Integrated Neurocognitive and Reading Biomarkers for Early Detection of Dyslexia Risk in Education

Approximately 5–10% of students experience dyslexia, a neurodevelopmental reading disorder, often identified only after prolonged difficulties in reading development. Standard screening approaches mainly focus on observable outcomes; such as decoding and fluency (1) which may not fully reflect underlying cognitive processes involved in reading acquisition. To address this limitation, performance was examined using a digital system that integrates eye-movement patterns, reading time, and comprehension measures to build a profile of reading behaviour during tasks. The system was implemented within SMCDSB to examine variation in classroom reading performance. Results indicated that combining timing, comprehension, and gaze measures produces distinct reading behaviour patterns across students. Early identification of these patterns may support earlier educational intervention for at-risk learners. Research in dyslexia suggests that early intervention can strengthen neural systems involved in reading (2), which supports that earlier detection may improve both the timing and effectiveness of educational support for struggling readers.

Why Dyslexia?

Dyslexia is a brain-based (neurodevelopmental) reading disorder characterized by persistent impairments in word recognition, decoding, spelling, and reading fluency despite adequate intelligence and educational exposure [4,2]. It affects approximately 5–10% of students [1] and comprises a substantial proportion of identified learning disabilities [6]. Delayed identification is associated with cumulative academic deficits and adverse psychosocial outcomes including reduced academic self-efficacy, increased risk of anxiety, and low self-esteem [7]. Students with dyslexia often have atypical development within reading networks of the brain, including reduced bilateral occipitotemporal and left temporo-parietal activations, which support phonological processing and the visual word form area supporting rapid orthographic recognition [8,9]. Early targeted interventions promote neuroplastic reorganization in reading networks, improving long-term reading acquisition outcomes [10].

Approach - NORA

The Neuro-Ocular Reading Analyzer (NORA) was developed and piloted as a classroom-based digital screening tool for reading performance. During standardized reading tasks, NORA records the following:

Reading time

Comprehension accuracy

Self-report responses

Eye-movement dynamics (fixations, saccades, and gaze patterns)

These measures are combined to create a profile of reading behaviour and estimate relative reading risk.

Objectives:

Examine whether reading-time and eye-movement patterns differ in students at varying levels of reading performance.

Evaluate the feasibility of integrating NORA into a classroom-based screening workflow for early identification and educational referral.

Determine whether a classroom-based digital screening system can be realistically used by teachers without disrupting normal instruction.

Methods & Development Process (NORA: Neuro-Ocular Reading Analyzer)

The NORA system was developed as a classroom-based screening prototype to evaluate reading behaviour across grade bands (Grades 3–12), with the intended implementation focused on Grade 3 as a critical early screening. The objective was to integrate observable indicators of reading behaviour into a unified, non-diagnostic index, including reading time, comprehension accuracy, self-reported reading effort, and webcam-based ocular gaze coordinates.

Background research

Background research used peer-reviewed literature on dyslexia and reading fluency, informed by models of reading comprehension and cognitive load theory.

Design process:

The prototype was built as a structured web application that guided the user through stages: grade level selection, ethical consent acknowledgment, timed reading, comprehension assessment, self-reflection survey, and automated scoring. Passage difficulty was divided into four grade bands (1–3, 4–6, 7–9, and 10–12) to control developmental level and linguistic complexity.

Data collection:

Each participant’s data included reading duration, 15 multiple-choice comprehension responses, and 15-item Likert-scale self-assessment responses. A webcam module recorded gaze coordinates to estimate visual attention distribution; however, this data was treated as supplementary due to variability in hardware and external conditions.

Variables:

Independent variable: selected reading level (grade band passage difficulty)

Dependent variables: comprehension accuracy, reading time, self-reported measures, and gaze dispersion (screen-coordinate variability)

Controlled elements: fixed passage texts, standardized question sets, and interface

Controls & limitations:

Environmental conditions such as lighting, device type, and screen size were not fully controlled, which may have affected eye-tracking and timing accuracy. Additionally, self-reported responses may be influenced by subjective bias.

Testing procedure:

The system was tested iteratively through internal trials to validate workflow logic, data storage, and scoring functionality using a Supabase backend.

The system was designed for rapid classroom administration (10–15 minutes), requiring minimal setup to ensure feasibility within standard instructional time.

Prototype Performance

The Neuro-Ocular Reading Analyzer (NORA) was evaluated using prototype testing across a range of reading levels with standardized reading passages, webcam-based gaze estimation, and computational processing of behavioural data. The system generated structured concern-level outputs immediately after task completion, supporting use within time-constrained classroom screening contexts. Repeated internal trials under consistent application conditions produced stable outputs; however, findings are constrained by the use of webcam-based gaze estimation rather than calibrated hardware eye-tracking.

Reading Performance Trends

Four primary variables were analyzed: reading time, comprehension accuracy, gaze-based positional variability metrics, and self-reported cognitive load. Higher-performing readers exhibited faster reading times, higher comprehension accuracy, more stable gaze dispersion patterns, and lower cognitive load ratings. Lower-performing readers showed slower reading times, reduced comprehension accuracy, increased variability in gaze position, less stable gaze patterns across the reading interval, and elevated self-reported difficulty. These patterns suggest that struggling readers require more time and attentional effort to process text efficiently. Trend analysis showed an inverse relationship between reading time and comprehension performance within the collected dataset, with longer reading times associated with lower comprehension scores across repeated trials.

Machine Learning Results

The current prototype uses a rule-based weighted scoring model rather than machine learning. This approach was selected to maintain transparency and interpretability during early-stage testing. Future work will evaluate supervised models following expansion of a validated dataset.

Statistical Analysis

Observed trends suggested relationships between gaze dispersion and reading performance. Increased variability in gaze position and longer reading times were associated with lower comprehension outcomes. Reading time alone demonstrated weaker predictive value due to variability in accuracy across participants. Basic performance comparisons were examined across variables. Integration of multiple behavioural measures produced more informative screening outputs than single-variable measures alone, supporting the use of combined indicators for screening applications.

Screening Output

The NORA system generated structured screening classifications categorizing participants into five-level concern classifications ranging from low to very high based on weighted scoring outputs. This tiered framework was designed to enhance interpretability in educational contexts and support early identification of students who may benefit from additional reading support. Outputs were generated in a standardized format to ensure consistency across participants and reduce interpretive bias. The system is intended strictly for screening purposes and not for clinical diagnosis.

Limitations & Future Improvements

This study is limited by a restricted sample size and reliance on webcam-based gaze estimation rather than calibrated hardware-based eye-tracking systems, which constrains measurement precision and external validity. Environmental factors, including lighting conditions, device variability, and screen size, were not controlled and may have influenced performance metrics among participants. Self-reported measures may also introduce subjective bias. Future work should incorporate larger and more diverse samples, validated eye-tracking hardware, and refinement of the scoring framework to improve generalizability across educational contexts. Additional improvements include longitudinal validation and improved calibration of passage difficulty across developmental reading levels.

Discussion & Implications:

The findings indicate that integrating behavioural indicators, including gaze-based metrics, reading time, comprehension accuracy, and self-reported reading effort, provides a more comprehensive representation of reading variability than outcome-only screening approaches. Unlike conventional assessments emphasizing accuracy or fluency, the NORA framework captures outcomes and reading processes, enabling improved differentiation of reading profiles.

Variation in gaze dispersion, reading time, and comprehension performance corresponded with differences in reading efficiency and consistency. These patterns suggest reading difficulty is reflected in outcomes and temporal and attentional characteristics of reading behaviour.

The primary contribution is not the tool itself, but its potential integration into a school-based early detection protocol. NORA can be administered in a classroom setting, generating structured concern-level outputs that support teacher interpretation, early identification, and referral decisions. Students identified as moderate or high concern may be reviewed alongside classroom performance and referred to literacy support or further assessment.

While evaluated across multiple grade levels, the primary application is early-stage screening, with Grade 3 representing a key developmental window.

These findings align with dyslexia research indicating reading differences arise from interactions across cognitive and perceptual systems. Early identification may improve access to targeted instructional support during critical developmental periods.

Conclusion:

The NORA framework demonstrates the feasibility of combining behavioural, gaze-based, and self-report measures into a unified screening approach for characterizing reading variability in educational settings. The results suggest that multi-dimensional assessment may improve the identification of students at risk for reading difficulties, supporting earlier and more targeted educational intervention.

Future & Improvements:

Future work will focus on improving the validity and applicability of NORA as a classroom-based screening protocol.

The next phase will prioritize a Grade 3 pilot implementation to evaluate feasibility. This includes standardized administration and teacher guidance for interpreting outputs. Students identified as moderate or high concern may be reviewed alongside classroom performance and referred to literacy support or further assessment.

Further validation will involve larger populations to improve generalizability. Longitudinal studies will assess predictive validity by examining relationships between early NORA scores and later reading outcomes.

Technical improvements will refine oculomotor calibration and scoring to improve consistency.

Acknowledgements:

I would like to thank Maria Correa (University of Guelph, Computer Science) for her sustained mentorship and technical guidance throughout the development of the NORA system. Her support in programming, debugging, and refining the web-based implementation was critical to building a functional and reliable digital screening tool, and significantly strengthened the design and stability of the system.

I would also like to thank Emily Goodson for her detailed feedback, structured suggestions, and ongoing support throughout the CWSF process, which improved the clarity, organization, and overall presentation quality of this project.

I would also like to thank Bianca Bono (Neuroscience, Carleton University) for her insightful feedback and contributions to refining the scientific clarity of this project.

I would like to thank the Simcoe Muskoka Catholic District School Board Research Advisory Committee for granting approval, enabling ethical classroom-based implementation during the 2025–2026 school year.

References

[1] Al-Lamki, L. (2012). Dyslexia: Its impact on the individual, parents and Society. Sultan Qaboos University medical journal. https://pmc.ncbi.nlm.nih.gov/articles/PMC3529660/

[2] Testing and evaluation. International Dyslexia Association. (2015, June 11). https://dyslexiaida.org/testing-and-evaluation/

[3] Munzer, T., Hussain, K., & Soares, N. (2020). Dyslexia: Neurobiology, clinical

features, Evaluation and Management. Translational pediatrics. https://pmc.ncbi.nlm.nih.gov/articles/PMC7082242/

[4] FragaGonzález, G., Karipidis, I. I., & Tijms, J. (2018). Dyslexia as a neurodevelopmental disorder and what makes it different from a chess disorder. Brain sciences. https://pmc.ncbi.nlm.nih.gov/articles/PMC6209961/

[5] Asadollahpour, F., Mazaheri, S., Shahouzaei, N., & Nakhaei, M. A. (2025). Prevalence of dyslexia and its relationship with decoding and comprehension in Persian-speaking elementary students. Indian journal of psychological medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC12592114/

[6] Dyslexia basics. (2020, March 10). International Dyslexia Association. Retrieved from: https://dyslexiaida.org/dyslexia-basics/

[7] Molly Hagan. (2026, January 21). Learning disabilities and depression. Child Mind Institute. Retrieved from:

https://childmind.org/article/learning-disabilities-and-depression/

[8] Norton, E. S., Beach, S. D., & Gabrieli, J. D. E. (2015). Neurobiology of dyslexia. Current opinion in neurobiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC4293303/

[9] Martin, L., Durisko, C., Moore, M. W., Coutanche, M. N., Chen, D., & Fiez, J. A. (2019). The VWFA is the home of orthographic learning when houses are used as letters. eNeuro. https://pmc.ncbi.nlm.nih.gov/articles/PMC6378324/

[10] Dorofeeva, S. V., & Дорофеева С. (2023). Neuroplasicity and the developmental dyslexia intervention. Genes & Cells. https://genescells.ru/2313-1829/article/view/623418

[11] Social and emotional problems related to dyslexia. (2023, March 12). International Dyslexia Association. Retrieved from: https://dyslexiaida.org/social-emotional/

[12] Dyslexia facts. (2021, February 1). Decoding Dyslexia Ontario. Retrieved from: https://decodingdyslexiaon.org/dyslexia-facts/#:~:text=Dyslexia%20is%20typically%20diagnosed%20or,a%20learning%20disability%20are%20dyslexic.

[13] BBC. (2017, February 23). Teenage reading ages “slip behind by up to three years.” BBC News. Retrieved from: https://www.bbc.com/news/uk-wales-39063304#:~:text=Reading%20ages%20of%20teenagers%20slip,that%20literacy%20skills%20help%20train.

[14] Bazen, L., van den Boer, M., de Jong, P. F., & de Bree, E. H. (2020). Early and late diagnosed dyslexia in secondary school: Performance on literacy skills and cognitive correlates. Dyslexia

https://pmc.ncbi.nlm.nih.gov/articles/PMC7687086/

[15] Sam McVancel. (2022, October 25). The intersection of dyslexia, struggles with reading, and mental health. Iowa Reading Research Center - The University of Iowa. Retrieved from: https://irrc.education.uiowa.edu/blog/2022/10/intersection-dyslexia-struggles-reading-and-mental-health

[16] Brain regions that make up the reading network and show consistent... | download scientific diagram. (n.d.-a). https://www.researchgate.net/figure/Brain-regions-that-make-up-the-reading-network-and-show-consistent-functional-and_fig1_328824514

[17] Hoeft, F., Hernandez, A., McMillon, G., Taylor-Hill, H., Martindale, J. L., Meyler, A., Keller, T. A., Siok, W. T., Deutsch, G. K., Just, M. A., Whitfield-Gabrieli, S., & Gabrieli, J. D. E. (2006). Neural basis of dyslexia: A comparison between dyslexic and nondyslexic children equated for reading ability. The Journal of neuroscience : https://pmc.ncbi.nlm.nih.gov/articles/PMC6674758/

[18] Tomaz Da Silva, L., Esper, N. B., Ruiz, D. D., Meneguzzi, F., & Buchweitz, A. (2021). Visual explanation for identification of the brain bases for developmental dyslexia on fMRI Data. Frontiers in computational neuroscience. https://pmc.ncbi.nlm.nih.gov/articles/PMC8458961/

[19] Brunyé, T. T., Drew, T., Weaver, D. L., & Elmore, J. G. (2019). A review of eye tracking for understanding and improving diagnostic interpretation. Cognitive research: principles and implications. https://pmc.ncbi.nlm.nih.gov/articles/PMC6515770/

AI or Large Language Model-generated text:

OpenAI. (2026). ChatGPT (Apr 30 version) [AI image generation model].

https://chat.openai.com/chat