The present study performs a comprehensive fairness analysis of machine learning (ML) models for the diagnosis of Mild Cognitive Impairment (MCI) and Alzheimer’s disease (AD) from MRI-derived neuroimaging features. B...
详细信息
Electroencephalography (EEG) and near-infrared spectroscopy (NIRS) are commonly used for neural source reconstruction, but have limited spatial or temporal resolution. Using NIRS reconstruction as a spatial prior for ...
详细信息
Second harmonic generation microscopy (SHG) is a powerful imaging modality which has found applications in investigating both biological and synthetic nanostructures. Like all optical microscopy techniques, the resolu...
详细信息
Deep learning (DL) applications in the medical field often face challenges related to limited data availability, resulting in issues like overfitting and imbalanced datasets. Synthetic data offers a promising solution...
详细信息
ISBN:
(数字)9798350313338
ISBN:
(纸本)9798350313345
Deep learning (DL) applications in the medical field often face challenges related to limited data availability, resulting in issues like overfitting and imbalanced datasets. Synthetic data offers a promising solution to these problems by enabling data augmentation and enhancing the performance of DL models. In this study, we trained the state-of-the-art generative model StyleGAN2-ADA on 1412 images from the Alzheimer’s disease neuroimaging initiative (ADNI) dataset to generate synthetic slices of T1-weighted brain MRI of healthy subjects. The quality of the synthetic images has been evaluated through quantitative and qualitative assessments, including a visual Turing test conducted by an expert observer with 2000 images. The observer achieved an accuracy of 52.95%, indicative of a performance level comparable to random guessing. These results demonstrate the capability of StyleGAN2-ADA to generate anatomically relevant synthetic brain MRI data.
Background and objectives: Depression inflicts significant harm on both society and family. Previous studies have indicated that the functional network of EEG signals worked well in recognizing major depression. This ...
详细信息
We will discuss the use of deep learning to infer fluorescence and nonlinear contrast from the texture and morphology of reflectance confocal microscopy. Our goal is to synthesize H&E-like contrast from existing F...
详细信息
Current image-correction frameworks for sensors that employ optically coherent detection attempt to estimate phase errors in the data, like those caused by aberrations, and simultaneously reconstruct digitally enhance...
详细信息
This paper concerns the energy efficiencyoptimization for distributed cooperative spectrum sensing. In the considered distributed spectrum sensing system, each sensor measures the local test statistic for the target s...
ISBN:
(数字)9781728149059
ISBN:
(纸本)9781728149066
This paper concerns the energy efficiencyoptimization for distributed cooperative spectrum sensing. In the considered distributed spectrum sensing system, each sensor measures the local test statistic for the target spectrum bands and these measurements will be combined together through a weighted consensus protocol. In this way, all the sensors are able to make contributions to the improvement of the spectrum sensing performance. However, the significance of a sensor's contribution depends on its local signal to noise ratio. From energy efficiency perspective, it is not reasonable to invest energy on the sensors that bring little benefits. In this paper, we formulate an energy efficiency optimization framework for distributed spectrum sensing. Our objective is to achieve the target spectrum sensing performance with as few sensors as possible. Accordingly, a genetic algorithm based approach and a particle swarm optimization based approach are proposed for this problem.
Introduction: Hand function is a central determinant of independence after stroke. Measuring hand use in the home environment is necessary to evaluate the impact of new interventions, and calls for novel wearable tech...
详细信息
Epilepsy has now become a serious global health issue. Approximately 1% of people globally experience epileptic seizures. Predicting seizures before they start helps use medication to avoid them. Using modern computer...
Epilepsy has now become a serious global health issue. Approximately 1% of people globally experience epileptic seizures. Predicting seizures before they start helps use medication to avoid them. Using modern computer tools, machine learning, and deep learning techniques, EEG has been used to predict seizures. The main difficulties in designing epilepsy prediction algorithms are feature selection and classification. Automatic detection and prediction of epileptic seizures are carried out much faster through deep learning techniques than traditional approaches. Designing a deep learning network from scratch or using a pre-trained network not only depends on the size, variability of data, complexity of the network, number of layers, network parameters, depth, etc. This work presents A Fast and Novel Deep Learning Approach for the Automatic Classification of Epileptic Seizures using Spectrograms. The seizure data was acquired from the TUH EEG Corpus. Pre-trained networks were used on four sets of spectrogram images of normal and seizure data and reorganized within the networks, tuned on the parameters, such as Convolution layers, Classification layers, etc. The above networks were trained and tested for each set of spectrograms, and the performance was evaluated based on testing accuracy and time for the training and validation process. The proposed method achieved remarkable accuracy with the capability of retrieving the signal information at each instant. Tuned ResNet-18 achieved an average accuracy of 98.9%, with an operating time of 1056 to 3606 seconds, while modified AlexNet proved to be the fastest, with an operating time of 421 to 425 seconds with an average accuracy of 98.2 for various features selected.
暂无评论