Due to the sensitive nature of diabetes-related data, preventing them from being easily shared between studies, and the wide discrepancies in their data processing pipeline, progress in the field of ...glucose prediction is hard to assess. To address this issue, we introduce GLYFE (GLYcemia Forecasting Evaluation), a benchmark of machine learning-based glucose predictive models. We present the accuracy and clinical acceptability of nine different models coming from the literature, from standard autoregressive to more complex neural network-based models. These results are obtained on two different datasets, namely UVA/Padova Type 1 Diabetes Metabolic Simulator (T1DMS) and Ohio Type-1 Diabetes Mellitus (OhioT1DM), featuring artificial and real type 1 diabetic patients respectively. By providing extensive details about the data flow as well as by providing the whole source code of the benchmarking process, we ensure the reproducibility of the results and the usability of the benchmark by the community. Those results serve as a basis of comparison for future studies. In a field where data are hard to obtain, and where the comparison of results from different studies is often irrelevant, GLYFE gives the opportunity of gathering researchers around a standardized common environment.
•In deep learning, multi-source transfer learning can alleviate the lack of data in healthcare.•Deep models learn to discriminate data with different origins while training on multiple ...sources.•Adversarial multi-source transfer learning incentivizes the learning of a more general feature representation.•Personalized glucose predictive models in diabetes are significantly improved by first training them on multiple patients.
Background and objectives: Deep learning has yet to revolutionize general practices in healthcare, despite promising results for some specific tasks. This is partly due to data being in insufficient quantities hurting the training of the models. To address this issue, data from multiple health actors or patients could be combined by capitalizing on their heterogeneity through the use of transfer learning.
Methods: To improve the quality of the transfer between multiple sources of data, we propose a multi-source adversarial transfer learning framework that enables the learning of a feature representation that is similar across the sources, and thus more general and more easily transferable. We apply this idea to glucose forecasting for diabetic people using a fully convolutional neural network. The evaluation is done by exploring various transfer scenarios with three datasets characterized by their high inter and intra variability.
Results: While transferring knowledge is beneficial in general, we show that the statistical and clinical accuracies can be further improved by using of the adversarial training methodology, surpassing the current state-of-the-art results. In particular, it shines when using data from different datasets, or when there is too little data in an intra-dataset situation. To understand the behavior of the models, we analyze the learnt feature representations and propose a new metric in this regard. Contrary to a standard transfer, the adversarial transfer does not discriminate the patients and datasets, helping the learning of a more general feature representation.
Conclusion: The adversarial training framework improves the learning of a general feature representation in a multi-source environment, enhancing the knowledge transfer to an unseen target. The proposed method can help improve the efficiency of data shared by different health actors in the training of deep models.
Finger-vein biometrics has been extensively investigated for personal verification. Despite recent advances in finger-vein verification, current solutions completely depend on domain knowledge and ...still lack the robustness to extract finger-vein features from raw images. This paper proposes a deep learning model to extract and recover vein features using limited a priori knowledge. First, based on a combination of the known state-of-the-art handcrafted finger-vein image segmentation techniques, we automatically identify two regions: a clear region with high separability between finger-vein patterns and background, and an ambiguous region with low separability between them. The first is associated with pixels on which all the above-mentioned segmentation techniques assign the same segmentation label (either foreground or background), while the second corresponds to all the remaining pixels. This scheme is used to automatically discard the ambiguous region and to label the pixels of the clear region as foreground or background. A training data set is constructed based on the patches centered on the labeled pixels. Second, a convolutional neural network (CNN) is trained on the resulting data set to predict the probability of each pixel of being foreground (i.e., vein pixel), given a patch centered on it. The CNN learns what a finger-vein pattern is by learning the difference between vein patterns and background ones. The pixels in any region of a test image can then be classified effectively. Third, we propose another new and original contribution by developing and investigating a fully convolutional network to recover missing finger-vein patterns in the segmented image. The experimental results on two public finger-vein databases show a significant improvement in terms of finger-vein verification accuracy.
Most of recent methods for action/activity recognition, usually based on static classifiers, have achieved improvements by integrating context of local interest point (IP) features such as ...spatiotemporal IPs by characterising their neighbourhood under different scales. In this study, the authors propose a new approach that explicitly models the sequential aspect of activities. First, a sliding window segmentation technique splits the video stream into overlapping short segments. Each window is characterised by a local bag of words of IPs encoded by motion information. A first-layer support vector machine provides for each window a vector of conditional class probabilities that summarises all discriminant information that is relevant for sequence recognition. The sequence of these stochastic vectors is then fed to a hidden conditional random field for inference at the sequence level. They also show how their approach can be naturally extended to the problem of conjoint segmentation and recognition of a sequence of action classes within a continuous video stream. They have tested their model on various human action and activity datasets and the obtained results compare favourably with current state of the art.
(1) Background: Diabetes mellitus (DM) is a chronic, metabolic disease characterized by elevated levels of blood glucose. Recently, some studies approached the diabetes care domain through the ...analysis of the modifications of cardiovascular system parameters. In fact, cardiovascular diseases are the first leading cause of death in diabetic subjects. Thanks to their cost effectiveness and their ease of use, electrocardiographic (ECG) and photoplethysmographic (PPG) signals have recently been used in diabetes detection, blood glucose estimation and diabetes-related complication detection. This review’s aim is to provide a detailed overview of all the published methods, from the traditional (non machine learning) to the deep learning approaches, to detect and manage diabetes using PPG and ECG signals. This review will allow researchers to compare and understand the differences, in terms of results, amount of data and complexity that each type of approach provides and requires. (2) Method: We performed a systematic review based on articles that focus on the use of ECG and PPG signals in diabetes care. The search was focused on keywords related to the topic, such as “Diabetes”, “ECG”, “PPG”, “Machine Learning”, etc. This was performed using databases, such as PubMed, Google Scholar, Semantic Scholar and IEEE Xplore. This review’s aim is to provide a detailed overview of all the published methods, from the traditional (non machine learning) to the deep learning approaches, to detect and manage diabetes using PPG and ECG signals. This review will allow researchers to compare and understand the differences, in terms of results, amount of data and complexity that each type of approach provides and requires. (3) Results: A total of 78 studies were included. The majority of the selected studies focused on blood glucose estimation (41) and diabetes detection (31). Only 7 studies focused on diabetes complications detection. We present these studies by approach: traditional, machine learning and deep learning approaches. (4) Conclusions: ECG and PPG analysis in diabetes care showed to be very promising. Clinical validation and data processing standardization need to be improved in order to employ these techniques in a clinical environment.
Finger-vein biometrics has been extensively investigated for personal authentication. One of the open issues in finger-vein verification is the lack of robustness against image-quality degradation. ...Spurious and missing features in poor-quality images may degrade the system's performance. Despite recent advances in finger-vein quality assessment, current solutions depend on domain knowledge. In this paper, we propose a deep neural network (DNN) for representation learning to predict image quality using very limited knowledge. Driven by the primary target of biometric quality assessment, i.e., verification error minimization, we assume that low-quality images are falsely rejected in a verification system. Based on this assumption, the low- and high-quality images are labeled automatically. We then train a DNN on the resulting data set to predict the image quality. To further improve the DNN's robustness, the finger-vein image is divided into various patches, on which a patch-based DNN is trained. The deepest layers associated with the patches form together a complementary and an over-complete representation. Subsequently, the quality of each patch from a testing image is estimated and the quality scores from the image patches are conjointly input to probabilistic support vector machines (P-SVM) to boost quality-assessment performance. To the best of our knowledge, this is the first proposed work of deep learning-based quality assessment, not only for finger-vein biometrics, but also for other biometrics in general. The experimental results on two public finger-vein databases show that the proposed scheme accurately identifies high- and low-quality images and significantly outperforms existing approaches in terms of the impact on equal error-rate decrease.
This Special Section of the Pattern Recognition Letters journal includes the best papers awarded at the 3rd International Conference on Pattern Recognition and Artificial Intelligence (ICPRAI) held ...in Paris, France, during 1-3 June 2022.A total of 153 papers were presented at the ICPRAI 2022 conference. At the closing ceremony, Manal Hamzaoui et al., received the best paper award for their manuscript "A Hierarchical Prototypical Network for Few-Shot Remote Sensing Scene Classification". The second best paper was awarded to Juan Olmos et al., for their paper "An Oculomotor Digital Parkinson Biomarker: From a Deep Riemannian Representation.", while the third best paper award was granted to Victor Fortier et al., for their paper "Robust Detection of Conversational Groups using a Voting Scheme and a Memory Process".The three awarded papers were invited to submit an extended manuscript to the Special Section of Pattern Recognition Letters journal. After an exhaustive review process involving two rounds, the three extended papers were accepted for publication.The paper by Hamzaoui et al. 1 proposes a hyperbolic prototypical network for few fhot remote sensing scene classification. The objective of this work was to examine the relevance of hyperbolic embeddings of Remote Sensing (RS) data, in the context of the few-shot remote sensing scene classification problem. To this end, the authors have adopted hyperbolic prototypical networks as a meta-learning approach to embed scene images along with a feature clipping technique to ensure a numerically steady model. These hyperbolic embeddings were then analyzed whether they provide a better representation than Euclidean representations and better reflect the underlying structure of scene classes. The experimental results on the NWPU-RESISC45 RS dataset have demonstrated that the hyperbolic embeddings outperform their Euclidean counterparts. This study suggests, therefore, that operating in hyperbolic spaces is an interesting alternative for the RS communityThe paper by Olmos et al. 2 proposes a Riemannian SPD learning scheme to characterize fixational oculomotor Parkinsonian abnormalities. It introduces a deep Riemannian framework to discover potential oculomotor patterns from non-invasive video analysis, with the aim at mitigating data scarcity and interpreting the latent space. A convolutional representation is first built and aggregated onto a symmetric positive definite matrix (SPD) to encode deep convolutional features’ second-order statistics. The latter are then fed to a non-linear hierarchical architecture that processes SPD data by maintaining them into their Riemannian manifold. This representation allows discriminating patients with Parkinson's disease (PD) from Healthy Controls, including at PD stages 2.5 and 3. Moreover, the proposed geometrical representation shows capabilities to statistically differentiate observations among Parkinson's stages. The developed tool also demonstrates coherent results from explainability maps back-propagated from output probabilities.The paper by Tosato et al. 3 proposes an approach for exploiting temporal information to detect conversational groups in videos and predict the next speaker. The work's first objective is to detect, in video sequences, F-formations, which are concepts describing spatial arrangements of participants during social interactions. The second objective is to predict the next speaker during a group conversation. The approach proposed by the authors harnesses time information and multimodal signals of humans extracted from video sequences, and uses the engagement level of people as a group belonging feature. It relies on a Long Short Term Memory neural network to predict who will take the speaker's turn in a conversation group. When tested on the MatchNMingle dataset, this scheme achieves an accuracy of 85% true positives in group detection and 98 % in predicting the next speaker.We want to thank the authors of the papers for submitting their work to this Special Section. We also want to thank the reviewers for their efforts in providing critical and constructive feedback. Finally, we thank Maria de Marsico and Mohammed Samiullah for supporting this Special Section.
•We propose a paradigm unveiling handwriting changes due to aging and Alzheimer‘s.•Our new semi-supervised learning and sequential representation learning are key.•Semi-supervised learning brings to ...light handwriting multimodal behavioral trends.•Our sequential representation learning uncovers temporal feature representations.•Classification based on temporal representations outperforms the state of the art.
We present, in this paper, a novel paradigm for assessing Alzheimer’s disease and aging by analyzing impairment of handwriting (HW) on tablets, a challenging problem that is still in its infancy. The state of the art is dominated by methods that assume a unique behavioral trend for each cognitive profile or age group, and that extract global kinematic parameters, assessed by standard statistical tests or classification models, for discriminating the neuropathological disorders (Alzheimer’s (AD), Mild Cognitive Impairment (MCI)) from Healthy Controls (HC), or HC age groups from each other. Our work tackles these two major limitations as follows. First, instead of considering a unique behavioral pattern for each cognitive profile or age group, we relax this heavy constraint by allowing the emergence of multimodal behavioral patterns. We achieve this by performing semi or unsupervised learning to uncover homogeneous clusters of subjects, and then we analyze how much information these clusters carry on the cognitive profiles (or age groups). Second, instead of relying on global kinematic parameters, mostly consisting of their average, we refine the encoding either by a semi-global parameterization, or by modeling the full dynamics of each parameter, harnessing thereby the rich temporal information inherently characterizing online HW. To illustrate the power of our paradigm, we present three studies, one regarding age, and two regarding Alzheimer’s. Thanks to our modeling, we obtain new findings that are the first of their kind on this research field. On aging, unlike previous works reporting only one pattern of HW change with age, our study, based on a semiglobal parametrization scheme, uncovers three major aging HW styles, one specific to aged subjects and two shared with other age groups. On Alzheimer’s, a striking finding is revealed: two major clusters are unveiled, one dominated by HC and MCI subjects, and one by MCI and ES-AD, thus revealing that MCI patients have fine motor skills leaning towards either HC’s or ES-AD’s. Our paper introduces also a new temporal representation learning from HW trajectories that uncovers a rich set of features simultaneously like the full velocity profile, size and slant, fluidity, and shakiness, and reveals, in a naturally explainable way, how these HW features conjointly characterize, with fine and subtle details, the cognitive profiles.
Building upon the recent advances and successes in the application of deep learning to the medical field, we propose in this work a new approach to detect and classify early-stage Alzheimer patients ...using online handwriting (HW) loop patterns. To cope with the lack of training data prevalent in the tasks of classification of neuro-degenerative diseases from behavioral data, we investigate several data augmentation techniques. In this respect, compared to the traditional data augmentation techniques proposed for HW-based Parkinson detection, we investigate a variant of Generative Adversarial Networks (GANs), DoppelGANger, especially tailored for times series and hence suitable for synthesizing realistic online handwriting sequences. Based on a 1D-Convolutional Neural Network (1D-CNN) to perform Alzheimer classification, we show, on a real dataset related to HW and Alzheimer, that our DoppelGANger-based augmentation model allow the CNN to significantly outperform both the current state of the art and the other data augmentation techniques.
Photoplethysmography (PPG) is a non-invasive and cost-efficient optical technique used to assess blood volume variations in the microcirculation. PPG technology is widely used in a variety of ...wearable sensors to investigate the cardiovascular system. Recent studies have demonstrated the utility of PPG analysis for carrying out large-scale screening to prevent and detect diabetes. However, most of these studies require feature extraction and/or several pre-processing steps. Over the past few years, the advent of deep learning has significantly impacted the analysis of biomedical signals. Despite their success in other fields, however, very few studies have focused on the application of deep learning to raw PPG signals for detecting diabetes. Existing studies have proposed large models trained on large amounts of data. In this paper, we present a Light CNN-based model for screening the presence of type 2 diabetes using a single raw pulse extracted from photoplethysmographic signals. In addition to the baseline architecture, we evaluate different model architectures that take as input age and biological sex or PPG handcrafted features. Furthermore, we apply transfer learning to all the tested architectures to evaluate the effectiveness of harnessing pre-trained models in detecting diabetes. We tested a model pre-trained on a general PPG shape dataset and another model pre-trained on a dataset containing hypertension PPG signals. Our model scored an AUC of 75.5 when trained with raw PPG waves, age, and biological sex without applying transfer learning, which is competitive with current state of the art.