Explore the vast range of titles available.

Endoscopy is a major tool for assessing the physiology of inner organs. Contemporary artificial intelligence methods are used to fully automatically label medical important classes on a pixel-by-pixel level. This so-called semantic segmentation is for example used to detect cancer tissue or to assess laryngeal physiology. However, due to the diversity of patients presenting, it is necessary to judge the segmentation quality. In this study, we present a fully automatic system to evaluate the segmentation performance in laryngeal endoscopy images. We showcase on glottal area segmentation that the predicted segmentation quality represented by the intersection over union metric is on par with human raters. Using a traffic light system, we are able to identify problematic segmentation frames to allow human-in-the-loop improvements, important for the clinical adaptation of automatic analysis procedures.

Deep Learning has a large impact on medical image analysis and lately has been adopted for clinical use at the point of care. However, there is only a small number of reports of long-term studies that show the performance of deep neural networks (DNNs) in such an environment. In this study, we measured the long-term performance of a clinically optimized DNN for laryngeal glottis segmentation. We have collected the video footage for two years from an AI-powered laryngeal high-speed videoendoscopy imaging system and found that the footage image quality is stable across time. Next, we determined the DNN segmentation performance on lossy and lossless compressed data revealing that only 9% of recordings contain segmentation artifacts. We found that lossy and lossless compression is on par for glottis segmentation, however, lossless compression provides significantly superior image quality. Lastly, we employed continual learning strategies to continuously incorporate new data into the DNN to remove the aforementioned segmentation artifacts. With modest manual intervention, we were able to largely alleviate these segmentation artifacts by up to 81%. We believe that our suggested deep learning-enhanced laryngeal imaging platform consistently provides clinically sound results, and together with our proposed continual learning scheme will have a long-lasting impact on the future of laryngeal imaging.

Mobile health (mHealth) apps are software interfaces that enable users to access and manage wearable technology through smartphones and tablet devices for health improvement purposes. However, many respiratory disease mHealth apps lack transparent development documentation, compromising user confidence in their quality, functionality, and usability. This study aimed to develop and evaluate AIrway, a companion mHealth app designed to interface with an in-house wearable device for monitoring airway symptoms following established mHealth development and reporting standards. The development cycle of AIrway comprised 2 study phases. In phase 1, AIrway, a native Android app, was developed following academic and industrial standards (Android material design and Morville's design principles) and privacy regulations (Personal Information Protection and Electronic Documents Act). Core functionalities included location-based environmental monitoring, a clinical diary with action plans, Bluetooth connectivity, and real-time data storage. In phase 2, the usability of AIrway was evaluated by software app developers using standardized assessment tools, namely, the User Version of the Mobile Application Rating Scale survey and the IQVIA questionnaire. AIrway successfully fulfilled 7 of 8 development criteria on usability, privacy, security, appropriateness, transparency, safety, and technical support, with only the technology aspects requiring refinement. Accessibility assessments confirmed that AIrway's content and interface were comprehensible to the general population (grade 9-10 reading level). Technical testing demonstrated reliable Bluetooth data transmission for up to 10 minutes without interruption. User evaluation scores for the User Version of the Mobile Application Rating Scale (3.6/5.0) and IQVIA (8/11) were comparable to those of similar mHealth apps on the market. By adhering to established mHealth app design principles, AIrway achieved the necessary accessibility standards and wireless communication capabilities for wearable device integration. Future development will focus on expanding cross-platform compatibility and conducting usability evaluation with intended patient populations to validate its clinical effectiveness and support ongoing improvements.

Mobile health wearables are often embedded with small processors for signal acquisition and analysis. These embedded wearable systems are, however, limited with low available memory and computational power. Advances in machine learning, especially deep neural networks (DNNs), have been adopted for efficient and intelligent applications to overcome constrained computational environments. Herein, evolutionary algorithms are used to find novel DNNs that are accurate in classifying airway symptoms while allowing wearable deployment. As opposed to typical microphone‐acoustic signals, mechano‐acoustic data signals, which did not contain identifiable speech information for better privacy protection, are acquired from laboratory‐generated and publicly available datasets. The optimized DNNs had a low model file size of less than 150 kB and predicted airway symptoms of interest with 81.49% accuracy on unseen data. By performing explainable AI techniques, namely occlusion experiments and class activation maps, mel‐frequency bands up to 8,000 Hz are found as the most important feature for the classification. It is further found that DNN decisions are consistently relying on these specific features, fostering trust and transparency of the proposed DNNs. The proposed efficient and explainable DNN is expected to support edge computing on mechano‐acoustic sensing wearables for remote, long‐term monitoring of airway symptoms. Deep neural networks are used to classify common airway‐related symptoms from mechano‐acoustic signals. To enable the classification on wearables, evolutionary algorithms are employed to find novel neural architectures optimized for accuracy and inference speed such that they operate with constrained memory and computational resources. This new architecture adapts to new data and learns explainable representations.

Deep Learning has a large impact on medical image analysis and lately has been adopted for clinical use at the point of care. However, there is only a small number of reports of long-term studies that show the performance of deep neural networks (DNNs) in such an environment. In this study, we measured the long-term performance of a clinically optimized DNN for laryngeal glottis segmentation. We have collected the video footage for two years from an AI-powered laryngeal high-speed videoendoscopy imaging system and found that the footage image quality is stable across time. Next, we determined the DNN segmentation performance on lossy and lossless compressed data revealing that only 9% of recordings contain segmentation artifacts. We found that lossy and lossless compression is on par for glottis segmentation, however, lossless compression provides significantly superior image quality. Lastly, we employed continual learning strategies to continuously incorporate new data into the DNN to remove the aforementioned segmentation artifacts. With modest manual intervention, we were able to largely alleviate these segmentation artifacts by up to 81%. We believe that our suggested deep learning-enhanced laryngeal imaging platform consistently provides clinically sound results, and together with our proposed continual learning scheme will have a long-lasting impact on the future of laryngeal imaging.

Benchmarking plays a pivotal role in assessing and enhancing the performance of compact deep learning models designed for execution on resource-constrained devices, such as microcontrollers. Our study introduces a novel, entirely artificially generated benchmarking dataset tailored for speech recognition, representing a core challenge in the field of tiny deep learning. SpokeN-100 consists of spoken numbers from 0 to 99 spoken by 32 different speakers in four different languages, namely English, Mandarin, German and French, resulting in 12,800 audio samples. We determine auditory features and use UMAP (Uniform Manifold Approximation and Projection for Dimension Reduction) as a dimensionality reduction method to show the diversity and richness of the dataset. To highlight the use case of the dataset, we introduce two benchmark tasks: given an audio sample, classify (i) the used language and/or (ii) the spoken number. We optimized state-of-the-art deep neural networks and performed an evolutionary neural architecture search to find tiny architectures optimized for the 32-bit ARM Cortex-M4 nRF52840 microcontroller. Our results represent the first benchmark data achieved for SpokeN-100.

Endoscopy is a major tool for assessing the physiology of inner organs. Contemporary artificial intelligence methods are used to fully automatically label medical important classes on a pixel-by-pixel level. This so-called semantic segmentation is for example used to detect cancer tissue or to assess laryngeal physiology. However, due to the diversity of patients presenting, it is necessary to judge the segmentation quality. In this study, we present a fully automatic system to evaluate the segmentation performance in laryngeal endoscopy images. We showcase on glottal area segmentation that the predicted segmentation quality represented by the intersection over union metric is on par with human raters. Using a traffic light system, we are able to identify problematic segmentation frames to allow human-in-the-loop improvements, important for the clinical adaptation of automatic analysis procedures.Competing Interest StatementThe authors have declared no competing interest.

Mobile health wearables are often embedded with small processors for signal acquisition and analysis. These embedded wearable systems are, however, limited with low available memory and computational power. Advances in machine learning, especially deep neural networks (DNNs), have been adopted for efficient and intelligent applications to overcome constrained computational environments. In this study, evolutionary optimized DNNs were analyzed to classify three common airway-related symptoms, namely coughs, throat clears and dry swallows. As opposed to typical microphone-acoustic signals, mechano-acoustic data signals, which did not contain identifiable speech information for better privacy protection, were acquired from laboratory-generated and publicly available datasets. The optimized DNNs had a low footprint of less than 150 kB and predicted airway symptoms of interests with 83.7% accuracy on unseen data. By performing explainable AI techniques, namely occlusion experiments and class activation maps, mel-frequency bands up to 8,000 Hz were found as the most important feature for the classification. We further found that DNN decisions were consistently relying on these specific features, fostering trust and transparency of proposed DNNs. Our proposed efficient and explainable DNN is expected to support edge computing on mechano-acoustic sensing wearables for remote, long-term monitoring of airway symptoms. Competing Interest Statement The authors have declared no competing interest.

Studies of the global sea-level budget (SLB) and the global ocean-mass budget (OMB) are essential to assess the reliability of our knowledge of sea-level change and its contributors. Here we present datasets for times series of the SLB and OMB elements developed in the framework of ESA's Climate Change Initiative. We use these datasets to assess the SLB and the OMB simultaneously, utilising a consistent framework of uncertainty characterisation. The time series, given at monthly sampling and available at https://doi.org/10.5285/17c2ce31784048de93996275ee976fff (Horwath et al., 2021), include global mean sea-level (GMSL) anomalies from satellite altimetry, the global mean steric component from Argo drifter data with incorporation of sea surface temperature data, the ocean-mass component from Gravity Recovery and Climate Experiment (GRACE) satellite gravimetry, the contribution from global glacier mass changes assessed by a global glacier model, the contribution from Greenland Ice Sheet and Antarctic Ice Sheet mass changes assessed by satellite radar altimetry and by GRACE, and the contribution from land water storage anomalies assessed by the global hydrological model WaterGAP (Water Global Assessment and Prognosis). Over the period January 1993–December 2016 (P1, covered by the satellite altimetry records), the mean rate (linear trend) of GMSL is 3.05 ± 0.24 mm yr−1. The steric component is 1.15 ± 0.12 mm yr−1 (38 % of the GMSL trend), and the mass component is 1.75 ± 0.12 mm yr−1 (57 %). The mass component includes 0.64  ± 0.03 mm yr−1 (21 % of the GMSL trend) from glaciers outside Greenland and Antarctica, 0.60 ± 0.04 mm yr−1 (20 %) from Greenland, 0.19 ± 0.04 mm yr−1 (6 %) from Antarctica, and 0.32 ± 0.10 mm yr−1 (10 %) from changes of land water storage. In the period January 2003–August 2016 (P2, covered by GRACE and the Argo drifter system), GMSL rise is higher than in P1 at 3.64 ± 0.26 mm yr−1. This is due to an increase of the mass contributions, now about 2.40 ± 0.13 mm yr−1 (66 % of the GMSL trend), with the largest increase contributed from Greenland, while the steric contribution remained similar at 1.19 ± 0.17 mm yr−1 (now 33 %). The SLB of linear trends is closed for P1 and P2; that is, the GMSL trend agrees with the sum of the steric and mass components within their combined uncertainties. The OMB, which can be evaluated only for P2, shows that our preferred GRACE-based estimate of the ocean-mass trend agrees with the sum of mass contributions within 1.5 times or 0.8 times the combined 1σ uncertainties, depending on the way of assessing the mass contributions. Combined uncertainties (1σ) of the elements involved in the budgets are between 0.29 and 0.42 mm yr−1, on the order of 10 % of GMSL rise. Interannual variations that overlie the long-term trends are coherently represented by the elements of the SLB and the OMB. Even at the level of monthly anomalies the budgets are closed within uncertainties, while also indicating possible origins of remaining misclosures.

Satellite gravimetry data acquired by the Gravity Recovery and Climate Experiment (GRACE) allows to derive the temporal evolution in ice mass for both the Antarctic Ice Sheet (AIS) and the Greenland Ice Sheet (GIS). Various algorithms have been used in a wide range of studies to generate Gravimetric Mass Balance (GMB) products. Results from different studies may be affected by substantial differences in the processing, including the applied algorithm, the utilised background models and the time period under consideration. This study gives a detailed description of an assessment of the performance of GMB algorithms using actual GRACE monthly solutions for a prescribed period as well as synthetic data sets. The inter-comparison exercise was conducted in the scope of the European Space Agency’s Climate Change Initiative (CCI) project for the AIS and GIS, and was, for the first time, open to everyone. GMB products generated by different groups could be evaluated and directly compared against each other. For the period from 2003-02 to 2013-12, estimated linear trends in ice mass vary between −99 Gt/yr and −108 Gt/yr for the AIS and between −252 Gt/yr and −274 Gt/yr for the GIS, respectively. The spread between the solutions is larger if smaller drainage basins or gridded GMB products are considered. Finally, findings from the exercise formed the basis to select the algorithms used for the GMB product generation within the AIS and GIS CCI project.