Details
| Original language | English |
|---|---|
| Pages (from-to) | 658-661 |
| Number of pages | 4 |
| Journal | Current Directions in Biomedical Engineering |
| Volume | 9 |
| Issue number | 1 |
| Publication status | Published - 1 Sept 2023 |
Abstract
The AUC margin loss is a valuable loss function for medical image classification as it addresses the problems of imbalanced and noisy labels. It is used by the current winner of the CheXpert competition. The CheXpert dataset is a large dataset (200k+ images), however datasets in the range of 1k-10k medical datasets are much more common. This raises the question if optimizing AUC margin loss also is effective in scenarios with limited data.We compare AUC margin loss optimization to binary cross-entropy on limited, imbalanced and noisy CheXpert5000, a subset of CheXpert dataset. We show that AUC margin loss is beneficial for limited data and considerably improves accuracy in the presence of label noise. It also improves out-of-box calibration.
Keywords
- CheXpert, Computer-Aided-Diagnosis, Deep Learning, Label Imbalance, Noisy Labels
ASJC Scopus subject areas
- Engineering(all)
- Biomedical Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: Current Directions in Biomedical Engineering, Vol. 9, No. 1, 01.09.2023, p. 658-661.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - AUC margin loss for limited, imbalanced and noisy medical image diagnosis
T2 - a case study on CheXpert5000
AU - Ihler, Sontje
AU - Kuhnke, Felix
PY - 2023/9/1
Y1 - 2023/9/1
N2 - The AUC margin loss is a valuable loss function for medical image classification as it addresses the problems of imbalanced and noisy labels. It is used by the current winner of the CheXpert competition. The CheXpert dataset is a large dataset (200k+ images), however datasets in the range of 1k-10k medical datasets are much more common. This raises the question if optimizing AUC margin loss also is effective in scenarios with limited data.We compare AUC margin loss optimization to binary cross-entropy on limited, imbalanced and noisy CheXpert5000, a subset of CheXpert dataset. We show that AUC margin loss is beneficial for limited data and considerably improves accuracy in the presence of label noise. It also improves out-of-box calibration.
AB - The AUC margin loss is a valuable loss function for medical image classification as it addresses the problems of imbalanced and noisy labels. It is used by the current winner of the CheXpert competition. The CheXpert dataset is a large dataset (200k+ images), however datasets in the range of 1k-10k medical datasets are much more common. This raises the question if optimizing AUC margin loss also is effective in scenarios with limited data.We compare AUC margin loss optimization to binary cross-entropy on limited, imbalanced and noisy CheXpert5000, a subset of CheXpert dataset. We show that AUC margin loss is beneficial for limited data and considerably improves accuracy in the presence of label noise. It also improves out-of-box calibration.
KW - CheXpert
KW - Computer-Aided-Diagnosis
KW - Deep Learning
KW - Label Imbalance
KW - Noisy Labels
UR - http://www.scopus.com/inward/record.url?scp=85173232202&partnerID=8YFLogxK
U2 - 10.1515/cdbme-2023-1165
DO - 10.1515/cdbme-2023-1165
M3 - Article
AN - SCOPUS:85173232202
VL - 9
SP - 658
EP - 661
JO - Current Directions in Biomedical Engineering
JF - Current Directions in Biomedical Engineering
IS - 1
ER -