Unraveling Brain Synchronisation Dynamics by Explainable Neural Networks using EEG Signals: Application to Dyslexia Diagnosis.

The electrical activity of the neural processes involved in cognitive functions is captured in EEG signals, allowing the exploration of the integration and coordination of neuronal oscillations across multiple spatiotemporal scales. We have proposed a novel approach that combines the transformation of EEG signal into image sequences, considering cross-frequency phase synchronisation (CFS) dynamics involved in low-level auditory processing, with the development of a two-stage deep learning model for the detection of developmental dyslexia (DD). This deep learning model exploits spatial and temporal information preserved in the image sequences to find discriminative patterns of phase synchronisation over time achieving a balanced accuracy of up to 83%. This result supports the existence of differential brain synchronisation dynamics between typical and dyslexic seven-year-old readers. Furthermore, we have obtained interpretable representations using a novel feature mask to link the most relevant regions during classification with the cognitive processes attributed to normal reading and those corresponding to compensatory mechanisms found in dyslexia.

Applications of machine learning and deep learning in SPECT and PET imaging: General overview, challenges and future prospects.

The integration of positron emission tomography (PET) and single-photon emission computed tomography (SPECT) imaging techniques with machine learning (ML) algorithms, including deep learning (DL) models, is a promising approach. This integration enhances the precision and efficiency of current diagnostic and treatment strategies while offering invaluable insights into disease mechanisms. In this comprehensive review, we delve into the transformative impact of ML and DL in this domain. Firstly, a brief analysis is provided of how these algorithms have evolved and which are the most widely applied in this domain. Their different potential applications in nuclear imaging are then discussed, such as optimization of image adquisition or reconstruction, biomarkers identification, multimodal fusion and the development of diagnostic, prognostic, and disease progression evaluation systems. This is because they are able to analyse complex patterns and relationships within imaging data, as well as extracting quantitative and objective measures. Furthermore, we discuss the challenges in implementation, such as data standardization and limited sample sizes, and explore the clinical opportunities and future horizons, including data augmentation and explainable AI. Together, these factors are propelling the continuous advancement of more robust, transparent, and reliable systems.

Use of Artificial Intelligence in the Identification and Diagnosis of Frailty Syndrome in Older Adults: Scoping Review.

BACKGROUND: Frailty syndrome (FS) is one of the most common noncommunicable diseases, which is associated with lower physical and mental capacities in older adults. FS diagnosis is mostly focused on biological variables; however, it is likely that this diagnosis could fail owing to the high biological variability in this syndrome. Therefore, artificial intelligence (AI) could be a potential strategy to identify and diagnose this complex and multifactorial geriatric syndrome. OBJECTIVE: The objective of this scoping review was to analyze the existing scientific evidence on the use of AI for the identification and diagnosis of FS in older adults, as well as to identify which model provides enhanced accuracy, sensitivity, specificity, and area under the curve (AUC). METHODS: A search was conducted using PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines on various databases: PubMed, Web of Science, Scopus, and Google Scholar. The search strategy followed Population/Problem, Intervention, Comparison, and Outcome (PICO) criteria with the population being older adults; intervention being AI; comparison being compared or not to other diagnostic methods; and outcome being FS with reported sensitivity, specificity, accuracy, or AUC values. The results were synthesized through information extraction and are presented in tables. RESULTS: We identified 26 studies that met the inclusion criteria, 6 of which had a data set over 2000 and 3 with data sets below 100. Machine learning was the most widely used type of AI, employed in 18 studies. Moreover, of the 26 included studies, 9 used clinical data, with clinical histories being the most frequently used data type in this category. The remaining 17 studies used nonclinical data, most frequently involving activity monitoring using an inertial sensor in clinical and nonclinical contexts. Regarding the performance of each AI model, 10 studies achieved a value of precision, sensitivity, specificity, or AUC ≥90. CONCLUSIONS: The findings of this scoping review clarify the overall status of recent studies using AI to identify and diagnose FS. Moreover, the findings show that the combined use of AI using clinical data along with nonclinical information such as the kinematics of inertial sensors that monitor activities in a nonclinical context could be an appropriate tool for the identification and diagnosis of FS. Nevertheless, some possible limitations of the evidence included in the review could be small sample sizes, heterogeneity of study designs, and lack of standardization in the AI models and diagnostic criteria used across studies. Future research is needed to validate AI systems with diverse data sources for diagnosing FS. AI should be used as a decision support tool for identifying FS, with data quality and privacy addressed, and the tool should be regularly monitored for performance after being integrated in clinical practice.

Enhancing Multimodal Patterns in Neuroimaging by Siamese Neural Networks with Self-Attention Mechanism.

The combination of different sources of information is currently one of the most relevant aspects in the diagnostic process of several diseases. In the field of neurological disorders, different imaging modalities providing structural and functional information are frequently available. Those modalities are usually analyzed separately, although a joint of the features extracted from both sources can improve the classification performance of Computer-Aided Diagnosis (CAD) tools. Previous studies have computed independent models from each individual modality and combined them in a subsequent stage, which is not an optimum solution. In this work, we propose a method based on the principles of siamese neural networks to fuse information from Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET). This framework quantifies the similarities between both modalities and relates them with the diagnostic label during the training process. The resulting latent space at the output of this network is then entered into an attention module in order to evaluate the relevance of each brain region at different stages of the development of Alzheimer's disease. The excellent results obtained and the high flexibility of the method proposed allow fusing more than two modalities, leading to a scalable methodology that can be used in a wide range of contexts.

Using Explainable Artificial Intelligence in the Clock Drawing Test to Reveal the Cognitive Impairment Pattern.

The prevalence of dementia is currently increasing worldwide. This syndrome produces a deterioration in cognitive function that cannot be reverted. However, an early diagnosis can be crucial for slowing its progress. The Clock Drawing Test (CDT) is a widely used paper-and-pencil test for cognitive assessment in which an individual has to manually draw a clock on a paper. There are a lot of scoring systems for this test and most of them depend on the subjective assessment of the expert. This study proposes a computer-aided diagnosis (CAD) system based on artificial intelligence (AI) methods to analyze the CDT and obtain an automatic diagnosis of cognitive impairment (CI). This system employs a preprocessing pipeline in which the clock is detected, centered and binarized to decrease the computational burden. Then, the resulting image is fed into a Convolutional Neural Network (CNN) to identify the informative patterns within the CDT drawings that are relevant for the assessment of the patient's cognitive status. Performance is evaluated in a real context where patients with CI and controls have been classified by clinical experts in a balanced sample size of [Formula: see text] drawings. The proposed method provides an accuracy of [Formula: see text] in the binary case-control classification task, with an AUC of [Formula: see text]. These results are indeed relevant considering the use of the classic version of the CDT. The large size of the sample suggests that the method proposed has a high reliability to be used in clinical contexts and demonstrates the suitability of CAD systems in the CDT assessment process. Explainable artificial intelligence (XAI) methods are applied to identify the most relevant regions during classification. Finding these patterns is extremely helpful to understand the brain damage caused by CI. A validation method using resubstitution with upper bound correction in a machine learning approach is also discussed.

Tiled Sparse Coding in Eigenspaces for Image Classification.

The automation in the diagnosis of medical images is currently a challenging task. The use of Computer Aided Diagnosis (CAD) systems can be a powerful tool for clinicians, especially in situations when hospitals are overflowed. These tools are usually based on artificial intelligence (AI), a field that has been recently revolutionized by deep learning approaches. These alternatives usually obtain a large performance based on complex solutions, leading to a high computational cost and the need of having large databases. In this work, we propose a classification framework based on sparse coding. Images are first partitioned into different tiles, and a dictionary is built after applying PCA to these tiles. The original signals are then transformed as a linear combination of the elements of the dictionary. Then, they are reconstructed by iteratively deactivating the elements associated with each component. Classification is finally performed employing as features the subsequent reconstruction errors. Performance is evaluated in a real context where distinguishing between four different pathologies: control versus bacterial pneumonia versus viral pneumonia versus COVID-19. Our system differentiates between pneumonia patients and controls with an accuracy of 97.74%, whereas in the 4-class context the accuracy is 86.73%. The excellent results and the pioneering use of sparse coding in this scenario evidence that our proposal can assist clinicians when their workload is high.

Neural representations of social valence bias economic interpersonal choices.

Prior personal information is highly relevant during social interactions. Such knowledge aids in the prediction of others, and it affects choices even when it is unrelated to actual behaviour. In this investigation, we aimed to study the neural representation of positive and negative personal expectations, how these impact subsequent choices, and the effect of mismatches between expectations and encountered behaviour. We employed functional Magnetic Resonance Imaging in combination with a version of the Ultimatum Game (UG) where participants were provided with information about their partners' moral traits previous to receiving their fair or unfair offers. Univariate and multivariate analyses revealed the implication of the supplementary motor area (SMA) and inferior frontal gyrus (IFG) in the representation of expectations about the partners in the game. Further, these regions also represented the valence of these expectations, together with the ventromedial prefrontal cortex (vmPFC). Importantly, the performance of multivariate classifiers in these clusters correlated with a behavioural choice bias to accept more offers following positive descriptions, highlighting the impact of the valence of the expectations on participants' economic decisions. Altogether, our results suggest that expectations based on social information guide future interpersonal decisions and that the neural representation of such expectations in the vmPFC is related to their influence on behaviour.

Neural representation of current and intended task sets during sequential judgements on human faces.

Engaging in a demanding activity while holding in mind another task to be performed in the near future requires the maintenance of information about both the currently-active task set and the intended one. However, little is known about how the human brain implements such action plans. While some previous studies have examined the neural representation of current task sets and others have investigated delayed intentions, to date none has examined the representation of current and intended task sets within a single experimental paradigm. In this fMRI study, we examined the neural representation of current and intended task sets, employing sequential classification tasks on human faces. Multivariate decoding analyses showed that current task sets were represented in the orbitofrontal cortex (OFC) and fusiform gyrus (FG), while intended tasks could be decoded from lateral prefrontal cortex (lPFC). Importantly, a ventromedial region in PFC/OFC contained information about both current and delayed tasks, although cross-classification between the two types of information was not possible. These results help delineate the neural representations of current and intended task sets, and highlight the importance of ventromedial PFC/OFC for maintaining task-relevant information regardless of when it is needed.

Atlas-Based Classification Algorithms for Identification of Informative Brain Regions in fMRI Data.

Multi-voxel pattern analysis (MVPA) has been successfully applied to neuroimaging data due to its larger sensitivity compared to univariate traditional techniques. Searchlight is the most widely employed approach to assign functional value to different regions of the brain. However, its performance depends on the size of the sphere, which can overestimate the region of activation when a large sphere size is employed. In the current study, we examined the validity of two different alternatives to Searchlight: an atlas-based local averaging method (ABLA, Schrouff et al. Neuroinformatics 16, 117-143, 2013a) and a Multi-Kernel Learning (MKL, Rakotomamonjy et al. Journal of Machine Learning 9, 2491-2521, 2008) approach, in a scenario where the goal is to find the informative brain regions that support certain mental operations. These methods employ weights to measure the informativeness of a brain region and highly reduce the large computational cost that Searchlight entails. We evaluated their performance in two different scenarios where the differential BOLD activation between experimental conditions was large vs. small, and employed nine different atlases to assess the influence of diverse brain parcellations. Results show that both methods were able to localize informative regions when differences between conditions were large, demonstrating a large sensitivity and stability in the identification of regions across atlases. Moreover, the sign of the weights reported by these methods provided the directionality of univariate approaches. However, when differences were small, only ABLA localized informative regions. Thus, our results show that atlas-based methods are useful alternatives to Searchlight, but that the nature of the classification to perform should be taken into account when choosing the specific method to implement.

Representational Organization of Novel Task Sets during Proactive Encoding.

Recent multivariate analyses of brain data have boosted our understanding of the organizational principles that shape neural coding. However, most of this progress has focused on perceptual visual regions (Connolly et al., 2012), whereas far less is known about the organization of more abstract, action-oriented representations. In this study, we focused on humans' remarkable ability to turn novel instructions into actions. While previous research shows that instruction encoding is tightly linked to proactive activations in frontoparietal brain regions, little is known about the structure that orchestrates such anticipatory representation. We collected fMRI data while participants (both males and females) followed novel complex verbal rules that varied across control-related variables (integrating within/across stimuli dimensions, response complexity, target category) and reward expectations. Using representational similarity analysis (Kriegeskorte et al., 2008), we explored where in the brain these variables explained the organization of novel task encoding, and whether motivation modulated these representational spaces. Instruction representations in the lateral PFC were structured by the three control-related variables, whereas intraparietal sulcus encoded response complexity and the fusiform gyrus and precuneus organized its activity according to the relevant stimulus category. Reward exerted a general effect, increasing the representational similarity among different instructions, which was robustly correlated with behavioral improvements. Overall, our results highlight the flexibility of proactive task encoding, governed by distinct representational organizations in specific brain regions. They also stress the variability of motivation-control interactions, which appear to be highly dependent on task attributes, such as complexity or novelty.SIGNIFICANCE STATEMENT In comparison with other primates, humans display a remarkable success in novel task contexts thanks to our ability to transform instructions into effective actions. This skill is associated with proactive task-set reconfigurations in frontoparietal cortices. It remains yet unknown, however, how the brain encodes in anticipation the flexible, rich repertoire of novel tasks that we can achieve. Here we explored cognitive control and motivation-related variables that might orchestrate the representational space for novel instructions. Our results showed that different dimensions become relevant for task prospective encoding, depending on the brain region, and that the lateral PFC simultaneously organized task representations following different control-related variables. Motivation exerted a general modulation upon this process, diminishing rather than increasing distances among instruction representations.

Transient and Sustained Control Mechanisms Supporting Novel Instructed Behavior.

The success of humans in novel environments is partially supported by our ability to implement new task procedures via instructions. This complex skill has been associated with the activity of control-related brain areas. Current models link fronto-parietal and cingulo-opercular networks with transient and sustained modes of cognitive control, based on observations during repetitive task settings or rest. The current study extends this dual model to novel instructed tasks. We employed a mixed design and an instruction-following task to extract phasic and tonic brain signals associated with the encoding and implementation of novel verbal rules. We also performed a representation similarity analysis to capture consistency in task-set encoding within trial epochs. Our findings show that both networks are involved while following novel instructions: transiently, during the implementation of the instruction, and in a sustained fashion, across novel trials blocks. Moreover, the multivariate results showed that task representations in the cingulo-opercular network were more stable than in the fronto-parietal one. Our data extend the dual model of cognitive control to novel demanding situations, highlighting the high flexibility of control-related regions in adopting different temporal profiles.

Influence of activation pattern estimates and statistical significance tests in fMRI decoding analysis.

The use of Multi-Voxel Pattern Analysis (MVPA) has increased considerably in recent functional magnetic resonance imaging (fMRI) studies. A crucial step consists in the choice of a method for the estimation of responses. However, a systematic comparison of the different estimation alternatives and their adequacy to predominant experimental design is missing. In the current study we compared three pattern estimation methods: Least-Squares Unitary (LSU), based on run-wise estimation, Least-Squares All (LSA) and Least-Squares Separate (LSS), which rely on trial-wise estimation. We compared the efficiency of these methods in an experiment where sustained activity needed to be isolated from zero-duration events as well as in a block-design approach and in a event-related design. We evaluated the sensitivity of the t-test in comparison with two non-parametric methods based on permutation testing: one proposed in Stelzer et al. (2013), equivalent to performing a permutation in each voxel separately and the Threshold-Free Cluster Enhancement. LSS resulted the most accurate approach to address the large overlap of signal among close events in the event-related designs. We found a larger sensitivity of Stelzer's method in all settings, especially in the event-related designs, where voxels close to surpass the statistical threshold with the other approaches were now marked as informative regions. Our results provide evidence that LSS is the most accurate approach for unmixing events with different duration and large overlap of signal. This is consistent with previous studies showing that LSS handles large collinearity better than other methods. Moreover, Stelzer's potentiates this better estimation with its large sensitivity.

Encoding, preparation and implementation of novel complex verbal instructions.

Verbal instructions allow humans to acquire and implement complex novel rules in few seconds. A major question that remains elusive is how the brain represents this information prior to successful task execution. In this experiment, we studied the brain regions involved in representing categorical stimulus information during the encoding of novel instructions, their preparation and also their implementation, as well as the relation of the fidelity of these representations to observable behavior. To do so, we devised a novel instructions paradigm to delimitate these three stages. Using univariate and multivariate analyses of functional magnetic resonance data, our study revealed that the semantic content (faces or letters) of complex novel instructions can be decoded several seconds before the onset of a target, as soon as instructions are encoded. Crucially, the quality of the information represented in domain-general and category-selective regions correlated with subsequent behavioral performance. This suggests that the rapid transformation of novel instructions into coherent behavior is supported by control mechanisms that use available, relevant information about the current rule prior to its execution. In addition, our results highlight the relation between these control processes and others such as prospective memory and maintenance of future intentions.