Abstract: Typically, high NREM stage N3 sleep detection accuracy is achieved using a frontal electrode referenced to an electrode at a distant location on the head (e.g., the mastoid, or the earlobe). For comfort and design considerations it is more convenient to have active and reference electrodes closely positioned on the frontal region of the head. This configuration, however, significantly attenuates the signal, which degrades sleep stage detection (e.g., N3) performance. The present disclosure describes a deep neural network (DNN) based solution developed to detect sleep using frontal electrodes only. N3 detection is enhanced through post-processing of the soft DNN outputs. Detection of slow-waves and sleep micro-arousals is accomplished using frequency domain thresholds. Volume modulation uses a high-frequency/low-frequency spectral ratio extracted from the frontal signal.

Abstract: The invention relates a computer-implemented method (500) of generating explainability information for explaining a model output of a trained model. The method uses one or more aspect recognition models configured to indicate a presence of respective characteristics in the input instance. A saliency method is applied to obtain a masked source representation of the input instance at a source layer of the trained model (e.g., the input layer or an internal layer), comprising those elements at the source layer relevant to the model output. The masked source representation is mapped to a target layer (e.g., input or internal layer) of an aspect recognition model, and the aspect recognition model is then applied to obtain a model output indicating a presence of the given characteristic relevant to the model output of the trained model. As explainability information, the characteristics indicated by the aspect recognition models are output.

Abstract: A concept for training a neural network model. The concept comprises receiving training data and test data, each comprising a set of annotated images. A neural network model is trained using the training data with an initial regularization parameter. Loss functions of the neural network for both the training data and the test data are used to modify the regularization parameter, and the neural network model is retrained using the modified regularization parameter. This process is iteratively repeated until the loss functions both converge. A system, method and a computer program product embodying this concept are disclosed.

Abstract: Aspects and embodiments relate to a method of providing a representation of a feature identified by a deep neural network as being relevant to an outcome, a computer program product and apparatus configured to perform that method.

Abstract: The present invention provides a method, computer program and processing system for identifies boundaries of lesions within image data. The image data is processed using a machine learning algorithm to generate probability data and uncertainty data. The probability data provides, for each image data point of the image data, a probability data points indicating a probability that said image data point is part of a lesion. The uncertainty data provides, for each probability data point, an uncertainty data point indicating an uncertainty of the said probability data point. The uncertainty data is processed to identify or correct boundaries of the lesions.

Abstract: Presented are concepts for feature identification in medical imaging of a subject. One such concept processes a medical image with a Bayesian deep learning network to determine a first image feature of interest and an associated uncertainty value, the first image feature being located in a first sub-region of the image. It also processes the medical image with a generative adversarial network to determine a second image feature of interest within the first sub-region of the image and an associated uncertainty value. Based on the first and second image features and their associated uncertainty values, the first sub-region of the image is classified.

Abstract: The invention relates to a trained model, such as a trained neural network, which is trained on training data. System and computer-implemented methods are provided for generating metadata which encodes a numerical characteristic of the training data of the trained model, and for using the metadata to determine conformance of input data of the trained model to the numerical characteristics of the training data. If the input data does not conform to the numerical characteristics, the use of the trained model on the input data may be considered out-of-specification (‘out-of-spec’). Accordingly, a system applying the trained model to the input data may, for example, warn a user of the non-conformance, or may decline to apply the trained model to the input data, etc.

Abstract: An adaptable neural network system (1) formed of two neural networks (4, 5). One of the neural networks (5) adjusts a structure of the other neural network (4) based on information about a specific task each time that new second input data (12) indicative of a desired task is received by the one neural network (5), so that the other neural network (4) is adapted to perform that specific task. Thus, an adaptable neural network system (1) capable of performing different tasks on input data (11) can be realized.

Abstract: A concept for training a neural network model. The concept comprises receiving training data and test data, each comprising a set of annotated images. A neural network model is trained using the training data with an initial regularization parameter. Loss functions of the neural network for both the training data and the test data are used to modify the regularization parameter, and the neural network model is retrained using the modified regularization parameter. This process is iteratively repeated until the loss functions both converge. A system, method and a computer program product embodying this concept are disclosed.

Abstract: Typically, high NREM stage N3 sleep detection accuracy is achieved using a frontal electrode referenced to an electrode at a distant location on the head (e.g., the mastoid, or the earlobe). For comfort and design considerations it is more convenient to have active and reference electrodes closely positioned on the frontal region of the head. This configuration, however, significantly attenuates the signal, which degrades sleep stage detection (e.g., N3) performance. The present disclosure describes a deep neural network (DNN) based solution developed to detect sleep using frontal electrodes only. N3 detection is enhanced through post-processing of the soft DNN outputs. Detection of slow-waves and sleep micro-arousals is accomplished using frequency domain thresholds. Volume modulation uses a high-frequency/low-frequency spectral ratio extracted from the frontal signal.

Abstract: Presented are concepts for feature identification in medical imaging of a subject. One such concept processes a medical image with a Bayesian deep learning network to determine a first image feature of interest and an associated uncertainty value, the first image feature being located in a first sub-region of the image. It also processes the medical image with a generative adversarial network to determine a second image feature of interest within the first sub-region of the image and an associated uncertainty value. Based on the first and second image features and their associated uncertainty values, the first sub-region of the image is classified.