Melanie Bernhardt

Melanie Bernhardt, Fabio De Sousa Ribeiro, Ben Glocker

Failure detection in automated image classification is a critical safeguard for clinical deployment. Detected failure cases can be referred to human assessment, ensuring patient safety in computer-aided clinical decision making. Despite its paramount importance, there is insufficient evidence about the ability of state-of-the-art confidence scoring methods to detect test-time failures of classification models in the context of medical imaging. This paper provides a reality check, establishing the performance of in-domain misclassification detection methods, benchmarking 9 widely used confidence scores on 6 medical imaging datasets with different imaging modalities, in multiclass and binary classification settings. Our experiments show that the problem of failure detection is far from being solved. We found that none of the benchmarked advanced methods proposed in the computer vision and machine learning literature can consistently outperform a simple softmax baseline, demonstrating that improved out-of-distribution detection or model calibration do not necessarily translate to improved in-domain misclassification detection. Our developed testbed facilitates future work in this important area

Ben Glocker, Charles Jones, Melanie Bernhardt et al.

It has been rightfully emphasized that the use of AI for clinical decision making could amplify health disparities. An algorithm may encode protected characteristics, and then use this information for making predictions due to undesirable correlations in the (historical) training data. It remains unclear how we can establish whether such information is actually used. Besides the scarcity of data from underserved populations, very little is known about how dataset biases manifest in predictive models and how this may result in disparate performance. This article aims to shed some light on these issues by exploring new methodology for subgroup analysis in image-based disease detection models. We utilize two publicly available chest X-ray datasets, CheXpert and MIMIC-CXR, to study performance disparities across race and biological sex in deep learning models. We explore test set resampling, transfer learning, multitask learning, and model inspection to assess the relationship between the encoding of protected characteristics and disease detection performance across subgroups. We confirm subgroup disparities in terms of shifted true and false positive rates which are partially removed after correcting for population and prevalence shifts in the test sets. We further find a previously used transfer learning method to be insufficient for establishing whether specific patient information is used for making predictions. The proposed combination of test-set resampling, multitask learning, and model inspection reveals valuable new insights about the way protected characteristics are encoded in the feature representations of deep neural networks.

Melanie Bernhardt, Daniel C. Castro, Ryutaro Tanno et al.

Imperfections in data annotation, known as label noise, are detrimental to the training of machine learning models and have an often-overlooked confounding effect on the assessment of model performance. Nevertheless, employing experts to remove label noise by fully re-annotating large datasets is infeasible in resource-constrained settings, such as healthcare. This work advocates for a data-driven approach to prioritising samples for re-annotation - which we term "active label cleaning". We propose to rank instances according to estimated label correctness and labelling difficulty of each sample, and introduce a simulation framework to evaluate relabelling efficacy. Our experiments on natural images and on a new medical imaging benchmark show that cleaning noisy labels mitigates their negative impact on model training, evaluation, and selection. Crucially, the proposed active label cleaning enables correcting labels up to 4 times more effectively than typical random selection in realistic conditions, making better use of experts' valuable time for improving dataset quality.

Shruthi Bannur, Ozan Oktay, Melanie Bernhardt et al.

Chest radiography has been a recommended procedure for patient triaging and resource management in intensive care units (ICUs) throughout the COVID-19 pandemic. The machine learning efforts to augment this workflow have been long challenged due to deficiencies in reporting, model evaluation, and failure mode analysis. To address some of those shortcomings, we model radiological features with a human-interpretable class hierarchy that aligns with the radiological decision process. Also, we propose the use of a data-driven error analysis methodology to uncover the blind spots of our model, providing further transparency on its clinical utility. For example, our experiments show that model failures highly correlate with ICU imaging conditions and with the inherent difficulty in distinguishing certain types of radiological features. Also, our hierarchical interpretation and analysis facilitates the comparison with respect to radiologists' findings and inter-variability, which in return helps us to better assess the clinical applicability of models.

Melanie Bernhardt, Valery Vishnevskiy, Richard Rau et al.

Speed-of-sound has been shown as a potential biomarker for breast cancer imaging, successfully differentiating malignant tumors from benign ones. Speed-of-sound images can be reconstructed from time-of-flight measurements from ultrasound images acquired using conventional handheld ultrasound transducers. Variational Networks (VN) have recently been shown to be a potential learning-based approach for optimizing inverse problems in image reconstruction. Despite earlier promising results, these methods however do not generalize well from simulated to acquired data, due to the domain shift. In this work, we present for the first time a VN solution for a pulse-echo SoS image reconstruction problem using diverging waves with conventional transducers and single-sided tissue access. This is made possible by incorporating simulations with varying complexity into training. We use loop unrolling of gradient descent with momentum, with an exponentially weighted loss of outputs at each unrolled iteration in order to regularize training. We learn norms as activation functions regularized to have smooth forms for robustness to input distribution variations. We evaluate reconstruction quality on ray-based and full-wave simulations as well as on tissue-mimicking phantom data, in comparison to a classical iterative (L-BFGS) optimization of this image reconstruction problem. We show that the proposed regularization techniques combined with multi-source domain training yield substantial improvements in the domain adaptation capabilities of VN, reducing median RMSE by 54% on a wave-based simulation dataset compared to the baseline VN. We also show that on data acquired from a tissue-mimicking breast phantom the proposed VN provides improved reconstruction in 12 milliseconds.