arXiv subjects

Emma A. M. Stanley

Publications and source records attributed to Emma A. M. Stanley.

7 recordsLinked to original sources

arXiv2026

Memorisation bias in medical AI

Medical AI models hold immense potential to improve patient outcomes, but they are also known to unintentionally memorise individual records from their training datasets. While such memorisation has been linked to targeted privacy attacks, its consequences for clinical deployment, where patients may be assessed by a model that saw their historical data during training, remain poorly understood. Here we show that predictions on a patient's unseen future data can change significantly if a model observed that same patient's anonymised historical data during training, a phenomenon we term "memorisation bias". We demonstrate that this bias exists across diverse data modalities and model architectures, and over prolonged time spans: in some cases, memorisation bias persists on future records acquired decades after the historical records used for training. Moreover, in simulated prospective deployment, memorisation bias has asymmetric effects on the diagnostic accuracy of returning data contributors. When a patient returned with a de novo condition absent from their historical records in the training dataset, diagnostic sensitivity decreased significantly compared to an otherwise identical model not trained on their historical data. Conversely, when their health state was unchanged, both sensitivity and specificity were significantly inflated. Our findings reveal a previously uncharacterised risk in medical AI that arises when a model is deployed on patients who contributed to its training data. This exposes a shortcoming of current model development practice: the de-identification measures designed to protect patients' privacy make it difficult to identify returning contributors and exclude them from the AI-assisted interpretation of their own future data. Mitigating memorisation risks may thus require changes to current model training and deployment protocols.

cs.LG↗

arXiv2026

Subgroup performance analysis of adaptation strategies for chest X-ray foundation models

Foundation models are increasingly adapted for downstream medical imaging tasks, yet the influence of the chosen adaptation strategy on subgroup fairness remains poorly understood. We investigate how three parameter-efficient adaptation techniques, including linear heads on the raw CLS token, an MLP, and an attention-pooling module over multi-layer patch features, affect both pathology classification performance and subgroup disparities when applied to the frozen Rad-DINO chest X-ray encoder. Using MIMIC-CXR, we evaluate eight pathologies across race, sex, and imaging-view subgroups on a prevalence-preserving, demographically balanced test set, and additionally probe how strongly each adapter encodes protected attributes. We find that attention pooling achieves the strongest overall discriminative performance and encodes attributes, particularly race, most strongly, but that improved overall performance does not consistently reduce subgroup disparities. Notably, stronger attribute encoding did not correspond to larger disparities: early network layers encoded race most weakly yet produced the largest subgroup performance gaps. Exploring different attention-pooling layer combinations further revealed no consistent relationship between the layers pooled, attribute encoding strength, and subgroup fairness. Our results indicate that richer, more expressive representations can improve accuracy while leaving fairness implications task-dependent and unpredictable, which must be assessed directly and per-task rather than inferred from encoding strength or overall performance alone.

cs.CV↗

arXiv2026

Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift

Foundation models are increasingly used as image feature extractors for mammography, but their robustness under external domain shift remains unclear. We benchmark 15 foundation-model backbones across breast density, BI-RADS severity, and cancer status using a unified frozen-backbone linear-probe protocol, training on 3 source datasets and evaluating on 12 task-compatible out-of-distribution (OOD) datasets after label harmonization. Mammography-specific vision-language models (Mammo-FM and MaMA) provide the strongest mean OOD performance, but robustness is not explained by mammography exposure alone. DINOv3 remains a competitive vision-only baseline, and mammography-adapted pretraining does not consistently improve generalization. Dataset-level analysis further shows that even leading models show heterogeneous performance across datasets. Feature-space inspection reveals that useful representations can preserve clinical signal while retaining dataset and acquisition structure. These findings highlight dataset-level OOD evaluation as a central criterion for assessing mammography representations. Our code is publicly available: https://github.com/biomedia-mira/mammo-ood.

cs.CV↗

arXiv2026

A Neuroimaging Simulation Framework for Developing and Evaluating Causal AI

Causally linking disease-related factors to image-derived biomarkers provides a powerful pathway to understanding disease mechanisms. Despite growing interest in applying causal artificial intelligence (AI) approaches for this task, these methods still need to be adapted for complex medical images, and especially, neuroimaging. However, the lack of ground-truth data presents a barrier to development. To bridge this gap, we developed and tested a method for generating synthetic neuroimages, which adhere to a user-specified causal structure describing the non-image to image variable relationships, permitting the creation of ground-truth neuroimaging datasets. In the simulated T1-weighted magnetic resonance images, anatomical variability is modeled by sampling from a subspace estimated from real data and deforming a template image to create unique simulated subjects. Causal relationships are encoded via precise volumetric changes of any region-of-interest without unwanted global artifacts. We achieved relative volume errors of 0.3-2.66% for the targeted regions-of-interest and demonstrate their statistically significant causal relationships, while maintaining mean absolute errors for non-target brain regions between 0.034-0.397ml. An initial evaluation of causal discovery methods exposes their limited ability to suppress spurious connections, highlighting the need for image-appropriate methods. Our framework is the first to enable the generation of realistic synthetic 3D neuroimages with explicit causal control that can serve as the missing ground-truth data necessary for the objective benchmarking and development of causal AI methods.

eess.IV↗

arXiv2026

Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers

We introduce the first generative foundation model for chest radiograph synthesis trained from scratch at the billion-parameter scale. Existing radiographic AI models often suffer from poor generalisation across patient subpopulations, institutions, and acquisition settings, resulting in limited real-world clinical utility. Controlled, high-fidelity synthesis of chest radiographs is a promising path toward diversifying clinical datasets and evaluating the robustness of diagnostic models. Therefore, we present the largest specialist generative foundation model for chest radiographs to date, with over 1.3B parameters, trained for 1.6T tokens on a curated, heterogeneous dataset comprising 1.2M radiographs and clinical expert-guided metadata. Our model supports controllable radiograph generation and editing across multiple demographic subgroups, acquisition views, and a dozen pathologies. Moreover, we significantly advance the state of the art in radiograph synthesis fidelity, producing images that are indistinguishable from real radiographs to clinical experts.

cs.CV↗

arXiv2025

Exploring the interplay of label bias with subgroup size and separability: A case study in mammographic density classification

Systematic mislabelling affecting specific subgroups (i.e., label bias) in medical imaging datasets represents an understudied issue concerning the fairness of medical AI systems. In this work, we investigated how size and separability of subgroups affected by label bias influence the learned features and performance of a deep learning model. Therefore, we trained deep learning models for binary tissue density classification using the EMory BrEast imaging Dataset (EMBED), where label bias affected separable subgroups (based on imaging manufacturer) or non-separable "pseudo-subgroups". We found that simulated subgroup label bias led to prominent shifts in the learned feature representations of the models. Importantly, these shifts within the feature space were dependent on both the relative size and the separability of the subgroup affected by label bias. We also observed notable differences in subgroup performance depending on whether a validation set with clean labels was used to define the classification threshold for the model. For instance, with label bias affecting the majority separable subgroup, the true positive rate for that subgroup fell from 0.898, when the validation set had clean labels, to 0.518, when the validation set had biased labels. Our work represents a key contribution toward understanding the consequences of label bias on subgroup fairness in medical imaging AI.

cs.CV↗

arXiv2024

Towards objective and systematic evaluation of bias in artificial intelligence for medical imaging

Artificial intelligence (AI) models trained using medical images for clinical tasks often exhibit bias in the form of disparities in performance between subgroups. Since not all sources of biases in real-world medical imaging data are easily identifiable, it is challenging to comprehensively assess how those biases are encoded in models, and how capable bias mitigation methods are at ameliorating performance disparities. In this article, we introduce a novel analysis framework for systematically and objectively investigating the impact of biases in medical images on AI models. We developed and tested this framework for conducting controlled in silico trials to assess bias in medical imaging AI using a tool for generating synthetic magnetic resonance images with known disease effects and sources of bias. The feasibility is showcased by using three counterfactual bias scenarios to measure the impact of simulated bias effects on a convolutional neural network (CNN) classifier and the efficacy of three bias mitigation strategies. The analysis revealed that the simulated biases resulted in expected subgroup performance disparities when the CNN was trained on the synthetic datasets. Moreover, reweighing was identified as the most successful bias mitigation strategy for this setup, and we demonstrated how explainable AI methods can aid in investigating the manifestation of bias in the model using this framework. Developing fair AI models is a considerable challenge given that many and often unknown sources of biases can be present in medical imaging datasets. In this work, we present a novel methodology to objectively study the impact of biases and mitigation strategies on deep learning pipelines, which can support the development of clinical AI that is robust and responsible.

cs.CV↗