An investigation into the causes of race bias in artificial intelligence–based cine cardiac magnetic resonance segmentation
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Aims: Artificial intelligence (AI) methods are being used increasingly for the automated segmentation of cine cardiac magnetic resonance (CMR) imaging. However, these methods have been shown to be subject to race bias; i.e. they exhibit different levels of performance for different races depending on the (im)balance of the data used to train the AI model. In this paper, we investigate the source of this bias, seeking to understand its root cause(s). Methods and results: We trained AI models to perform race classification on cine CMR images and/or segmentations from White and Black subjects from the UK Biobank and found that the classification accuracy for images was higher than for segmentations. Interpretability methods showed that the models were primarily looking at non-heart regions. Cropping images tightly around the heart caused classification accuracy to drop to almost chance level. Visualizing the latent space of AI segmentation models showed that race information was encoded in the models. Training segmentation models using cropped images reduced but did not eliminate the bias. A number of possible confounders for the bias in segmentation model performance were identified for Black subjects but none for White subjects. Conclusion: Distributional differences between annotated CMR data of White and Black races, which can lead to bias in trained AI segmentation models, are predominantly image-based, not segmentation-based. Most of the differences occur in areas outside the heart, such as subcutaneous fat. These findings will be important for researchers investigating performance of AI models on different races.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it