There are 4 groups defined to the tuples (y, s).

Machine learning models have achieved high overall accuracy in medical image analysis. However, performance disparities on specific patient groups pose challenges to their clinical utility, safety, and fairness. This can affect known patient groups - such as those based on sex, age, or disease subtype - as well as previously unknown and unlabeled groups. Furthermore, the root cause of such observed performance disparities is often challenging to uncover, hindering mitigation efforts. In this paper, to address these issues, we leverage Slice Discovery Methods (SDMs) to identify interpretable underperforming subsets of data and formulate hypotheses regarding the cause of observed performance disparities. We introduce a novel SDM and apply it in a case study on the classification of pneumothorax and atelectasis from chest x-rays. Our study demonstrates the effectiveness of SDMs in hypothesis formulation and yields an explanation of previously observed but unexplained performance disparities between male and female patients in widely used chest X-ray datasets and models. Our findings indicate shortcut learning in both classification tasks, through the presence of chest drains and ECG wires, respectively. Sex-based differences in the prevalence of these shortcut features appear to cause the observed classification performance gap, representing a previously underappreciated interaction between shortcut learning and model fairness analyses.

Deep learning could be prone to learning shortcuts raised by dataset bias and result in inaccurate, unreliable, and unfair models, which impedes its adoption in real-world clinical applications. Despite its significance, there is a dearth of research in the medical image classification domain to address dataset bias. Furthermore, the bias labels are often agnostic, as identifying biases can be laborious and depend on post-hoc interpretation. This paper proposes learning Adaptive Agreement from a Biased Council (Ada-ABC), a debiasing framework that does not rely on explicit bias labels to tackle dataset bias in medical images. Ada-ABC develops a biased council consisting of multiple classifiers optimized with generalized cross entropy loss to learn the dataset bias. A debiasing model is then simultaneously trained under the guidance of the biased council. Specifically, the debiasing model is required to learn adaptive agreement with the biased council by agreeing on the correctly predicted samples and disagreeing on the wrongly predicted samples by the biased council. In this way, the debiasing model could learn the target attribute on the samples without spurious correlations while also avoiding ignoring the rich information in samples with spurious correlations. We theoretically demonstrated that the debiasing model could learn the target features when the biased model successfully captures dataset bias. Moreover, to our best knowledge, we constructed the first medical debiasing benchmark from four datasets containing seven different bias scenarios. Our extensive experiments practically showed that our proposed Ada-ABC outperformed competitive approaches, verifying its effectiveness in mitigating dataset bias for medical image classification. The codes and organized benchmark datasets will be made publicly available.

Biomedical imaging datasets are often small and biased, meaning that real-world performance of predictive models can be substantially lower than expected from internal testing. This work proposes using generative image editing to simulate dataset shifts and diagnose failure modes of biomedical vision models; this can be used in advance of deployment to assess readiness, potentially reducing cost and patient harm. Existing editing methods can produce undesirable changes, with spurious correlations learned due to the co-occurrence of disease and treatment interventions, limiting practical applicability. To address this, we train a text-to-image diffusion model on multiple chest X-ray datasets and introduce a new editing method RadEdit that uses multiple masks, if present, to constrain changes and ensure consistency in the edited images. We consider three types of dataset shifts: acquisition shift, manifestation shift, and population shift, and demonstrate that our approach can diagnose failures and quantify model robustness without additional data collection, complementing more qualitative tools for explainable AI.

Deep Neural Networks (DNNs) are prone to learning spurious features that correlate with the label during training but are irrelevant to the learning problem. This hurts model generalization and poses problems when deploying them in safety-critical applications. This paper aims to better understand the effects of spurious features through the lens of the learning dynamics of the internal neurons during the training process. We make the following observations: (1) While previous works highlight the harmful effects of spurious features on the generalization ability of DNNs, we emphasize that not all spurious features are harmful. Spurious features can be "benign" or "harmful" depending on whether they are "harder" or "easier" to learn than the core features for a given model. This definition is model and dataset-dependent. (2) We build upon this premise and use instance difficulty methods (like Prediction Depth (Baldock et al., 2021)) to quantify "easiness" for a given model and to identify this behavior during the training phase. (3) We empirically show that the harmful spurious features can be detected by observing the learning dynamics of the DNN's early layers. In other words, easy features learned by the initial layers of a DNN early during the training can (potentially) hurt model generalization. We verify our claims on medical and vision datasets, both simulated and real, and justify the empirical success of our hypothesis by showing the theoretical connections between Prediction Depth and information-theoretic concepts like V-usable information (Ethayarajh et al., 2021). Lastly, our experiments show that monitoring only accuracy during training (as is common in machine learning pipelines) is insufficient to detect spurious features. We, therefore, highlight the need for monitoring early training dynamics using suitable instance difficulty metrics.

Machine learning models for medical image analysis often suffer from poor performance on important subsets of a population that are not identified during training or testing. For example, overall performance of a cancer detection model may be high, but the model still consistently misses a rare but aggressive cancer subtype. We refer to this problem as hidden stratification, and observe that it results from incompletely describing the meaningful variation in a dataset. While hidden stratification can substantially reduce the clinical efficacy of machine learning models, its effects remain difficult to measure. In this work, we assess the utility of several possible techniques for measuring and describing hidden stratification effects, and characterize these effects both on multiple medical imaging datasets and via synthetic experiments on the well-characterised CIFAR-100 benchmark dataset. We find evidence that hidden stratification can occur in unidentified imaging subsets with low prevalence, low label quality, subtle distinguishing features, or spurious correlates, and that it can result in relative performance differences of over 20% on clinically important subsets. Finally, we explore the clinical implications of our findings, and suggest that evaluation of hidden stratification should be a critical component of any machine learning deployment in medical imaging.