Explainable AI: A review of applications to neuroimaging data.

Deep neural networks (DNNs) have transformed the field of computer vision and currently constitute some of the best models for representations learned via hierarchical processing in the human brain. In medical imaging, these models have shown human-level performance and even higher in the early diagnosis of a wide range of diseases. However, the goal is often not only to accurately predict group membership or diagnose but also to provide explanations that support the model decision in a context that a human can readily interpret. The limited transparency has hindered the adoption of DNN algorithms across many domains. Numerous explainable artificial intelligence (XAI) techniques have been developed to peer inside the "black box" and make sense of DNN models, taking somewhat divergent approaches. Here, we suggest that these methods may be considered in light of the interpretation goal, including functional or mechanistic interpretations, developing archetypal class instances, or assessing the relevance of certain features or mappings on a trained model in a post-hoc capacity. We then focus on reviewing recent applications of post-hoc relevance techniques as applied to neuroimaging data. Moreover, this article suggests a method for comparing the reliability of XAI methods, especially in deep neural networks, along with their advantages and pitfalls.

The Chaotic Behavior of the Spread of Infection during the COVID-19 Pandemic in Japan.

In December 2019, China reported a new virus identified as SARS-CoV-2, causing COVID-19, which soon spread to other countries and led to a global pandemic. Although many countries imposed strict actions to control the spread of the virus, the COVID-19 pandemic resulted in unprecedented economic and social consequences in 2020 and early 2021. To understand the dynamics of the spread of the virus, we evaluated its chaotic behavior in Japan. A 0-1 test was applied to the time-series data of daily COVID-19 cases from January 26, 2020 to August 5, 2021 (3 days before the end of the Tokyo Olympic Games). Additionally, the influence of hosting the Olympic Games in Tokyo was assessed in data including the post-Olympic period until October 8, 2021. Even with these extended time period data, although the time-series data for the daily infections across Japan were not found to be chaotic, more than 76.6% and 55.3% of the prefectures in Japan showed chaotic behavior in the pre- and post-Olympic Games periods, respectively. Notably, Tokyo and Kanagawa, the two most populous cities in Japan, did not show chaotic behavior in their time-series data of daily COVID-19 confirmed cases. Overall, the prefectures with the largest population centers showed non-chaotic behavior, whereas the prefectures with smaller populations showed chaotic behavior. This phenomenon was observed in both of the analyzed time periods (pre- and post-Olympic Games); therefore, more attention should be paid to prefectures with smaller populations, in which controlling and preventing the current pandemic is more difficult.

Decoding Task-Based fMRI Data with Graph Neural Networks, Considering Individual Differences.

Task fMRI provides an opportunity to analyze the working mechanisms of the human brain during specific experimental paradigms. Deep learning models have increasingly been applied for decoding and encoding purposes study to representations in task fMRI data. More recently, graph neural networks, or neural networks models designed to leverage the properties of graph representations, have recently shown promise in task fMRI decoding studies. Here, we propose an end-to-end graph convolutional network (GCN) framework with three convolutional layers to classify task fMRI data from the Human Connectome Project dataset. We compared the predictive performance of our GCN model across four of the most widely used node embedding algorithms-NetMF, RandNE, Node2Vec, and Walklets-to automatically extract the structural properties of the nodes in the functional graph. The empirical results indicated that our GCN framework accurately predicted individual differences (0.978 and 0.976) with the NetMF and RandNE embedding methods, respectively. Furthermore, to assess the effects of individual differences, we tested the classification performance of the model on sub-datasets divided according to gender and fluid intelligence. Experimental results indicated significant differences in the classification predictions of gender, but not high/low fluid intelligence fMRI data. Our experiments yielded promising results and demonstrated the superior ability of our GCN in modeling task fMRI data.

The COVID-19 Infection Diffusion in the US and Japan: A Graph-Theoretical Approach.

Coronavirus disease 2019 (COVID-19) was first discovered in China; within several months, it spread worldwide and became a pandemic. Although the virus has spread throughout the globe, its effects have differed. The pandemic diffusion network dynamics (PDND) approach was proposed to better understand the spreading behavior of COVID-19 in the US and Japan. We used daily confirmed cases of COVID-19 from 5 January 2020 to 31 July 2021, for all states (prefectures) of the US and Japan. By applying the pandemic diffusion network dynamics (PDND) approach to COVID-19 time series data, we developed diffusion graphs for the US and Japan. In these graphs, nodes represent states and prefectures (regions), and edges represent connections between regions based on the synchrony of COVID-19 time series data. To compare the pandemic spreading dynamics in the US and Japan, we used graph theory metrics, which targeted the characterization of COVID-19 bedhavior that could not be explained through linear methods. These metrics included path length, global and local efficiency, clustering coefficient, assortativity, modularity, network density, and degree centrality. Application of the proposed approach resulted in the discovery of mostly minor differences between analyzed countries. In light of these findings, we focused on analyzing the reasons and defining research hypotheses that, upon addressing, could shed more light on the complex phenomena of COVID-19 virus spread and the proposed PDND methodology.

Neural Decoding of EEG Signals with Machine Learning: A Systematic Review.

Electroencephalography (EEG) is a non-invasive technique used to record the brain's evoked and induced electrical activity from the scalp. Artificial intelligence, particularly machine learning (ML) and deep learning (DL) algorithms, are increasingly being applied to EEG data for pattern analysis, group membership classification, and brain-computer interface purposes. This study aimed to systematically review recent advances in ML and DL supervised models for decoding and classifying EEG signals. Moreover, this article provides a comprehensive review of the state-of-the-art techniques used for EEG signal preprocessing and feature extraction. To this end, several academic databases were searched to explore relevant studies from the year 2000 to the present. Our results showed that the application of ML and DL in both mental workload and motor imagery tasks has received substantial attention in recent years. A total of 75% of DL studies applied convolutional neural networks with various learning algorithms, and 36% of ML studies achieved competitive accuracy by using a support vector machine algorithm. Wavelet transform was found to be the most common feature extraction method used for all types of tasks. We further examined the specific feature extraction methods and end classifier recommendations discovered in this systematic review.

Optimizing COVID-19 vaccine distribution across the United States using deterministic and stochastic recurrent neural networks.

Optimizing COVID-19 vaccine distribution can help plan around the limited production and distribution of vaccination, particularly in early stages. One of the main criteria for equitable vaccine distribution is predicting the geographic distribution of active virus at the time of vaccination. This research developed sequence-learning models to predict the behavior of the COVID-19 pandemic across the US, based on previously reported information. For this objective, we used two time-series datasets of confirmed COVID-19 cases and COVID-19 effective reproduction numbers from January 22, 2020 to November 26, 2020 for all states in the US. The datasets have 310 time-steps (days) and 50 features (US states). To avoid training the models for all states, we categorized US states on the basis of their similarity to previously reported COVID-19 behavior. For this purpose, we used an unsupervised self-organizing map to categorize all states of the US into four groups on the basis of the similarity of their effective reproduction numbers. After selecting a leading state (the state with earliest outbreaks) in each group, we developed deterministic and stochastic Long Short Term Memory (LSTM) and Mixture Density Network (MDN) models. We trained the models with data from each leading state to make predictions, then compared the models with a baseline linear regression model. We also remove seasonality and trends from a dataset of non-stationary COVID-19 cases to determine the effects on prediction. We showed that the deterministic LSTM model trained on the COVID-19 effective reproduction numbers outperforms other prediction methods.

Predicting the Dynamics of the COVID-19 Pandemic in the United States Using Graph Theory-Based Neural Networks.

The COVID-19 pandemic has had unprecedented social and economic consequences in the United States. Therefore, accurately predicting the dynamics of the pandemic can be very beneficial. Two main elements required for developing reliable predictions include: (1) a predictive model and (2) an indicator of the current condition and status of the pandemic. As a pandemic indicator, we used the effective reproduction number (Rt), which is defined as the number of new infections transmitted by a single contagious individual in a population that may no longer be fully susceptible. To bring the pandemic under control, Rt must be less than one. To eliminate the pandemic, Rt should be close to zero. Therefore, this value may serve as a strong indicator of the current status of the pandemic. For a predictive model, we used graph neural networks (GNNs), a method that combines graphical analysis with the structure of neural networks. We developed two types of GNN models, including: (1) graph-theory-based neural networks (GTNN) and (2) neighborhood-based neural networks (NGNN). The nodes in both graphs indicated individual states in the United States. While the GTNN model's edges document functional connectivity between states, those in the NGNN model link neighboring states to one another. We trained both models with Rt numbers collected over the previous four days and asked them to predict the following day for all states in the United States. The performance of these models was evaluated with the datasets that included Rt values reflecting conditions from 22 January through 26 November 2020 (before the start of COVID-19 vaccination in the United States). To determine the efficiency, we compared the results of two models with each other and with those generated by a baseline Long short-term memory (LSTM) model. The results indicated that the GTNN model outperformed both the NGNN and LSTM models for predicting Rt.

A Study of the Effects of the COVID-19 Pandemic on the Experience of Back Pain Reported on Twitter® in the United States: A Natural Language Processing Approach.

The COVID-19 pandemic has changed our lifestyles, habits, and daily routine. Some of the impacts of COVID-19 have been widely reported already. However, many effects of the COVID-19 pandemic are still to be discovered. The main objective of this study was to assess the changes in the frequency of reported physical back pain complaints reported during the COVID-19 pandemic. In contrast to other published studies, we target the general population using Twitter as a data source. Specifically, we aim to investigate differences in the number of back pain complaints between the pre-pandemic and during the pandemic. A total of 53,234 and 78,559 tweets were analyzed for November 2019 and November 2020, respectively. Because Twitter users do not always complain explicitly when they tweet about the experience of back pain, we have designed an intelligent filter based on natural language processing (NLP) to automatically classify the examined tweets into the back pain complaining class and other tweets. Analysis of filtered tweets indicated an 84% increase in the back pain complaints reported in November 2020 compared to November 2019. These results might indicate significant changes in lifestyle during the COVID-19 pandemic, including restrictions in daily body movements and reduced exposure to routine physical exercise.

Convolutional Neural Networks-Based Image Analysis for the Detection and Quantification of Neutrophil Extracellular Traps.

Over a decade ago, the formation of neutrophil extracellular traps (NETs) was described as a novel mechanism employed by neutrophils to tackle infections. Currently applied methods for NETs release quantification are often limited by the use of unspecific dyes and technical difficulties. Therefore, we aimed to develop a fully automatic image processing method for the detection and quantification of NETs based on live imaging with the use of DNA-staining dyes. For this purpose, we adopted a recently proposed Convolutional Neural Network (CNN) model called Mask R-CNN. The adopted model detected objects with quality comparable to manual counting-Over 90% of detected cells were classified in the same manner as in manual labelling. Furthermore, the inhibitory effect of GW 311616A (neutrophil elastase inhibitor) on NETs release, observed microscopically, was confirmed with the use of the CNN model but not by extracellular DNA release measurement. We have demonstrated that a modern CNN model outperforms a widely used quantification method based on the measurement of DNA release and can be a valuable tool to quantitate the formation process of NETs.

Competitive fitness analysis using Convolutional Neural Network.

We developed a procedure for estimating competitive fitness by using Caenorhabditis elegans as a model organism and a Convolutional Neural Network (CNN) as a tool. Competitive fitness is usually the most informative fitness measure, and competitive fitness assays often rely on green fluorescent protein (GFP) marker strains. CNNs are a class of deep learning neural networks, which are well suited for image analysis and object classification. Our model analyses involved image classification of nematodes as wild-type vs. GFP-expressing, and counted both categories. The performance was analyzed with (i) precision and recall parameters, and (ii) comparison of the wild-type frequency calculated from the model against that obtained by visual scoring of the same images. The average precision and recall varied from 0.79 to 0.87 and from 0.84 to 0.92, respectively, depending on worm density in the images. Compared with manual counting, the model decreased counting time at least 20-fold while preventing human errors. Given the rapid development in the field of CNN, the model, which is fully available on GitHub, can be further optimized and adapted for other image-based uses.

Physiological parameters depending on two different types of manual wheelchair propulsion.

OBJECTIVE: The purpose of this study was to compare aerobic parameters in the multistage field test (MFT) in hand rim wheelchair propulsion and lever wheelchair propulsion. METHODS: Twenty-one men performed MFT using two different types of propulsion, i.e., lever and hand rim wheelchair propulsion. The covered distance and physiological variables (oxygen uptake (VO2), minute ventilation (VE), carbon dioxide output (VCO2), respiratory coefficient (RQ), and heart rate (HR)) were observed. Physiological variables were measured with Cosmed K5 system. Kolmogorov-Smirnov test, t-test, Wilcoxon test and effect sizes (ESs) were used to assess differences. Statistical significance was set at p < .05. RESULTS: A significantly longer distance was observed in lever wheelchair propulsion than in hand rim wheelchair propulsion (1,194 and 649 m, respectively). VO2max and RQ were higher in hand rim wheelchair propulsion. All physiological variables for the last (fifth) level of the test in hand rim propulsion were significantly higher than in lever wheelchair propulsion. ES was large for each observed difference. CONCLUSION: The lever wheelchair propulsion movement is less demanding than hand rim wheelchair propulsion and longer distances can be achieved by the user. There is a need to check lever wheelchair propulsion in different types of field tests.

Comparison of muscle activity during hand rim and lever wheelchair propulsion over flat terrain.

PURPOSE: The aim of this study was to compare the activity of upper limb muscles during hand rim wheelchair propulsion and lever wheelchair propulsion at two different velocity levels. METHODS: Twenty male volunteers with physical impairments participated in this study. Their task was to push a lever wheelchair and a hand rim wheelchair on a mechanical wheelchair treadmill for 4 minutes at a speed of 3.5 km/h and 4.5 km/h in a flat race setting (conditions of moving over flat terrain). During these trials, activity of eight muscles of upper limbs were examined using surface electromyography. RESULTS: The range of motion in the elbow joint was significantly higher in lever wheelchair propulsion (59.8 ± 2.43°) than in hand rim wheelchair propulsion (43.9 ± 0.26°). Such values of kinematics resulted in a different activity of muscles. All the muscles were more active during lever wheelchair propulsion at both velocity levels. The only exceptions were extensor and flexor carpi muscles which were more active during hand rim wheelchair propulsion due to the specificity of a grip. In turn, the examined change in the velocity (by 1 km/h) while moving over flat terrain also caused a different EMG timing of muscle activation depending on the type of propulsion. CONCLUSIONS: Lever wheelchair propulsion seems to be a good alternative to hand rim wheelchair propulsion owing to a different movement technique and a different EMG timing of muscle activity. Therefore, we believe that lever wheelchair propulsion should serve as supplement to traditional propulsion.

How does lever length and the position of its axis of rotation influence human performance during lever wheelchair propulsion?

The purpose of this study was to investigate empirically how lever length and its axis of rotation position influences human performance during lever wheelchair propulsion. In order to fulfill this goal, a dedicated test stand allowing easy implementation of various lever positions was created. In the experiment, 10 young, healthy, male subjects performed 8 tests consisting of propulsion work with levers of different lengths and lever axis of rotation positions. During tests heart rate, oxygen consumption and EMG assessment of 6 muscles was carried out. Measurements of power output on the test stand were done as well. Together with oxygen consumption analysis, this allowed calculation of human work efficiency. The results show significant (p<0.05 and p<0.001) differences between lever configurations when comparing various parameters values. From the carried out experiments, the authors conclude that levers' length and their axis of rotation position significantly influence human performance during lever wheelchair propulsion. For the examined subjects, placing the levers' axis of rotation close behind the back wheels axis of rotation offered advantageous work conditions.