The effect of dataset confounding on predictions of deep neural networks for medical imaging

in Vol. 3 (2022): Proceedings of the Northern Lights Deep Learning Workshop 2022 (2022, April 18)

The use of Convolutional Neural Networks (CNN) in medical imaging has often outperformed previous solutions and even specialists, becoming a promising technology for Computer-aided-Diagnosis (CAD) systems ... [more ▼]

The use of Convolutional Neural Networks (CNN) in medical imaging has often outperformed previous solutions and even specialists, becoming a promising technology for Computer-aided-Diagnosis (CAD) systems. However, recent works suggested that CNN may have poor generalisation on new data, for instance, generated in different hospitals. Uncontrolled confounders have been proposed as a common reason. In this paper, we experimentally demonstrate the impact of confounding data in unknown scenarios. We assessed the effect of four confounding configurations: total, strong, light and balanced. We found the confounding effect is especially prominent in total confounder scenarios, while the effect on light and strong confounding scenarios may depend on the dataset robustness. Our findings indicate that the confounding effect is independent of the architecture employed. These findings might explain why models can report good metrics during the development stage but fail to translate to real-world settings. We highlight the need for thorough consideration of these commonly unattended aspects, to develop safer CNN-based CAD systems. [less ▲]

Public Covid-19 X-ray datasets and their impact on model bias - a systematic review of a significant problem

in Medical Image Analysis (2021), 74

Computer-aided diagnosis and stratification of COVID-19 based on chest X-ray suffers from weak bias assessment and limited quality-control. Undetected bias induced by inappropriate use of datasets, and ... [more ▼]

Computer-aided diagnosis and stratification of COVID-19 based on chest X-ray suffers from weak bias assessment and limited quality-control. Undetected bias induced by inappropriate use of datasets, and improper consideration of confounders prevents the translation of prediction models into clinical practice. By adopting established tools for model evaluation to the task of evaluating datasets, this study provides a systematic appraisal of publicly available COVID-19 chest X-ray datasets, determining their potential use and evaluating potential sources of bias. Only 9 out of more than a hundred identified datasets met at least the criteria for proper assessment of the risk of bias and could be analysed in detail. Remarkably most of the datasets utilised in 201 papers published in peer-reviewed journals, are not among these 9 datasets, thus leading to models with a high risk of bias. This raises concerns about the suitability of such models for clinical use. This systematic review highlights the limited description of datasets employed for modelling and aids researchers to select the most suitable datasets for their task. [less ▲]

On the Composition and Limitations of Publicly Available COVID-19 X-Ray Imaging Datasets

E-print/Working paper (2020)

 Machine learning based methods for diagnosis and progression prediction of COVID-19 from imaging data have gained significant attention in the last months, in particular by the use of deep learning ... [more ▼]

 Machine learning based methods for diagnosis and progression prediction of COVID-19 from imaging data have gained significant attention in the last months, in particular by the use of deep learning models. In this context hundreds of models where proposed with the majority of them trained on public datasets. Data scarcity, mismatch between training and target population, group imbalance, and lack of documentation are important sources of bias, hindering the applicability of these models to real-world clinical practice. Considering that datasets are an essential part of model building and evaluation, a deeper understanding of the current landscape is needed. This paper presents an overview of the currently public available COVID-19 chest X-ray datasets. Each dataset is briefly described and potential strength, limitations and interactions between datasets are identified. In particular, some key properties of current datasets that could be potential sources of bias, impairing models trained on them are pointed out. These descriptions are useful for model building on those datasets, to choose the best dataset according the model goal, to take into account the specific limitations to avoid reporting overconfident benchmark results, and to discuss their impact on the generalisation capabilities in a specific clinical setting. [less ▲]