The limited ability of Convolutional Neural Networks to generalize to images from previously unseen domains is a major limitation, in particular, for safety-critical clinical tasks such as dermoscopic skin cancer classification. In order to translate CNN-based applications into the clinic, it is essential that they are able to adapt to domain shifts. Such new conditions can arise through the use of different image acquisition systems or varying lighting conditions. In dermoscopy, shifts can also occur as a change in patient age or occurence of rare lesion localizations (e.g. palms). These are not prominently represented in most training datasets and can therefore lead to a decrease in performance. In order to verify the generalizability of classification models in real world clinical settings it is crucial to have access to data which mimics such domain shifts. To our knowledge no dermoscopic image dataset exists where such domain shifts are properly described and quantified. We therefore grouped publicly available images from ISIC archive based on their metadata (e.g. acquisition location, lesion localization, patient age) to generate meaningful domains. To verify that these domains are in fact distinct, we used multiple quantification measures to estimate the presence and intensity of domain shifts. Additionally, we analyzed the performance on these domains with and without an unsupervised domain adaptation technique. We observed that in most of our grouped domains, domain shifts in fact exist. Based on our results, we believe these datasets to be helpful for testing the generalization capabilities of dermoscopic skin cancer classifiers.
相關內容
The estimands framework outlined in ICH E9 (R1) describes the components needed to precisely define the effects to be estimated in clinical trials, which includes how post-baseline "intercurrent" events (IEs) are to be handled. In late-stage clinical trials, it is common to handle intercurrent events like "treatment discontinuation" using the treatment policy strategy and target the treatment effect on all outcomes regardless of treatment discontinuation. For continuous repeated measures, this type of effect is often estimated using all observed data before and after discontinuation using either a mixed model for repeated measures (MMRM) or multiple imputation (MI) to handle any missing data. In basic form, both of these estimation methods ignore treatment discontinuation in the analysis and therefore may be biased if there are differences in patient outcomes after treatment discontinuation compared to patients still assigned to treatment, and missing data being more common for patients who have discontinued treatment. We therefore propose and evaluate a set of MI models that can accommodate differences between outcomes before and after treatment discontinuation. The models are evaluated in the context of planning a phase 3 trial for a respiratory disease. We show that analyses ignoring treatment discontinuation can introduce substantial bias and can sometimes underestimate variability. We also show that some of the MI models proposed can successfully correct the bias but inevitably lead to increases in variance. We conclude that some of the proposed MI models are preferable to the traditional analysis ignoring treatment discontinuation, but the precise choice of MI model will likely depend on the trial design, disease of interest and amount of observed and missing data following treatment discontinuation.
White matter bundle segmentation is a cornerstone of modern tractography to study the brain's structural connectivity in domains such as neurological disorders, neurosurgery, and aging. In this study, we present FIESTA (FIbEr Segmentation in Tractography using Autoencoders), a reliable and robust, fully automated, and easily semi-automatically calibrated pipeline based on deep autoencoders that can dissect and fully populate white matter bundles. This pipeline is built upon previous works that demonstrated how autoencoders can be used successfully for streamline filtering, bundle segmentation, and streamline generation in tractography. Our proposed method improves bundle segmentation coverage by recovering hard-to-track bundles with generative sampling through the latent space seeding of the subject bundle and the atlas bundle. A latent space of streamlines is learned using autoencoder-based modeling combined with contrastive learning. Using an atlas of bundles in standard space (MNI), our proposed method segments new tractograms using the autoencoder latent distance between each tractogram streamline and its closest neighbor bundle in the atlas of bundles. Intra-subject bundle reliability is improved by recovering hard-to-track streamlines, using the autoencoder to generate new streamlines that increase the spatial coverage of each bundle while remaining anatomically correct. Results show that our method is more reliable than state-of-the-art automated virtual dissection methods such as RecoBundles, RecoBundlesX, TractSeg, White Matter Analysis and XTRACT. Our framework allows for the transition from one anatomical bundle definition to another with marginal calibration efforts. Overall, these results show that our framework improves the practicality and usability of current state-of-the-art bundle segmentation framework.
Sparsity is a highly desired feature in deep neural networks (DNNs) since it ensures numerical efficiency, improves the interpretability of models (due to the smaller number of relevant features), and robustness. In machine learning approaches based on linear models, it is well known that there exists a connecting path between the sparsest solution in terms of the $\ell^1$ norm (i.e., zero weights) and the non-regularized solution, which is called the regularization path. Very recently, there was a first attempt to extend the concept of regularization paths to DNNs by means of treating the empirical loss and sparsity ($\ell^1$ norm) as two conflicting criteria and solving the resulting multiobjective optimization problem. However, due to the non-smoothness of the $\ell^1$ norm and the high number of parameters, this approach is not very efficient from a computational perspective. To overcome this limitation, we present an algorithm that allows for the approximation of the entire Pareto front for the above-mentioned objectives in a very efficient manner. We present numerical examples using both deterministic and stochastic gradients. We furthermore demonstrate that knowledge of the regularization path allows for a well-generalizing network parametrization.
The choice to participate in a data-driven service, often made on the basis of quality of that service, influences the ability of the service to learn and improve. We study the participation and retraining dynamics that arise when both the learners and sub-populations of users are \emph{risk-reducing}, which cover a broad class of updates including gradient descent, multiplicative weights, etc. Suppose, for example, that individuals choose to spend their time amongst social media platforms proportionally to how well each platform works for them. Each platform also gathers data about its active users, which it uses to update parameters with a gradient step. For this example and for our general class of dynamics, we show that the only asymptotically stable equilibria are segmented, with sub-populations allocated to a single learner. Under mild assumptions, the utilitarian social optimum is a stable equilibrium. In contrast to previous work, which shows that repeated risk minimization can result in representation disparity and high overall loss for a single learner \citep{hashimoto2018fairness,miller2021outside}, we find that repeated myopic updates with multiple learners lead to better outcomes. We illustrate the phenomena via a simulated example initialized from real data.
Group sparsity in Machine Learning (ML) encourages simpler, more interpretable models with fewer active parameter groups. This work aims to incorporate structured group sparsity into the shared parameters of a Multi-Task Learning (MTL) framework, to develop parsimonious models that can effectively address multiple tasks with fewer parameters while maintaining comparable or superior performance to a dense model. Sparsifying the model during training helps decrease the model's memory footprint, computation requirements, and prediction time during inference. We use channel-wise l1/l2 group sparsity in the shared layers of the Convolutional Neural Network (CNN). This approach not only facilitates the elimination of extraneous groups (channels) but also imposes a penalty on the weights, thereby enhancing the learning of all tasks. We compare the outcomes of single-task and multi-task experiments under group sparsity on two publicly available MTL datasets, NYU-v2 and CelebAMask-HQ. We also investigate how changing the sparsification degree impacts both the performance of the model and the sparsity of groups.
Despite recent availability of large transcribed Kinyarwanda speech data, achieving robust speech recognition for Kinyarwanda is still challenging. In this work, we show that using self-supervised pre-training, following a simple curriculum schedule during fine-tuning and using semi-supervised learning to leverage large unlabelled speech data significantly improve speech recognition performance for Kinyarwanda. Our approach focuses on using public domain data only. A new studio-quality speech dataset is collected from a public website, then used to train a clean baseline model. The clean baseline model is then used to rank examples from a more diverse and noisy public dataset, defining a simple curriculum training schedule. Finally, we apply semi-supervised learning to label and learn from large unlabelled data in four successive generations. Our final model achieves 3.2% word error rate (WER) on the new dataset and 15.9% WER on Mozilla Common Voice benchmark, which is state-of-the-art to the best of our knowledge. Our experiments also indicate that using syllabic rather than character-based tokenization results in better speech recognition performance for Kinyarwanda.
Statistics is sometimes described as the science of reasoning under uncertainty. Statistical models provide one view of this uncertainty, but what is frequently neglected is the invisible portion of uncertainty: that assumed not to exist once a model has been fitted to some data. Systematic errors, i.e. bias, in data relative to some model and inferential goal can seriously undermine research conclusions, and qualitative and quantitative techniques have been created across several disciplines to quantify and generally appraise such potential biases. Perhaps best known are so-called risk of bias assessment instruments used to investigate the likely quality of randomised controlled trials in medical research. However, the logic of assessing the risks caused by various types of systematic error to statistical arguments applies far more widely. This logic applies even when statistical adjustment strategies for potential biases are used, as these frequently make assumptions (e.g. data missing at random) that can never be guaranteed in finite samples. Mounting concern about such situations can be seen in the increasing calls for greater consideration of biases caused by nonprobability sampling in descriptive inference (i.e. survey sampling), and the statistical generalisability of in-sample causal effect estimates in causal inference; both of which relate to the consideration of model-based and wider uncertainty when presenting research conclusions from models. Given that model-based adjustments are never perfect, we argue that qualitative risk of bias reporting frameworks for both descriptive and causal inferential arguments should be further developed and made mandatory by journals and funders. It is only through clear statements of the limits to statistical arguments that consumers of research can fully judge their value for any specific application.
We combine Kronecker products, and quantitative information flow, to give a novel formal analysis for the fine-grained verification of utility in complex privacy pipelines. The combination explains a surprising anomaly in the behaviour of utility of privacy-preserving pipelines -- that sometimes a reduction in privacy results also in a decrease in utility. We use the standard measure of utility for Bayesian analysis, introduced by Ghosh at al., to produce tractable and rigorous proofs of the fine-grained statistical behaviour leading to the anomaly. More generally, we offer the prospect of formal-analysis tools for utility that complement extant formal analyses of privacy. We demonstrate our results on a number of common privacy-preserving designs.
Large, curated datasets are required to leverage speech-based tools in healthcare. These are costly to produce, resulting in increased interest in data sharing. As speech can potentially identify speakers (i.e., voiceprints), sharing recordings raises privacy concerns. We examine the re-identification risk for speech recordings, without reference to demographic or metadata, using a state-of-the-art speaker recognition system. We demonstrate that the risk is inversely related to the number of comparisons an adversary must consider, i.e., the search space. Risk is high for a small search space but drops as the search space grows ($precision >0.85$ for $<1*10^{6}$ comparisons, $precision <0.5$ for $>3*10^{6}$ comparisons). Next, we show that the nature of a speech recording influences re-identification risk, with non-connected speech (e.g., vowel prolongation) being harder to identify. Our findings suggest that speaker recognition systems can be used to re-identify participants in specific circumstances, but in practice, the re-identification risk appears low.
Recent advances in 3D fully convolutional networks (FCN) have made it feasible to produce dense voxel-wise predictions of volumetric images. In this work, we show that a multi-class 3D FCN trained on manually labeled CT scans of several anatomical structures (ranging from the large organs to thin vessels) can achieve competitive segmentation results, while avoiding the need for handcrafting features or training class-specific models. To this end, we propose a two-stage, coarse-to-fine approach that will first use a 3D FCN to roughly define a candidate region, which will then be used as input to a second 3D FCN. This reduces the number of voxels the second FCN has to classify to ~10% and allows it to focus on more detailed segmentation of the organs and vessels. We utilize training and validation sets consisting of 331 clinical CT images and test our models on a completely unseen data collection acquired at a different hospital that includes 150 CT scans, targeting three anatomical organs (liver, spleen, and pancreas). In challenging organs such as the pancreas, our cascaded approach improves the mean Dice score from 68.5 to 82.2%, achieving the highest reported average score on this dataset. We compare with a 2D FCN method on a separate dataset of 240 CT scans with 18 classes and achieve a significantly higher performance in small organs and vessels. Furthermore, we explore fine-tuning our models to different datasets. Our experiments illustrate the promise and robustness of current 3D FCN based semantic segmentation of medical images, achieving state-of-the-art results. Our code and trained models are available for download: //github.com/holgerroth/3Dunet_abdomen_cascade.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191