Luis Oala

Healthcare systems are currently adapting to digital technologies, producing large quantities of novel data. Based on these data, machine-learning algorithms have been developed to support practitioners in labor-intensive workflows such as diagnosis, prognosis, triage or treatment of disease. However, their translation into medical practice is often hampered by a lack of careful evaluation in different settings. Efforts have started worldwide to establish guidelines for evaluating machine learning for health (ML4H) tools, highlighting the necessity to evaluate models for bias, interpretability, robustness, and possible failure modes. However, testing and adopting these guidelines in practice remains an open challenge. In this work, we target the paper-to-practice gap by applying an ML4H audit framework proposed by the ITU/WHO Focus Group on Artificial Intelligence for Health (FG-AI4H) to three use cases: diagnostic prediction of diabetic retinopathy, diagnostic prediction of Alzheimer’s disease, and cytomorphologic classification for leukemia diagnostics. The assessment comprises dimensions such as bias, interpretability, and robustness. Our results highlight the importance of fine-grained and caseadapted quality assessment, provide support for incorporating proposed quality assessment considerations of ML4H during the entire development life cycle, and suggest improvements for future ML4H reference evaluation frameworks.

The quantitative detection of failure modes is important for making deep neural networks reliable and usable at scale. We consider three examples for failure modes in image reconstruction problems and demonstrate the potential of uncertainty quantification as a fine-grained alarm system. We propose a deterministic, modular and lightweight approach, called Interval Neural Networks, that produces fast and easy to interpret uncertainty scores which improve the detection of failure modes across four out of five image reconstruction experiments.

Addressing shifts in data distributions is an important prerequisite for the deployment of deep learning models to real-world settings. A general approach to this problem involves the adjustment of models to a new domain through transfer learning. However, in many cases, this is not applicable in a post-hoc manner to deployed models and further parameter adjustments jeopardize safety certifications that were established beforehand. In such a context, we propose to deal with changes in the data distribution via guided data homogenization which shifts the burden of adaptation from the model to the data. This approach makes use of information about the training data contained implicitly in the deep learning model to learn a domain transfer function. This allows for a targeted deployment of models to unknown scenarios without changing the model itself. We demonstrate the potential of data homogenization through experiments on the CIFAR-10 and MNIST data sets.

In absence of explicit knowledge about data models it is often assumed that a semantic match between labelled and unlabelled data is desirable in the context of semi-supervised deep learning (SSDL). In this work, we demonstrate the limits of semantic data set matching. We present and make available a comprehensive simulation sandbox, called non-IID-SSDL, for stress testing SSDL algorithms under various non-IID configurations. We show that semantic data set matching can degrade the performance for state of the art SSDL. In addition, we demonstrate that simple dissimilarity measures in the feature space of a generic classifier offer a promising and more reliable matching criterion.

This work investigates the detection of instabilities that may occur when utilizing deep learning models for image reconstruction tasks. Although neural networks often empirically outperform traditional reconstruction methods, their usage for sensitive medical applications remains controversial. Indeed, in a recent series of works, it has been demonstrated that deep learning approaches are susceptible to various types of instabilities, caused for instance by adversarial noise or out-ofdistribution features. It is argued that this phenomenon can be observed regardless of the underlying architecture and that there is no easy remedy. Based on this insight, the present work demonstrates, how uncertainty quantification methods can be employed as instability detectors. In particular, it is shown that the recently proposed Interval Neural Networks are highly effective in revealing instabilities of reconstructions. Such an ability is crucial to ensure a safe use of deep learning-based methods for medical image reconstruction.

Purpose The quantitative detection of failure modes is important for making deep neural networks reliable and usable at scale. We consider three examples for common failure modes in image reconstruction and demonstrate the potential of uncertainty quantification as a fine-grained alarm system. Methods We propose a deterministic, modular and lightweight approach called Interval Neural Network (INN) that produces fast and easy to interpret uncertainty scores for deep neural networks. Importantly, INNs can be constructed post hoc for already trained prediction networks. We compare it against state-of-the-art baseline methods (MCDROP, PROBOUT). Results We demonstrate on controlled, synthetic inverse problems the capacity of INNs to capture uncertainty due to noise as well as directional error information. On a real-world inverse problem with human CT scans, we can show that INNs produce uncertainty scores which improve the detection of all considered failure modes compared to the baseline methods. Conclusion Interval Neural Networks offer a promising tool to expose weaknesses of deep image reconstruction models and ultimately make them more reliable. The fact that they can be applied post hoc to equip already trained deep neural network models with uncertainty scores makes them particularly interesting for deployment.