Methods, apparatus, and devices for child reading and attention deficit risk screening

Abstract

This application relates to a method, apparatus, and device for screening the risk of reading and attention deficit disorder in children. The method includes acquiring multi-channel raw brain blood oxygenation signals under task-induced conditions using a specific layout fNIRS array integrated into a wearable headband, based on a rapid naming cognitive paradigm. Based on the raw brain blood oxygenation signals, a fusion feature vector representing the reading and attention networks is generated by calculating temporal waveform features and frontotemporal functional connectivity strength. The multi-dimensional fusion feature vector is then processed and analyzed using a Transformer classification model to generate classification results indicating the risk level of reading disorders and comorbid ADHD. This application achieves portable and rapid brain function signal acquisition by integrating a targeted fNIRS array with a standardized cognitive paradigm. By fusing temporal dynamics and brain network connectivity features, a multi-dimensional neural representation is constructed. Finally, a lightweight Transformer model is used to output the risk level of reading disorders and comorbid ADHD end-to-end, achieving high-precision automated assisted screening.

Description

Technical Field

[0001] This application relates to the field of medical diagnostic technology, and in particular to a method, device, and equipment for screening the risk of reading and attention deficit in children. Background Technology

[0002] Early identification and screening of dyslexia and attention deficit are important issues in the field of children's neurodevelopmental health, and are crucial for timely educational intervention and improving children's long-term academic and social functioning. With the development of cognitive neuroscience, achieving objective and quantitative assessments using brain function testing technologies has become an important direction for improving screening effectiveness. However, how to build a large-scale, rapid screening technology system suitable for primary healthcare and educational institutions while ensuring scientific validity remains a significant challenge.

[0003] Currently, mainstream screening methods in clinical and school settings heavily rely on standardized behavioral scales (such as the Chinese Reading Diagnostic Test). These methods have inherent technical limitations: First, their implementation is subject to a significant age window delay, typically requiring children to have received systematic literacy education (generally ≥6 years old), thus missing the golden period for earlier intervention; second, assessment results are highly susceptible to interference from non-target factors such as the examiner's testing experience, the child's emotional state on the day of the test, testing motivation, and level of cooperation, leading to fluctuations in reliability and validity, strong subjectivity, and difficulty in establishing a stable and objective baseline.

[0004] In recent years, cognitive neuroscience research has seen the emergence of paradigms that utilize techniques such as high temporal resolution event-related potentials (ERPs) or high spatial resolution functional magnetic resonance imaging (fMRI) to explore brain mechanisms. While these techniques can provide objective neurophysiological indicators, their application scenarios are severely limited: experimental task design is complex, a single test is time-consuming (often 20–40 minutes), the required equipment is extremely expensive, and specialized shielded environments and professional technicians are required for operation. These factors collectively result in such methods being costly, inefficient, and inaccessible. Essentially, they are laboratory tools geared towards in-depth scientific research and are completely unsuitable for the large-scale screening needs in outpatient, school, or community settings where timeliness, convenience, and cost-effectiveness are paramount.

[0005] Therefore, there is a significant gap between the existing technological system and its accessibility. The industry urgently needs a new brain function testing solution that can ensure the objectivity of neurophysiological testing while also being quick to operate, portable, environmentally adaptable, and cost-effective, in order to fill the technological gap between laboratory research and large-scale field applications. Summary of the Invention

[0006] Based on this, this application provides a method for screening children's reading and attention deficit risks, including: Based on the cognitive paradigm of rapid naming, multi-channel raw cerebral blood oxygenation signals under task-induced conditions are obtained through a specially laid-out fNIRS array integrated into a wearable headband. Based on the original cerebral blood oxygenation signal, a fusion feature vector representing the reading and attention network is generated by calculating the temporal waveform features and the frontotemporal functional connectivity strength. Based on the multi-dimensional fused feature vectors, the data are processed and analyzed using the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

[0007] Optionally, the acquisition of multi-channel raw cerebral blood oxygenation signals induced by a task, based on a cognitive paradigm of rapid naming and through a specifically laid-out fNIRS array integrated into a wearable headband, includes: According to the rapid naming paradigm, the stimulus presentation module is controlled by a script embedded in the FPGA to sequentially present a calibration blank screen and a digital matrix on the display screen to induce target cognitive operation. Based on the preset frontotemporal reading network and parietal attention network targets, light signals from multiple brain regions during the task are collected at a preset sampling rate using a multi-channel dual-wavelength near-infrared light pole integrated into a wearable structure. Based on the acquired raw optical signals, the raw data of HbO and HbR concentration changes over time in each channel are obtained through analog-to-digital conversion and signal processing.

[0008] Optionally, the step of generating a fusion feature vector representing the reading and attention network based on the original cerebral blood oxygenation signal by calculating temporal waveform features and frontotemporal functional connectivity strength includes: Based on the HbO and HbR time-series data of each channel, the temporal characteristics reflecting the activation intensity and dynamics of the brain region are obtained by calculating the waveform centroid value, integral area and instantaneous slope. Based on the HbO signals of key channels in the left frontal and temporal lobes, the functional connectivity strength is quantified and calculated using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network. Based on all extracted temporal and functional connectivity features, a unified multi-dimensional fusion feature vector is generated for subsequent model input through concatenation and layer normalization operations.

[0009] Optionally, the step of processing and analyzing multi-dimensional fused feature vectors using a Transformer classification model to generate classification results identifying the risk level of dyslexia and comorbid attention deficit includes: Based on the fused feature vector, after embedding position encoding and modality encoding, it is input into a lightweight Transformer encoder for feature transformation and attention calculation; Based on the sequence output by the Transformer, the representation vector corresponding to its CLS token is extracted and processed by a linear layer and a Softmax function to generate probability distributions corresponding to the three risk subtypes. Based on the obtained classification probabilities, by determining the category corresponding to the highest probability, an objective classification result is finally generated: low risk of reading disability, high risk of reading disability, or high risk of comorbid ADHD.

[0010] Optionally, the step of acquiring light signals from multiple brain regions during the task at a preset sampling rate using a multi-channel dual-wavelength near-infrared light electrode integrated into a wearable structure, based on preset frontotemporal reading network and parietal attention network targets, includes: Based on the anatomical location of the frontotemporal reading network and the dorsal attention network, light sources and detectors are arranged at preset intervals to form a combination of light pole channels covering key brain regions. Near-infrared light is emitted to the scalp through photoelectric electrodes, and the intensity signal of the emitted light is received by the brain tissue. Based on the received light intensity signal, raw electrophysiological data reflecting changes in cortical blood oxygen concentration are obtained through analog-to-digital conversion and signal conditioning circuits.

[0011] Optionally, the step of obtaining temporal features reflecting the activation intensity and dynamics of brain regions by calculating the waveform centroid value, integral area, and instantaneous slope based on the HbO and HbR time-series data of each channel includes: Based on the time sequence of HbO and HbR concentration changes, the centroid value characteristic representing the overall temporal centroid of the cerebral blood flow response is obtained by calculating the weighted average position of the sequence on the time axis. Let the centroid value characteristic be The time corresponding to the sampling point is In time The concentration of HbO or HbR at that location is ,but: ; Based on the HbO and HbR concentration sequences, the integral area feature characterizing the overall energy of the cerebral blood flow response is obtained by calculating the area of ​​the region enclosed between the sequence curve and the time axis. Let the area of ​​integration be The time corresponding to the sampling point is In time The concentration of HbO or HbR at that location is The sampling time interval is The duration of the analysis task is ,but: ； Based on the HbO and HbR concentration sequences, the slope characteristics characterizing the velocity and direction of cerebral blood flow response are obtained by calculating the rate of change of the sequences. Let the slope be The time corresponding to the sampling point is In time The concentration of HbO or HbR at that location is ,but: .

[0012] Optionally, the step of inputting the fused feature vector, after embedding position encoding and modality encoding, into a lightweight Transformer encoder for feature transformation and attention calculation includes: Based on the physical meaning and source channels of different feature components in the fused feature vector, temporal features and functional connectivity features are distinguished by adding learnable modal embedding vectors to the feature vectors. Based on the order relationship of each element in the feature vector, spatial sequence information of the features is preserved by adding a position embedding vector to the feature vector; Based on the feature vectors with embedded information, the multi-head self-attention mechanism and feedforward neural network layer in the Transformer encoder are used to perform cross-feature and cross-brain region information interaction and deep abstraction to generate high-level task representations.

[0013] This application also provides a device for screening children's reading and attention deficit risks, the device comprising: The information acquisition module is used to acquire multi-channel raw cerebral blood oxygenation signals under task-induced conditions through a specific layout fNIRS array integrated into a wearable headband, based on a cognitive paradigm of rapid naming. The feature generation module is used to generate a fusion feature vector representing the reading and attention network by calculating the temporal waveform features and the frontotemporal functional connectivity strength based on the original cerebral blood oxygenation signal. The results output module is used to process and analyze multi-dimensional fused feature vectors through the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

[0014] Optionally, the feature generation module further includes: The temporal feature extraction module is used to obtain a temporal feature set reflecting the local activation intensity and dynamics of brain regions by calculating the waveform centroid value, integral area and slope based on the time series data of HbO and HbR concentrations in each channel. The brain network connectivity analysis module is used to quantify the functional connectivity strength of key channels in the left frontal and temporal lobes using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network. The feature fusion and normalization module is used to generate a unified and regular multi-dimensional fusion feature vector based on all extracted temporal and functional connectivity features through concatenation and layer normalization operations, for use by the classification model.

[0015] This application also provides an electronic device for implementing any of the described methods for screening for child reading and attention deficit risks, including: The processor is used to execute the complete computation and control process from controlling the presentation of stimuli in the fast naming paradigm, acquiring and processing multi-channel fNIRS signals, extracting temporal and functional connectivity features, to generating risk classification results through a lightweight Transformer model. The memory is used to store the paradigm scripts, classification model parameters, intermediate data during processing, and final classification results necessary to implement the complete process. The hardware components include a specifically laid-out fNIRS photoarray covering key reading and attention brain regions, a display and audio unit for presenting task stimuli, acquisition and conditioning circuitry for converting optical signals into digital signals, and a communication unit for data exchange.

[0016] The beneficial effects of this application are as follows: By integrating a minimalist multi-channel fNIRS array into an all-in-one wearable headband and adopting a standardized rapid naming cognitive paradigm, child-friendly and rapidly deployable objective brain function screening is achieved, significantly improving the accessibility and efficiency of screening in on-site or grassroots environments; by simultaneously calculating the temporal dynamic characteristics of brain blood oxygenation signals and the functional connectivity strength of the frontotemporal lobe brain network, and fusing them to generate multi-dimensional feature vectors, efficient extraction and representation of multi-scale and complementary information on reading and attention-related neural activities are achieved; and by feeding the fused feature vectors into a lightweight Transformer classification model for end-to-end processing, the risk level of reading disorders and their comorbid attention deficits is directly output, achieving high-precision and automated objective auxiliary diagnosis, effectively reducing reliance on subjective behavioral scales and professionals. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings required in the description of the embodiments or the prior art are briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a specific embodiment of this application shows a method for screening the risk of reading and attention deficit in children; Figure 2This is a block diagram of a device for screening the risk of reading and attention deficit in children, according to a specific embodiment of this application. Detailed Implementation

[0019] Various exemplary embodiments, features, and aspects of this application will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0021] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0022] Furthermore, to better illustrate this application, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that this application can be implemented without certain specific details. In some instances, methods, means, components, and circuits well-known to those skilled in the art have not been described in detail in order to highlight the main points of this application.

[0023] This application proposes a method for screening reading and attention deficit risk in children, aiming to address the problems of existing assessment methods being subjective, inefficient, and difficult to implement at the grassroots level. It combines a minimally targeted fNIRS hardware design, a standardized rapid naming cognitive paradigm, and deep learning-based multi-dimensional neural feature fusion analysis. Specifically, the device uses a specially laid-out multi-channel fNIRS array integrated into a wearable headband to collect brain blood oxygenation signals from the frontotemporal lobe reading network and parietal lobe attention network induced by a rapid naming task. Then, the algorithm simultaneously extracts temporal features characterizing local activation intensity and dynamics, as well as functional connectivity features reflecting inter-brain region collaboration, from the signals. These multi-scale features are fused to construct a unified representation vector. Finally, this fused feature vector is input into a lightweight Transformer classification model to achieve end-to-end automatic analysis and classification, directly outputting objective results of low-risk reading disorder, high-risk reading disorder, or high-risk comorbid ADHD. By miniaturizing hardware and standardizing paradigms, rapid screening can be completed in minutes. At the same time, the accuracy and objectivity of the screening are ensured by multi-dimensional neural feature fusion and lightweight deep learning models, providing an efficient, reliable, and portable solution for the early detection of large-scale children's reading development and attention problems.

[0024] Example 1 like Figure 1 The diagram shown is a flowchart of a method for screening the risk of reading and attention deficit in children according to an embodiment of this application. The method specifically includes the following: S100, based on a rapid naming cognitive paradigm, acquires multi-channel raw brain oxygenation signals induced by a task through a specially laid-out fNIRS array integrated into a wearable headband.

[0025] Specifically, the stimulation process is designed based on the standardized cognitive psychological paradigm of rapid naming. A fixed program script controls the standardized presentation and synchronous triggering of visual stimuli to induce target neural activity closely related to reading and attention functions. Based on the anatomical localization of the frontotemporal reading network and the parietal attention network, a minimally oriented near-infrared light source and detector of a specific wavelength are integrated at a preset spacing into a wearable headband to construct a minimally oriented fNIRS signal acquisition array targeting these key brain regions. This array emits near-infrared light to the scalp and receives the emitted light intensity after scattering by brain tissue. Then, through analog front-end conditioning and high-precision analog-to-digital conversion, digital light intensity signals reflecting changes in cortical blood oxygen dynamics are acquired. Finally, the dual-wavelength light intensity signals are calculated according to the modified Lambert-Beer law to reconstruct the original time-series data of the continuous changes in oxyhemoglobin and deoxyhemoglobin concentrations in each channel over time, laying the signal foundation for feature analysis.

[0026] S200, based on the original cerebral blood oxygenation signal, by calculating the temporal waveform features and the frontotemporal functional connectivity strength, a fusion feature vector representing the reading and attention network is generated.

[0027] Specifically, from the perspective of reflecting local brain region blood oxygenation dynamics, based on the time-series data of HbO and HbR concentrations in each channel, the centroid value representing the overall response time pattern is obtained by calculating the weighted average position of their waveforms on the time axis; the area under the concentration-time curve is obtained by calculating the area representing the overall response energy; and the slope value representing the response speed and direction is obtained by calculating the rate of change of the sequence in key periods. This constitutes a time-domain feature set characterizing the activation intensity and dynamics of brain regions. From the perspective of reflecting cross-brain region functional coordination, based on the HbO signals of corresponding channels in the core reading brain regions of the left frontal and temporal lobes, the synchronicity intensity of their neural activity during the task is quantified using methods such as wavelet coherence analysis, thereby obtaining functional connectivity features characterizing the collaborative working efficiency of the reading network. Finally, by concatenating and layer-normalizing the time-domain feature sets of all channels with the functional connectivity features of key channel pairs, a unified and regular multi-dimensional feature vector that integrates local activation and network coordination information and can be used as model input is generated.

[0028] S300, based on multi-dimensional fused feature vectors, is processed and analyzed using the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

[0029] Specifically, to adapt to the structural information of features processed and encoded by sequence models, based on the fused feature vector, positional encoding is embedded to preserve feature order, and learnable modal encoding is embedded to distinguish different feature subsets such as temporal and connectivity features. The embedded feature sequence is input into a lightweight Transformer encoder with optimized parameters. Through its multi-head self-attention mechanism, it automatically learns and models the complex nonlinear interactions and dependencies between features, achieving deep abstraction and context-aware representation transformation of cross-feature and cross-brain region information. Based on the feature sequence output by the encoder, the global aggregate representation corresponding to the dedicated classification token is extracted and processed by a fully connected layer and a Softmax function to map it into a probability distribution corresponding to different risk categories. Finally, by determining the category corresponding to the highest probability, objective classification results of low-risk dyslexia, high-risk dyslexia, or high-risk dyslexia comorbid with ADHD are generated, completing the end-to-end automated mapping from neural features to clinical decision support.

[0030] In summary, this application breaks away from the traditional subjective screening model that relies on behavioral scales and single EEG indicators, and constructs a complete objective data acquisition system from standardized cognitive evoked responses and targeted near-infrared optical acquisition to high-fidelity blood oxygenation signal reconstruction. By solidifying a rapid naming paradigm script, it ensures strict consistency in the temporal and content aspects of the cognitive task evoked process, providing repeatable standardized event anchors for neural responses. Through a wearable fNIRS array optimized based on the anatomical localization of reading and attention networks, it achieves efficient, targeted, and synchronous capture of neural blood oxygenation activity in core brain regions. Furthermore, based on the modified Lambert-Beer law, it achieves accurate reconstruction from raw optical signals to HbO / HbR concentration temporal data, laying a high-quality, directly computable physiological signal foundation for subsequent analysis. Secondly, addressing the challenge of comprehensively and quantitatively characterizing reading and attentional functional states from multi-channel brain blood oxygenation signals, it designs a multi-dimensional feature extraction and fusion framework that integrates local activation dynamics and cross-brain region functional synergy. By calculating the centroid value, integral area, and instantaneous slope of the signal waveforms in each channel, the activation characteristics of local brain regions were characterized from three dimensions: temporal pattern, response intensity, and rate of change. Wavelet coherence analysis was used to quantify the synchronicity of neural activity between key brain regions in the left frontal and temporal lobes, revealing the functional integration efficiency of the reading core circuit at the network level. Finally, the aforementioned temporal feature set and functional connectivity features were concatenated and normalized to generate a multi-dimensional fusion feature vector that can comprehensively and complementaryly represent the individual brain functional state, overcoming the limitation of single-type feature representation capabilities. Finally, an end-to-end automated classification decision-making link based on a lightweight deep learning model was constructed, realizing intelligent mapping from multi-dimensional neural features to specific risk levels. By embedding positional and modal encodings into the fusion features, they were adapted to the Transformer architecture while retaining the structured information of the features. Through the multi-head self-attention mechanism of the lightweight Transformer encoder, complex nonlinear interaction relationships between features were automatically discovered and modeled, achieving deep information abstraction. Finally, the global aggregated representation was mapped to three-class probabilities through the classification head, and objective risk level results were output according to the decision rules. This link achieves full automation and high objectivity in the screening process, providing a reliable technical closed loop for rapid and accurate brain function assisted screening.

[0031] As an optional implementation of this application, optionally, in step S100, based on the cognitive paradigm of rapid naming, multi-channel raw cerebral blood oxygenation signals induced by a task are acquired through a specific layout fNIRS array integrated into a wearable headband, including: 110. Following the rapid naming paradigm, the stimulus presentation module, controlled by a script embedded in the FPGA, sequentially presents a calibration blank screen and a digital matrix on the display screen to induce target cognitive operations.

[0032] Specifically, the device's main control unit, based on a standardized process using a rapid naming paradigm, employs a hard-core control script embedded in a field-programmable gate array (FPGA) to perform high-precision, low-latency timing scheduling of the visual stimulus presentation process. This process begins with the presentation of a calibration blank screen, whose core function is to allow the subject's visual system to adapt to the display environment and provide a stable baseline starting point for neural activity. Subsequently, the control logic drives the display screen to sequentially present a series of stimulus matrices composed of numbers. The presentation duration, interval, and combination sequence of numbers for each matrix strictly adhere to experimental psychology paradigms to systematically induce target cognitive operations related to speech retrieval, symbol recognition, and executive control. This hardware-based control method fundamentally ensures millisecond-level timing accuracy of stimulus presentation and absolute consistency with the task flow, eliminating timing jitter caused by software scheduling or operating system delays, and providing a reliable time anchor for the precise alignment of subsequent brain signals and cognitive events. The entire stimulus control process operates independently of the upper-level application software, forming a stable and repeatable cognitive task induction engine, which is a prerequisite for achieving objective and standardized brain function testing.

[0033] 120. Based on the preset frontotemporal reading network and parietal attention network targets, light signals from multiple brain regions are collected during the task using a multi-channel dual-wavelength near-infrared light pole integrated into a wearable structure at a preset sampling rate.

[0034] Specifically, to capture specific neural circuit activity induced by the task, a wearable near-infrared optical acquisition component designed based on anatomical targeting principles is employed. The core of this component is a flexible headband integrating a specific arrangement of light sources and detectors, its physical layout strictly optimized according to the cortical projection positions of the frontotemporal reading network (e.g., the left inferior frontal gyrus and middle temporal gyrus) and the parietal attention network (e.g., the internal parietal sulcus). By pairing light sources emitting near-infrared light of specific wavelengths with detectors receiving scattered light signals at a preset optimal optical distance, a series of effective measurement channels spanning the target brain regions are formed. During task execution, these light sources emit near-infrared light towards the scalp at a preset constant frequency. Photons enter the brain tissue via the scalp and skull, and during scattering, their intensity specifically attenuates due to changes in the concentration of oxyhemoglobin and deoxyhemoglobin in the cerebral cortex. The detector array synchronously receives the emitted light intensity signals scattered back from the brain tissue. These weak analog light intensity signals are transmitted in real time to the analog front end, and after a series of low-noise amplification and filtering adjustments, they are sent to a high-precision analog-to-digital converter circuit and converted into raw digital electrophysiological data streams. This enables direct, synchronous, multi-point optical measurement of blood oxygenation dynamics in preset target brain regions.

[0035] 130. Based on the acquired raw optical signal, the raw data of HbO and HbR concentration changes over time in each channel are obtained through analog-to-digital conversion and signal processing.

[0036] Specifically, the acquired raw digital light intensity signals undergo post-processing to convert them into time-series blood oxygen concentration data with clear physiological significance. This process first performs quality checks and preprocessing on the raw dual-wavelength (e.g., 730nm and 850nm) light intensity data received from each channel. This includes removing signal spikes caused by motion artifacts and potentially applying bandpass filtering to isolate specific frequency bands related to neural activity. Subsequently, the path-length corrected dual-wavelength light intensity changes are modeled and calculated. By solving a system of equations composed of the two wavelength light intensity changes, the relative changes in oxyhemoglobin and deoxyhemoglobin concentrations within the measurement region relative to the pre-task baseline can be calculated. Finally, two time-series concentration sequences—HbO and HbR—are output for each measurement channel, strictly synchronized in time and continuously varying in numerical value. These two sequences constitute the most direct signal basis for characterizing the neurovascular coupling response of each brain region and are the data source for all subsequent higher-order feature calculations.

[0037] As an optional embodiment of this application, optionally, in step S200, based on the original cerebral blood oxygenation signal, a fusion feature vector representing the reading and attention network is generated by calculating the temporal waveform features and the frontotemporal functional connectivity strength, including: 210. Based on the HbO and HbR time series data of each channel, the temporal characteristics reflecting the activation intensity and dynamics of the brain region are obtained by calculating the waveform centroid value, integral area and instantaneous slope.

[0038] Specifically, a set of core temporal features quantitatively characterizing the local brain region's blood oxygenation response pattern is extracted from the time-series data of HbO and HbR concentrations in each channel. This process first calculates the centroid value feature, which represents the overall temporal centroid of the blood flow response. This feature is obtained by calculating the weighted average position of the concentration sequence on the time axis, and its mathematical expression is as follows: ,in for Concentration value at time, This value, corresponding to the time frame, reflects the timing of the peak in the blood oxygenation response and is a key indicator for assessing the temporal characteristics of neural activation. Secondly, the integral area characteristic, representing the overall energy magnitude of the response, is calculated. This characteristic is obtained by calculating the area enclosed by the concentration-time curve and the baseline, mathematically approximated as... , The sampling interval, this value quantifies the net load of brain region blood oxygenation changes during the task and is correlated with the overall intensity of neural activity. Finally, a slope feature characterizing response velocity and direction is calculated. This feature is obtained by calculating the rate of change of the concentration sequence at specific key time points (such as the initial rise in the response), expressed as... These three features characterize the rate and trend of changes in blood oxygen concentration. They provide a comprehensive digital description of the activation intensity and dynamic characteristics of a single brain region from three dimensions: temporal distribution of response, energy accumulation, and rate of change.

[0039] 220. Based on the HbO signals of key channels in the left frontal and temporal lobes, the functional connectivity strength is quantified and calculated using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network.

[0040] Specifically, to assess the functional synergy between core brain regions in the reading process, this study focuses on analyzing the synchronization intensity of neural activity between key pathways in the left frontal and temporal lobes. HbO signals recorded in pathways located in the left inferior frontal gyrus (typically involved in speech processing and executive control) and key temporal lobe regions (such as the posterior middle temporal gyrus, involved in visual word recognition) were selected as the analysis objects. Wavelet coherence analysis, a time-frequency domain method, was used to quantify these two time series. This method can assess the phase consistency intensity of two signals at specific frequency components over time in a two-dimensional time-frequency plane. The calculated coherence coefficient is the functional connectivity strength index, ranging from 0 to 1. The closer the value is to 1, the higher the synchronization of neural activity between the two brain regions at a specific oscillation frequency during the task, and the stronger the functional connectivity. This connectivity feature directly reflects the efficiency of information transmission and collaborative work between different functional modules in the reading network from a neurophysiological perspective. It is an important network-level indicator for identifying the functional integration status of the reading network, complementing the time-domain features that only reflect local activation.

[0041] 230. Based on all extracted temporal and functional connectivity features, a unified multi-dimensional fusion feature vector is generated for subsequent model input through concatenation and layer normalization operations.

[0042] Specifically, after extracting temporal and functional connectivity features separately, they need to be integrated into a unified and standardized data structure for the classification model to process. First, feature concatenation is performed, connecting all temporal features from all channels (three features each for HbO and HbR for each channel) and the calculated functional connectivity features between all key channel pairs in a predetermined fixed order to form a long one-dimensional feature vector. This vector integrates information from multiple dimensions, including spatial (different brain regions), modal (local activation and network connectivity), and attribute (time, energy, rate, synchronicity). Subsequently, to avoid adverse effects on model training due to large differences in the dimensions and numerical ranges of different features, layer normalization is performed on the fused feature vector. This operation standardizes the vector using the mean and standard deviation of all its elements, making the processed feature vector data distribution close to zero mean and unit variance, thereby accelerating model convergence and improving its generalization performance. The multi-dimensional fused feature vector generated after concatenation and normalization is a comprehensive and standardized mathematical representation of an individual's reading and attention-related brain functional states in a standardized task.

[0043] As an optional implementation of this application, optionally, in step S300, based on the multi-dimensional fused feature vector, the data is processed and analyzed using a Transformer classification model to generate classification results identifying the risk level of reading disorders and comorbid attention deficit, including: 310. Based on the fused feature vector, after embedding position encoding and modality encoding, it is input into a lightweight Transformer encoder for feature transformation and attention calculation.

[0044] Specifically, to input the fused feature vectors into a Transformer-based model for deep analysis, they first need to be embedded and encoded to inject necessary structured information. This process involves adding two types of embedding vectors: learnable positional encoding, which assigns a unique positional embedding vector based on the feature's index in the vector, enabling the model to perceive and utilize the sequential relationship of features; and learnable modal encoding, which assigns corresponding modal embedding vectors to different subsets of features based on their physical origin, such as distinguishing between temporal centroid features, area features, and slope features originating from specific channels, and functional connectivity features originating from specific channel pairs, to clarify their semantic categories. After element-wise addition of the original feature vectors to these two embedding vectors, an input sequence rich in sequential and semantic information is obtained. Subsequently, this sequence is fed into a lightweight Transformer encoder with optimized parameters. The encoder, through its core multi-head self-attention mechanism, calculates the association weights between any two feature elements in the sequence, achieving global information interaction and weighted aggregation across different brain regions and feature types, and then performs a nonlinear transformation via a feedforward neural network. By stacking multiple layers of such a structure, the model can abstract and deeply integrate the input features layer by layer, and finally output a new sequence containing high-level semantic task representations.

[0045] 320. Based on the sequence output by the Transformer, the representation vector corresponding to its CLS token is extracted and processed by a linear layer and a Softmax function to generate probability distributions corresponding to the three risk subtypes.

[0046] Specifically, the feature sequence output by the Transformer encoder contains context-aware information from all input features after deep interaction and abstraction. To extract an aggregated representation for global classification from the sequence, a specific trainable vector, called a CLS token, is pre-inserted at the beginning of the initial input sequence. This token, along with all features (including positional and modal information), undergoes multiple layers of attention and transformation by the encoder. Its corresponding output vector is considered to encapsulate the global contextual information of the entire input sequence, making it suitable as a holistic representation of the task. Subsequently, the output vector corresponding to this CLS token is input into a classification head composed of fully connected layers. This fully connected layer maps the high-dimensional representation to a low-dimensional space equal to the number of target categories. Finally, the Softmax function is applied to normalize this three-dimensional vector, transforming it into a probability distribution. The three probability values ​​in this distribution correspond to the predicted probabilities of three preset risk subtypes: low risk of dyslexia, high risk of dyslexia, and high risk of dyslexia comorbid ADHD, with the sum of all probabilities being 1. This process completes the mapping from complex neural feature sequences to concise classification probabilities.

[0047] 330. Based on the obtained classification probabilities, by determining the category corresponding to the highest probability, an objective classification result is finally generated: low risk of reading disability, high risk of reading disability, or high risk of comorbid ADHD.

[0048] Specifically, based on the probability distribution output by the classification model, the final decision-making logic is executed to generate a clear classification result. The decision-making process involves selecting the risk category corresponding to the highest probability value from the three probability values ​​generated by the Softmax function. This risk category is then used as the system's final output for the test. For example, if the probability value corresponding to "high risk of dyslexia comorbid with ADHD" is the highest, then that category is output. The generated classification result is the final output of the entire system's technical process, objectively and quantitatively grading the risk of brain function patterns exhibited by subjects in a standardized rapid naming task. This result differs from traditional subjective behavioral scale scores; it is an objective indicator calculated based on multi-dimensional neurophysiological characteristics and deep learning models. It aims to provide professionals with an auxiliary, neural mechanism-based screening reference, ultimately achieving a closed loop for rapid and objective brain function risk assessment of dyslexia and its common comorbid attention deficit problems.

[0049] Example 2 As an application example of this application, the specific application is as follows: 1.1 Overall Approach A "rapid naming" paradigm was designed to induce millisecond-level responses in the frontotemporal reading network and the dorsal parietal attention network (DAN). Using a simplified 8-channel fNIRS optical array (2 middle temporal gyri, 2 inferior temporal gyri, 2 inferior frontal gyri, and 2 auxiliary motor areas), HbO / HbR dual-wavelength slopes (fNIRS mode) were acquired. Utilizing waveform features (centroid value, integral value) and frontotemporal functional connectivity values, a deep learning transformer model was used to output: the presence of reading impairment risk and the coexistence of attention deficits.

[0050] 1.2 Hardware Composition a) Signal acquisition module – Or 8-channel fNIRS (dual wavelength 760 / 850 nm, staggered light source-detector spacing of 30 mm); – 24-bit ADC, sampling rate 500 Hz (ERP) / 10 Hz (fNIRS), bandwidth 0.1–100 Hz.

[0051] b) Stimulus Presentation Module – 6.5-inch OLED, 60 Hz, brightness 300 cd / m², built-in 24-bit audio DAC; – The paradigm script is embedded in the FPGA, requiring no computer and running immediately upon power-on.

[0052] c) Host processing module – ARM Cortex-M7 600 MHz + NPU (0.5 TOPS), running a lightweight Transformer in real time (≤2.1 M parameters); – Upload raw data and results via Bluetooth 5.2; d) Wearable structure – The elastic sports headband is molded in one piece and features a quick-release optical buckle, allowing for easy wearing in 30 seconds; – The device weighs less than 85g, has a battery life of 4 hours, and supports Type-C fast charging for 15 minutes.

[0053] 1.3 Paradigm Flow (Time: 120 s) Step 1 Calibrate for 30 seconds: Look at a blank screen and establish an individual baseline; Step 2 Quick Naming Task (45 seconds): Present an 8-column, 6-row number matrix for 45 seconds, requiring children to "read it out quietly and quickly and accurately"; Step 3 Instantaneous detection for 45 seconds: gaze at a blank screen, and the system records the brain response; Step 4 is complete.

[0054] 1.4 Algorithm Core fNIRS: Calculates the centroid values, integral area, and slope of HbO and HbR, and quantifies the left frontotemporal functional connectivity strength using wavelet coherence; concatenates the 8-channel features into a 32-dimensional vector, feeds it into a lightweight Transformer with a "reading-attention dual-factor" architecture, and performs the following end-to-end joint loss: X = LayerNorm(Concat[ Centroid_HbO, Centroid_HbR, # 2×1 Center of Gravity Area_HbO, Area_HbR, # 2×1 Integral Area Slope_HbO, Slope_HbR, # 2×1 slope FC_left_frontotemporal # k×1 connection vector ]) → [1, 6+k, d_model] (1) Z = X + ModalityEmb + PosEmb (2) ŷ = Softmax(Linear(Transformer(Z)[:, 0])) # Get the CLS token and output the probabilities of the three classes (3) 1.5 Output and Explanation –SubType=1 → Indicates a low risk of reading difficulties; –SubType=2 → Indicates a high risk of reading difficulties; –SubType=3 → Indicates a high risk of dyslexia and ADHD; – The report automatically generates a QR code, which parents can scan with WeChat to read.

[0055] 1.6 Alternative Solutions Number of channels: can be reduced to 4 channels (1 middle temporal gyrus, 1 inferior temporal gyrus, 1 inferior frontal gyrus, 1 auxiliary motor area), with an accuracy decrease of ≤5%, still meeting the screening requirements; Paradigm Stimulation: Numbers can be replaced with Chinese characters and color, as long as the "rapid presentation" cognitive operation is maintained; Headband structure: Can be changed to "helmet style", with the optical position corresponding to the 10-20 system; Output indicators: You can output only the activation value or only the subtype classification, which still falls within the protection scope.

[0056] Example 3 Based on the same principle as the aforementioned methods, a device for screening the risk of reading and attention deficit in children is also proposed, see [link to relevant documentation]. Figure 2 An embodiment of this disclosure provides a device 100 for screening the risk of reading and attention deficit in children, comprising: Information acquisition module 110 is used to acquire multi-channel raw cerebral blood oxygenation signals under task-induced conditions by using a specific layout fNIRS array integrated into a wearable headband, based on a cognitive paradigm of rapid naming. The feature generation module 120 is used to generate a fusion feature vector representing the reading and attention network by calculating the temporal waveform features and the frontotemporal functional connectivity strength based on the original cerebral blood oxygenation signal. The output module 130 is used to process and analyze multi-dimensional fused feature vectors through the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

[0057] As an optional implementation of this application, the feature generation module 120 may further include: The temporal feature extraction module 121 is used to obtain a temporal feature set reflecting the local activation intensity and dynamics of the brain region by calculating the waveform centroid value, integral area and slope based on the time series data of HbO and HbR concentrations in each channel. The brain network connectivity analysis module 122 is used to quantify and calculate the functional connectivity strength of the key channels in the left frontal lobe and temporal lobe using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network. The feature fusion and normalization module 123 is used to generate a unified and regular multi-dimensional fusion feature vector based on all extracted temporal features and functional connectivity features through concatenation and layer normalization operations, for use by the classification model.

[0058] Obviously, those skilled in the art should understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the control methods described above. The modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, this application is not limited to any specific hardware and software combination.

[0059] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the control methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0060] Example 4 Furthermore, this application proposes an electronic device for implementing any of the described methods for screening for child reading and attention deficit risks, comprising: The processor is used to execute the complete computation and control process from controlling the presentation of stimuli in the fast naming paradigm, acquiring and processing multi-channel fNIRS signals, extracting temporal and functional connectivity features, to generating risk classification results through a lightweight Transformer model. The memory is used to store the paradigm scripts, classification model parameters, intermediate data during processing, and final classification results necessary to implement the complete process. The hardware components include a specifically laid-out fNIRS photoarray covering key reading and attention brain regions, a display and audio unit for presenting task stimuli, acquisition and conditioning circuitry for converting optical signals into digital signals, and a communication unit for data exchange.

[0061] The electronic device of this disclosure includes a processor and a memory for storing processor-executable instructions. The processor is configured to implement any of the preceding methods for screening for child reading and attention deficit risks when executing the executable instructions.

[0062] It should be noted that the number of processors can be one or more. Furthermore, the electronic device in this embodiment may also include input devices and output devices. The processor, memory, input devices, and output devices can be connected via a bus or other means, without specific limitations herein.

[0063] The memory, serving as a computer-readable storage medium for automated fault handling and self-learning methods for modules, can be used to store software programs, computer-executable programs, and various modules, such as the program or module corresponding to the child reading and attention deficit risk screening method in this disclosure embodiment. The processor executes various functional applications and data processing of the electronic device by running the software program or module stored in the memory.

[0064] The hardware components include: an fNIRS photoelectric array integrated into an elastic headband, with its light source and detectors arranged at preset intervals, and the photoelectric poles covering at least four key brain regions in the middle temporal gyrus, inferior temporal gyrus, inferior frontal gyrus, and supplementary motor area; a display and audio unit, including a display screen and an audio output unit, controlled by a processor to present visual and auditory task stimuli; an acquisition and conditioning circuit, including an analog front-end amplifier circuit, a filter circuit, and an analog-to-digital converter, used to convert the photoelectric signals into digital signals; and a communication unit used to upload raw data, feature vectors, or classification results to an external terminal.

[0065] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for screening children for risk of reading and attention deficits, comprising: include: Based on the cognitive paradigm of rapid naming, multi-channel raw cerebral blood oxygenation signals under task-induced conditions are obtained through a specially laid-out fNIRS array integrated into a wearable headband. Based on the original cerebral blood oxygenation signal, a fusion feature vector representing the reading and attention network is generated by calculating the temporal waveform features and the frontotemporal functional connectivity strength. Based on the multi-dimensional fused feature vectors, the data are processed and analyzed using the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

2. The method for risk screening of reading and attention deficit in children of claim 1, wherein, The method, based on a rapid naming cognitive paradigm, acquires multi-channel raw cerebral oxygenation signals under task-induced conditions through a specifically designed fNIRS array integrated into a wearable headband, including: According to the rapid naming paradigm, the stimulus presentation module is controlled by a script embedded in the FPGA to sequentially present a calibration blank screen and a digital matrix on the display screen to induce target cognitive operation. Based on the preset frontotemporal reading network and parietal attention network targets, light signals from multiple brain regions during the task are collected at a preset sampling rate using a multi-channel dual-wavelength near-infrared light pole integrated into a wearable structure. Based on the acquired raw optical signals, the raw data of HbO and HbR concentration changes over time in each channel are obtained through analog-to-digital conversion and signal processing.

3. The method for risk screening of reading and attention deficit in children as claimed in claim 1, wherein, The step involves generating a fusion feature vector representing the reading and attention networks based on the original cerebral blood oxygenation signal by calculating temporal waveform features and frontotemporal functional connectivity strength, including: Based on the HbO and HbR time-series data of each channel, the temporal characteristics reflecting the activation intensity and dynamics of the brain region are obtained by calculating the waveform centroid value, integral area and instantaneous slope. Based on the HbO signals of key channels in the left frontal and temporal lobes, the functional connectivity strength is quantified and calculated using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network. Based on all extracted temporal and functional connectivity features, a unified multi-dimensional fusion feature vector is generated for subsequent model input through concatenation and layer normalization operations.

4. The method for risk screening of reading and attention deficit in children of claim 1, wherein, The process involves fusing multi-dimensional feature vectors and analyzing them using a Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit, including: Based on the fused feature vector, after embedding position encoding and modality encoding, it is input into a lightweight Transformer encoder for feature transformation and attention calculation; Based on the sequence output by the Transformer, the representation vector corresponding to its CLS token is extracted and processed by a linear layer and a Softmax function to generate probability distributions corresponding to the three risk subtypes. Based on the obtained classification probabilities, by determining the category corresponding to the highest probability, an objective classification result is finally generated: low risk of reading disability, high risk of reading disability, or high risk of comorbid ADHD.

5. The method for risk screening of reading and attention deficit in children as claimed in claim 2, wherein, The method involves acquiring light signals from multiple brain regions during a task at a preset sampling rate, based on preset targets in the frontotemporal reading network and parietal attention network, using a multi-channel dual-wavelength near-infrared light electrode integrated into a wearable structure. This includes: Based on the anatomical location of the frontotemporal reading network and the dorsal attention network, light sources and detectors are arranged at preset intervals to form a combination of light pole channels covering key brain regions. Near-infrared light is emitted to the scalp through photoelectric electrodes, and the intensity signal of the emitted light is received by the brain tissue. Based on the received light intensity signal, raw electrophysiological data reflecting changes in cortical blood oxygen concentration are obtained through analog-to-digital conversion and signal conditioning circuits.

6. The method for risk screening of reading and attention deficit in children as claimed in claim 3, wherein, The process involves obtaining temporal characteristics reflecting the activation intensity and dynamics of brain regions by calculating the waveform centroid value, integral area, and instantaneous slope based on the HbO and HbR time-series data of each channel. This includes: Based on the time sequence of HbO and HbR concentration changes, the centroid value characteristic representing the overall temporal centroid of the cerebral blood flow response is obtained by calculating the weighted average position of the sequence on the time axis. Let the center of gravity value feature be , the time corresponding to the sampling point be , the HbO or HbR concentration value at the time be , then: ； Based on the HbO and HbR concentration sequences, the integral area feature characterizing the overall energy of the cerebral blood flow response is obtained by calculating the area of ​​the region enclosed between the sequence curve and the time axis. Let the integral area be , the time corresponding to the sampling point be , the HbO or HbR concentration value at time be , the sampling time interval be , and the analysis task period length be , then: ； Based on the HbO and HbR concentration sequences, the slope characteristics characterizing the velocity and direction of cerebral blood flow response are obtained by calculating the rate of change of the sequences. Let the slope be The time corresponding to the sampling point is In time The concentration of HbO or HbR at that location is ,but: 。 7. The method for screening children's reading and attention deficit risks as described in claim 4, characterized in that, The step of embedding positional encoding and modality encoding based on the fused feature vector, and then inputting it into a lightweight Transformer encoder for feature transformation and attention calculation includes: Based on the physical meaning and source channels of different feature components in the fused feature vector, temporal features and functional connectivity features are distinguished by adding learnable modal embedding vectors to the feature vectors. Based on the order relationship of each element in the feature vector, spatial sequence information of the features is preserved by adding a position embedding vector to the feature vector; Based on the feature vectors with embedded information, the multi-head self-attention mechanism and feedforward neural network layer in the Transformer encoder are used to perform cross-feature and cross-brain region information interaction and deep abstraction to generate high-level task representations.

8. A device for screening the risk of reading and attention deficit in children, the device comprising: The information acquisition module is used to acquire multi-channel raw cerebral blood oxygenation signals under task-induced conditions through a specific layout fNIRS array integrated into a wearable headband, based on a cognitive paradigm of rapid naming. The feature generation module is used to generate a fusion feature vector representing the reading and attention network by calculating the temporal waveform features and the frontotemporal functional connectivity strength based on the original cerebral blood oxygenation signal. The results output module is used to process and analyze multi-dimensional fused feature vectors through the Transformer classification model to generate classification results that identify the risk level of reading disorders and comorbid attention deficit.

9. The device for screening the risk of reading and attention deficit in children according to claim 8, wherein the feature generation module further comprises: The temporal feature extraction module is used to obtain a temporal feature set reflecting the local activation intensity and dynamics of brain regions by calculating the waveform centroid value, integral area and slope based on the time series data of HbO and HbR concentrations in each channel. The brain network connectivity analysis module is used to quantify the functional connectivity strength of key channels in the left frontal and temporal lobes using wavelet coherence analysis to obtain connectivity features that reflect the collaborative work of the reading network. The feature fusion and normalization module is used to generate a unified and regular multi-dimensional fusion feature vector based on all extracted temporal and functional connectivity features through concatenation and layer normalization operations, for use by the classification model.

10. An electronic device for implementing the method for screening for reading and attention deficit risks in children as described in any one of claims 1 to 7, comprising: The processor is used to execute the complete computation and control process from controlling the presentation of fast naming paradigm stimuli, acquiring and processing multi-channel fNIRS signals, extracting temporal and functional connectivity features, to generating risk classification results through a lightweight Transformer model. The memory is used to store the paradigm scripts, classification model parameters, intermediate data during processing, and final classification results necessary to implement the complete process. The hardware components include a specifically laid-out fNIRS photoarray covering key reading and attention brain regions, a display and audio unit for presenting task stimuli, acquisition and conditioning circuitry for converting optical signals into digital signals, and a communication unit for data exchange.

Images