully used to tackle both between class and within class imbalance issues. We test our technique on the Digital Database for Screening Mammography (DDSM) and the Wisconsin Breast Cancer (WBC) datasets and compare Area Under the Curve (AUC) results with an ensemble of the top three performing classifiers from a set of eight standard ML classifiers with varying degrees of interpretability. We demonstrate that the GE-derived models present the best AUC while still maintaining interpretable solutions.

Interpretable Solutions for Breast Cancer Diagnosis with Grammatical Evolution and Data Augmentation

TL;DR

This work tackles the dual challenges of imbalanced data and interpretability in breast cancer diagnosis by pairing Grammatical Evolution (GE) with the STEM data-augmentation technique. Using Haralick texture features extracted from DDSM and Wisconsin Breast Cancer datasets, the authors show that GE models achieve superior AUC while remaining inherently interpretable, outperforming an ensemble of eight traditional classifiers. Bayesian signed-rank analysis further establishes STEM as the top augmentation strategy across multiple data configurations, highlighting the method's robustness. The approach holds promise for clinically actionable decision support by delivering transparent, high-performing models and guiding future improvements with feature-level insights and cross-dataset validation.

Abstract

Medical imaging diagnosis increasingly relies on Machine Learning (ML) models. This is a task that is often hampered by severely imbalanced datasets, where positive cases can be quite rare. Their use is further compromised by their limited interpretability, which is becoming increasingly important. While post-hoc interpretability techniques such as SHAP and LIME have been used with some success on so-called black box models, the use of inherently understandable models makes such endeavors more fruitful. This paper addresses these issues by demonstrating how a relatively new synthetic data generation technique, STEM, can be used to produce data to train models produced by Grammatical Evolution (GE) that are inherently understandable. STEM is a recently introduced combination of the Synthetic Minority Oversampling Technique (SMOTE), Edited Nearest Neighbour (ENN), and Mixup; it has previously been successfully used to tackle both between class and within class imbalance issues. We test our technique on the Digital Database for Screening Mammography (DDSM) and the Wisconsin Breast Cancer (WBC) datasets and compare Area Under the Curve (AUC) results with an ensemble of the top three performing classifiers from a set of eight standard ML classifiers with varying degrees of interpretability. We demonstrate that the GE-derived models present the best AUC while still maintaining interpretable solutions.
Paper Structure (13 sections, 2 equations, 4 figures, 4 tables)

This paper contains 13 sections, 2 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Methodology
  4. Experimental Details
  5. System Settings
  6. Performance Metric
  7. Class Balancing
  8. Grammatical Evolution
  9. Other Classifiers
  10. Results and Discussion
  11. Statistical Analysis
  12. Conclusion and Future Work
  13. ACKNOWLEDGEMENTS

Figures (4)

  • Figure 1: Outline of the proposed approach for breast cancer classification using GE and other classifiers.
  • Figure 2: Concentric ring chart for setup description. Rings are setups, and the coloured areas indicate training positive percent. Legend includes the training positive and negative total samples.
  • Figure 3: Boxplot analysis comparing opponent approaches and their AUC distributions across multiple runs
  • Figure 4: The illustration of the overall results acquired from the Bayesian signed-rank test is shown here. The cumulative score is the total number of times one approach outperforms the other. STEM obtained a cumulative score of 23 where the maximum possible is 40 (comparing one versus another 8 approaches in 5 setups), outperforming the other approaches. Each color represents distinct test setups used for the evaluation.