A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer
Tools based on deep learning models have been created in recent years to aid radiologists in the diagnosis of breast cancer from mammograms. However, the datasets used to train these models may suffer from class imbalance, i.e., there are often fewer malignant samples than benign or healthy cases, w...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-12-01
|
| Series: | Diagnostics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-4418/13/1/67 |
| _version_ | 1797626029152403456 |
|---|---|
| author | Ricky Walsh Mickael Tardy |
| author_facet | Ricky Walsh Mickael Tardy |
| author_sort | Ricky Walsh |
| collection | DOAJ |
| description | Tools based on deep learning models have been created in recent years to aid radiologists in the diagnosis of breast cancer from mammograms. However, the datasets used to train these models may suffer from class imbalance, i.e., there are often fewer malignant samples than benign or healthy cases, which can bias the model towards the healthy class. In this study, we systematically evaluate several popular techniques to deal with this class imbalance, namely, class weighting, over-sampling, and under-sampling, as well as a synthetic lesion generation approach to increase the number of malignant samples. These techniques are applied when training on three diverse Full-Field Digital Mammography datasets, and tested on in-distribution and out-of-distribution samples. The experiments show that a greater imbalance is associated with a greater bias towards the majority class, which can be counteracted by any of the standard class imbalance techniques. On the other hand, these methods provide no benefit to model performance with respect to Area Under the Curve of the Recall Operating Characteristic (AUC-ROC), and indeed under-sampling leads to a reduction of 0.066 in AUC in the case of a 19:1 benign to malignant imbalance. Our synthetic lesion methodology leads to better performance in most cases, with increases of up to 0.07 in AUC on out-of-distribution test sets over the next best experiment. |
| first_indexed | 2024-03-11T10:04:48Z |
| format | Article |
| id | doaj.art-476c91479ecb4290819a0158a53b9aee |
| institution | Directory Open Access Journal |
| issn | 2075-4418 |
| language | English |
| last_indexed | 2024-03-11T10:04:48Z |
| publishDate | 2022-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Diagnostics |
| spelling | doaj.art-476c91479ecb4290819a0158a53b9aee2023-11-16T15:08:22ZengMDPI AGDiagnostics2075-44182022-12-011316710.3390/diagnostics13010067A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast CancerRicky Walsh0Mickael Tardy1ISTIC, Campus Beaulieu, Université de Rennes 1, 35700 Rennes, FranceHera-MI SAS, 44800 Saint-Herblain, FranceTools based on deep learning models have been created in recent years to aid radiologists in the diagnosis of breast cancer from mammograms. However, the datasets used to train these models may suffer from class imbalance, i.e., there are often fewer malignant samples than benign or healthy cases, which can bias the model towards the healthy class. In this study, we systematically evaluate several popular techniques to deal with this class imbalance, namely, class weighting, over-sampling, and under-sampling, as well as a synthetic lesion generation approach to increase the number of malignant samples. These techniques are applied when training on three diverse Full-Field Digital Mammography datasets, and tested on in-distribution and out-of-distribution samples. The experiments show that a greater imbalance is associated with a greater bias towards the majority class, which can be counteracted by any of the standard class imbalance techniques. On the other hand, these methods provide no benefit to model performance with respect to Area Under the Curve of the Recall Operating Characteristic (AUC-ROC), and indeed under-sampling leads to a reduction of 0.066 in AUC in the case of a 19:1 benign to malignant imbalance. Our synthetic lesion methodology leads to better performance in most cases, with increases of up to 0.07 in AUC on out-of-distribution test sets over the next best experiment.https://www.mdpi.com/2075-4418/13/1/67mammographymedical imagingbreast cancerclass imbalancedeep learningsynthetic data |
| spellingShingle | Ricky Walsh Mickael Tardy A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer Diagnostics mammography medical imaging breast cancer class imbalance deep learning synthetic data |
| title | A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer |
| title_full | A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer |
| title_fullStr | A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer |
| title_full_unstemmed | A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer |
| title_short | A Comparison of Techniques for Class Imbalance in Deep Learning Classification of Breast Cancer |
| title_sort | comparison of techniques for class imbalance in deep learning classification of breast cancer |
| topic | mammography medical imaging breast cancer class imbalance deep learning synthetic data |
| url | https://www.mdpi.com/2075-4418/13/1/67 |
| work_keys_str_mv | AT rickywalsh acomparisonoftechniquesforclassimbalanceindeeplearningclassificationofbreastcancer AT mickaeltardy acomparisonoftechniquesforclassimbalanceindeeplearningclassificationofbreastcancer AT rickywalsh comparisonoftechniquesforclassimbalanceindeeplearningclassificationofbreastcancer AT mickaeltardy comparisonoftechniquesforclassimbalanceindeeplearningclassificationofbreastcancer |