The inverse problem of supervised reconstruction of depth-variable (time-dependent) parameters in ordinary differential equations is considered, with the typical application of finding weights of a neural ordinary differential equation (NODE) for a residual network with time continuous layers. The differential equation is treated as an abstract and isolated entity, termed a standalone NODE (sNODE), to facilitate for a wide range of applications. The proposed parameter reconstruction is performed by minimizing a cost functional covering a variety of loss functions and penalty terms. Regularization via penalty terms is incorporated to enhance ethical and trustworthy AI formulations. A nonlinear conjugate gradient mini-batch optimization scheme (NCG) is derived for the training having the benefit of including a sensitivity problem. The model (differential equation)-based approach is thus combined with a data-driven learning procedure. Mathematical properties are stated for the differential equation and the cost functional. The adjoint problem needed is derived together with the sensitivity problem. The sensitivity problem itself can estimate changes in the output under perturbation of the trained parameters. To preserve smoothness during the iterations, the Sobolev gradient is calculated and incorporated. Numerical results are included to validate the procedure for a NODE and synthetic datasets and compared with standard gradient approaches. For stability, using the sensitivity problem, a strategy for adversarial attacks is constructed, and it is shown that the given method with Sobolev gradients is more robust than standard approaches for parameter identification.

Detecting novelties given unlabeled examples of normal data is a challenging task in machine learning, particularly when the novel and normal categories are semantically close. Large deep models pretrained on massive datasets can provide a rich representation space in which the simple k-nearest neighbor distance works as a novelty measure. However, as we show in this paper, the basic k-NN method might be insufficient in this context due to ignoring the 'local geometry' of the distribution over representations as well as the impact of irrelevant 'background features'. To address this, we propose a fully unsupervised novelty detection approach that integrates the flexibility of k-NN with a locally adapted scaling of dimensions based on the 'neighbors of nearest neighbor' and computing a 'likelihood ratio' in pretrained (self-supervised) representation spaces. Our experiments with image data show the advantage of this method when off-the-shelf vision transformers (e.g., pretrained by DINO) are used as the feature extractor without any fine-tuning.

Training a deep neural network is computationally expensive, but achieving the same network performance with less computation is possible if the training data is carefully chosen. However, selecting input samples during training is challenging as their true importance for the optimization is unknown. Furthermore, evaluation of the importance of individual samples must be computationally efficient and unbiased. In this paper, we present a new input data importance sampling strategy for reducing the training time of deep neural networks. We investigate different importance metrics that can be efficiently retrieved as they are available during training, i.e., the training loss and gradient norm. We found that choosing only samples with large loss or gradient norm, which are hard for the network to learn, is not optimal for the network performance. Instead, we introduce an importance sampling strategy that selects samples based on the cumulative distribution function of the loss and gradient norm, thereby making it more likely to choose hard samples while still including easy ones. The behavior of the proposed strategy is first analyzed on a synthetic dataset, and then evaluated in the application of classification of malignant cancer in digital pathology image patches. As pathology images contain many repetitive patterns, there could be significant gains in focusing on features that contribute stronger to the optimization. Finally, we show how the importance sampling process can be used to gain insights about the input data through visualization of samples that are found most or least useful for the training.

Multiple-instance learning (MIL) is an attractive approach for digital pathology applications as it reduces the costs related to data collection and labelling. However, it is not clear how sensitive MIL is to clinically realistic domain shifts, i.e., differences in data distribution that could negatively affect performance, and if already existing metrics for detecting domain shifts work well with these algorithms. We trained an attention-based MIL algorithm to classify whether a whole-slide image of a lymph node contains breast tumour metastases. The algorithm was evaluated on data from a hospital in a different country and various subsets of this data that correspond to different levels of domain shift. Our contributions include showing that MIL for digital pathology is affected by clinically realistic differences in data, evaluating which features from a MIL model are most suitable for detecting changes in performance, and proposing an unsupervised metric named Fréchet Domain Distance (FDD) for quantification of domain shifts. Shift measure performance was evaluated through the mean Pearson correlation to change in classification performance, where FDD achieved 0.70 on 10-fold cross-validation models. The baselines included Deep ensemble, Difference of Confidence, and Representation shift which resulted in 0.45, -0.29, and 0.56 mean Pearson correlation, respectively. FDD could be a valuable tool for care providers and vendors who need to verify if a MIL system is likely to perform reliably when implemented at a new site, without requiring any additional annotations from pathologists.

As the problem of reconstructing high dynamic range (HDR) imagesfrom a single exposure has attracted much research effort, it isessential to provide a robust protocol and clear guidelines on howto evaluate and compare new methods. In this work, we comparedsix recent single image HDR reconstruction (SI-HDR) methodsin a subjective image quality experiment on an HDR display. Wefound that only two methods produced results that are, on average,more preferred than the unprocessed single exposure images. Whenthe same methods are evaluated using image quality metrics, astypically done in papers, the metric predictions correlate poorlywith subjective quality scores. The main reason is a significant toneand color difference between the reference and reconstructed HDRimages. To improve the predictions of image quality metrics, we propose correcting for the inaccuracies of the estimated cameraresponse curve before computing quality values. We further analyzethe sources of prediction noise when evaluating SI-HDR methodsand demonstrate that existing metrics can reliably predict onlylarge quality differences.

Unsupervised learning has made substantial progress over the last few years, especially by means of contrastive self-supervised learning. The dominating dataset for benchmarking self-supervised learning has been ImageNet, for which recent methods are approaching the performance achieved by fully supervised training. The ImageNet dataset is however largely object-centric, and it is not clear yet what potential those methods have on widely different datasets and tasks that are not object-centric, such as in digital pathology.While self-supervised learning has started to be explored within this area with encouraging results, there is reason to look closer at how this setting differs from natural images and ImageNet. In this paper we make an in-depth analysis of contrastive learning for histopathology, pin-pointing how the contrastive objective will behave differently due to the characteristics of histopathology data. Using SimCLR and H&E stained images as a representative setting for contrastive self-supervised learning in histopathology, we bring forward a number of considerations, such as view generation for the contrastive objectiveand hyper-parameter tuning. In a large battery of experiments, we analyze how the downstream performance in tissue classification will be affected by these considerations. The results point to how contrastive learning can reduce the annotation effort within digital pathology, but that the specific dataset characteristics need to be considered. To take full advantage of the contrastive learning objective, different calibrations of view generation and hyper-parameters are required. Our results pave the way for realizing the full potential of self-supervised learning for histopathology applications. Code and trained models are available at https://github.com/k-stacke/ssl-pathology.

Insufficient training data is a major bottleneck for most deep learning practices, not least in medical imaging where data is difficult to collect and publicly available datasets are scarce due to ethics and privacy. This work investigates the use of synthetic images, created by generative adversarial networks (GANs), as the only source of training data. We demonstrate that for this application, it is of great importance to make use of multiple GANs to improve the diversity of the generated data, i.e. to sufficiently cover the data distribution. While a single GAN can generate seemingly diverse image content, training on this data in most cases lead to severe over-fitting. We test the impact of ensembled GANs on synthetic 2D data as well as common image datasets (SVHN and CIFAR-10), and using both DCGANs and progressively growing GANs. As a specific use case, we focus on synthesizing digital pathology patches to provide anonymized training data.

This paper presents an evaluation of how data augmentation and inter-class transformations can be used to synthesize training data in low-data scenarios for single-image weather classification. In such scenarios, augmentations is a critical component, but there is a limit to how much improvements can be gained using classical augmentation strategies. Generative adversarial networks (GAN) have been demonstrated to generate impressive results, and have also been successful as a tool for data augmentation, but mostly for images of limited diversity, such as in medical applications. We investigate the possibilities in using generative augmentations for balancing a small weather classification dataset, where one class has a reduced number of images. We compare intra-class augmentations by means of classical transformations as well as noise-to-image GANs, to interclass augmentations where images from another class are transformed to the underrepresented class. The results show that it is possible to take advantage of GANs for inter-class augmentations to balance a small dataset for weather classification. This opens up for future work on GAN-based augmentations in scenarios where data is both diverse and scarce.

The high capacity of neural networks allows fitting models to data with high precision, but makes generalization to unseen data a challenge. If a domain shift exists, i.e. differences in image statistics between training and test data, care needs to be taken to ensure reliable deployment in real-world scenarios. In digital pathology, domain shift can be manifested in differences between whole-slide images, introduced by for example differences in acquisition pipeline - between medical centers or over time. In order to harness the great potential presented by deep learning in histopathology, and ensure consistent model behavior, we need a deeper understanding of domain shift and its consequences, such that a model's predictions on new data can be trusted. This work focuses on the internal representation learned by trained convolutional neural networks, and shows how this can be used to formulate a novel measure - the representation shift - for quantifying the magnitude of model specific domain shift. We perform a study on domain shift in tumor classification of hematoxylin and eosin stained images, by considering different datasets, models, and techniques for preparing data in order to reduce the domain shift. The results show how the proposed measure has a high correlation with drop in performance when testing a model across a large number of different types of domain shifts, and how it improves on existing techniques for measuring data shift and uncertainty. The proposed measure can reveal how sensitive a model is to domain variations, and can be used to detect new data that a model will have problems generalizing to. We see techniques for measuring, understanding and overcoming the domain shift as a crucial step towards reliable use of deep learning in the future clinical pathology applications.

Machine learning (ML) algorithms are optimized for the distribution represented by the training data. For outlier data, they often deliver predictions with equal confidence, even though these should not be trusted. In order to deploy ML-based digital pathology solutions in clinical practice, effective methods for detecting anomalous data are crucial to avoid incorrect decisions in the outlier scenario. We propose a new unsupervised learning approach for anomaly detection in histopathology data based on generative adversarial networks (GANs). Compared to the existing GAN-based methods that have been used in medical imaging, the proposed approach improves significantly on performance for pathology data. Our results indicate that histopathology imagery is substantially more complex than the data targeted by the previous methods. This complexity requires not only a more advanced GAN architecture but also an appropriate anomaly metric to capture the quality of the reconstructed images.