Identifying reading disabilities through eye movements: a validation study using Lexplore and AI-driven technology

Abstract

This study explores the potential of eye-tracking technology combined with artificial intelligence (AI) to identify early reading disabilities (RD). Findings indicate significant differences in fixation patterns between elementary students with and without RD, suggesting eye-tracking as a promising screening tool for classroom use. Students’ reading failure is a systemic problem across the United States. Early reading interventions show promise in addressing reading disabilities (RD); however, traditional measures are costly, time-consuming, and unreliable. Recent researchers conclude that AI tools, such as eye-tracking, could detect RD for earlier intervention. This study investigated the potential impact of eye-tracking data during a Lexplore reading screening to determine if differences existed between the (a) average fixation time and (b) proportions of fixations to total stimuli duration while reading with 12 elementary students with RD and 17 students without RD. Although Lexplore incorporates AI and previously trained machine learning algorithms, the present study's goals were not to train, modify, or validate AI models. Instead, researchers analyzed pre-generated eye tracking output using traditional statistical methods to examine group differences regarding reading difficulties. This study showed statistically significant differences between groups in both average and proportions of eye fixations. The findings from this exploratory study indicate a need for further investigations regarding eye-tracking devices to screen, identify, and monitor elementary students potentially at risk for RD.

Introduction

Often considered the foundation for learning, reading is a fundamental skill critical for success in all aspects of life (Moats, 2020; Schoenbach et al., 2023). Struggling elementary readers are at higher risk of dropping out of high school, often suffer from low self-esteem, and have minimal academic achievement (Kaniuka, 2010). However, with appropriate intervention and support, struggling readers can develop essential literacy skills (Van Der Kleij et al., 2019). Identifying deficits impeding student reading progress as early as possible allows for targeted intervention (Kaye et al., 2022).

To address reading failure, public schools typically use a tiered model known as Response to Intervention (RTI) or Multi-Tiered Systems of Support (MTSS). These models require teams to identify research-based instructional interventions, progress monitoring, and data-based decision-making related to reading deficits (Fuchs and Fuchs, 2001; NRCLD, 2003). Identifying reading disabilities (RD) is a crucial first step in the RTI process (Catts et al., 2009; Vaughn and Fletcher, 2020), yet current identification tools are often deemed unreliable (Davis et al., 2007; Glover and Albers, 2007; Odegard et al., 2020; Petscher et al., 2019; Ritchie and Speece, 2004). Universal reading screenings work to predict students at substantial risk for reading failure, but the major flaw is that many may under- or over-identify cases (Glover and Albers, 2007; Odegard et al., 2020). Screenings also usurp actual instruction time, often overlook students in need, or provide unnecessary referrals to special education (Fuchs et al., 2012; Vaughn and Wanzek, 2014) dysl.

Approximately 10% to 15% of U.S. students have an RD (Moats and Dakin, 2008; Shaywitz and Shaywitz, 2008), which is defined as one who may or may not have a cognitive or developmental delay with symptoms that affect other aspects of life, including reading (5th ed.; DSM-5; American Psychiatric Association, 2013). Despite standard IQs and adequate educational opportunities, a person with RD has significant difficulty in acquiring and using reading skills, including difficulties in word recognition, decoding, comprehension (Adams and Wilson, 2022), or spelling (Pugh et al., 2001).

Using data from eye-tracking during the act of reading reveal the larger processes of reading, integrating the visual (Jimenez and Meyer, 2016), cognitive, and linguistic processes (Shaywitz et al., 2006). Obviously, the visual system plays a crucial role in reading, as readers rely heavily on their eyes to perceive and process written text (Stein, 2014). Eye movements during reading, such as fixations and saccades, are important indicators of cognitive processing (Rayner, 2009; Vidyasagar, 2013). Understanding the relationship between visual processing and reading is critical for improving instruction and developing effective interventions for individuals with reading difficulties (Lobier et al., 2013).

Eye-tracking technology, combined with artificial intelligence (AI) and previously established machine learning software programs, shows promise in streamlining the identification of RD (Benfatto et al., 2016; Gran Ekstrand et al., 2021). Eye-tracking can provide valuable information about how readers process text, including how they allocate attention, identify, and distinguish letters and words, and comprehend text (Raney et al., 2014). By analyzing eye movements, AI and machine learning algorithms can identify unique patterns associated with RDs, allowing for accurate and efficient identification of individuals requiring additional support or interventions (Sarker, 2021). Additionally, eye-tracking technologies provide real-time feedback and progress monitoring, allowing educators to adapt and track interventions over time (Benfatto, 2021).

According to the National Institute of Child Health and Human Development (2020), interventions for struggling readers should begin as early, ideally in kindergarten or first grade. Research shows that unaddressed reading difficulties persist with negative impacts on academic and social development (Torgesen et al., 1999). With a short timeline to address reading issues for maximum results (Finn et al., 2001; Snowling et al., 2007; Vaughn and Fletcher, 2020), traditional reading screenings for students at risk are time-consuming, often render unreliable results, and miss the critical timeline (Davis et al., 2007; Odegard et al., 2020; Petscher et al., 2019; Ritchie and Speece, 2004). Recent research (Alqahtani, 2025) supports early identification of RD before students even advance to fluency. Furthermore, cumbersome screening can lead to many undiagnosed RD students, especially for those living in poverty (Pugh et al., 2001).

In a deficit-based approach, students are only identified as having an RD when inadequacies persist after years and as long-term deficits become apparent (McGrath et al., 2010). First, identifying a student is at risk for RD begins with a universal reading screening assessment (Johnson et al., 2010). Davis and colleagues’ seminal work (2007) reviewed classroom universal reading screening methods and concluded no one screening method is best. This documented lack of consensus is compounded by the fact that individual reading screenings take hours due to multiple measures within one screening (Davis et al., 2007; Ritchie and Speece, 2004). Yet, researchers Benfatto et al. (2016) found that an individual eye-tracking reading screening assessment on a computer, as opposed to traditional screenings, was accurate and expedient. Therefore, this study sought to explore the gap in the existing literature on using eye-tracking to detect RD in students in the U.S.

Literature review

Reading screenings’ alignment to eye movements

Eye movement metrics allow researchers to analyze gaze data during reading quantitatively (Bixler and Mello, 2016). The main types of eye movements measured are saccades, regressions, fixations, smooth pursuit movements, vergence movements, and vestibular-ocular movements (Purves, 2008). Fixation is the most common eye movement and is defined as the time of one specific stopping of the eye on a line of print. During fixations, the eyes stop and hold the field of vision in one place (Types of Eye Movements, 2022), fixating on one area of interest (AOI) while the gaze remains (Rayner, 2009). In contrast with fixations, Purves (2008) describes saccades as when the eyes move quickly and abruptly, changing their place of focus between one AOI and another. During a saccadic movement, both eyes simultaneously move in the same direction. Saccades are another common eye-tracking metric measured during reading research using machine learning.

Eye-tracking research

In 1989, the longitudinal Kronoberg Reading Development Research Project emerged and followed a cohort of 185 s graders for over 20 years until 2010 (Benfatto, 2021). Fouganthine (2012) originally aimed to investigate the relationship between eye movements and visual attention during expository, narrative, and instructional texts. The research found that readers tend to fixate frequently on key content words important for comprehension and make longer fixations on difficult, unknown words with complex spelling or syntax. Overall, the Kronoberg study provided valuable insights into the relationships across eye-tracking and reading cognitive processes.

In 2013, Benfatto developed the first machine-learning model in Sweden using eye-tracking data from the Kronoberg study. Benfatto et al. (2016) conducted a study in Sweden using machine learning algorithms to process eye-tracking data, which yielded 86% reliability and 97% accuracy in detecting RDs. Then, Seimyr and Benfatto (2017) created Lexplore by incorporating eye-tracking, research-based machine learning algorithms, assessment tools, deficit-based intervention techniques, and progress monitoring. The current study utilizes Lexplore assessment, because it incorporates eye-tracking data as a validated measure to predict RD (Lexplore, n.d.).

Benfatto et al. (2016) highlighted gaps in the field related to eye-tracking to detect RD in children as opposed to adults. One challenge in RD identification is limited research (Caldani et al., 2020). Existing research on eye-tracking primarily centers around adult studies of eye-tracking while reading (Kim and Wiseheart, 2017; Lukasova et al., 2018; McChesney and Bond, 2019). Few studies specifically examine using eye-tracking for the early identification of elementary students (Wanzek et al., 2018) or to detect differences between students with RD and those without (Cui et al., 2020). Studies with students have relied on outdated data recordings as secondary sources, which do not adequately capture the current landscape of RD (Vajs et al., 2023). This gap further points to needed research directly investigating eye-tracking in elementary students and its relationship to RD (Hessel et al., 2021).

Perhaps unrealized, eye-tracking research in the literacy field began well over a century ago (Huey, 1908), yet researchers are still developing applications for this technology, in particular, to detect RD in elementary students (Benfatto et al., 2016). According to Tobii Connect (2022), capturing eye movements and eye-tracking data can offer insight into reading behavior (Morrison et al., 2011).

Eye-tracking data

Researchers from the following systematic literature reviews examined eye-tracking data during reading to gain a comprehensive understanding of identifying persons who are potentially at risk of having RD (Rayner et al., 2006). By examining the relationships between eye movement data while reading, researchers identified potential areas of weakness or atypical reading behaviors indicative of RD (Cui et al., 2020; Prabha and Bhargan, 2020; Raatikainet et al., 2021). Identifying those potentially at risk of having RDs using eye movement metrics involves assessing and analyzing specific aspects of eye movement changes during reading and performance (Benfatto, 2013; Benfatto et al., 2016; Kim and Wiseheart, 2017; McChesney and Bond, 2019; Rello and Bastelaros, 2015). Eye movement patterns during reading provide an indirect but sensitive indicator of cognitive load, as increased fixation duration and frequency often reflect greater processing demands and inefficient integration of visual and linguistic information. Prior research demonstrates that children with reading difficulties exhibit eye movement behaviors associated with heightened cognitive load during reading tasks, supporting the use of eye tracking as a tool for identifying reading disabilities (Ozeri-Rotstain et al., 2020).

Eye-tracking studies machine learning and artificial intelligence

Young elementary students with RD often exhibit different eye movement patterns compared to typically developing readers. For example, they may have longer fixation durations, an increased number of fixations, and more regressions during reading (Sims and Conati, 2020; Sluis and Broek, 2023), suggesting challenges in processing and integrating text information efficiently. Further, eye-tracking studies show that primary students with RD may have difficulties in accurately fixating and recognizing words (Alqahtani, 2025). They may display shorter fixations on target words, more frequent total fixations, increased regressions, and increased attention to non-target areas (e.g., surrounding letters; Antúnez et al., 2022). Students with RD may allocate their visual attention differently compared to their peers without disabilities (Bundesen, 1990). They may show reduced sensitivity to orthographic and phonological information, resulting in less efficient visual search and difficulty in attending to relevant linguistic cues (Jimenez and Meyer, 2016; Shaywitz et al., 2006).

Simultaneously, the field of AI has grown significantly in the past decade (Zhang and Aslan 2021), creating a potential for use with machine learning for RD identification and remediation using eye-tracking (García Chimeno et al., 2014; Prado et al., 2007). Few studies explore eye-tracking data within software programs that include established machine-learning algorithms to identify elementary students potentially at risk for RD (Riedl, 2019).

Presently, limited research exists on how eye-tracking may identify elementary students potentially at risk of RD (Benfatto et al., 2016; Cui et al., 2020; Gran Ekstrand et al., 2021; Prabha and Bhargavi, 2020; Raatikainen et al., 2021). The current literature is void of eye-tracking to identify elementary students potentially at risk for RD research in the U.S (Caldani et al., 2020).

Most salient to the current study, Kim and Wiseheart (2017) analyzed cognitive processing and ontological deficits with graph reading by individuals with and without dyslexia, a type of RD. Dyslexia is a phonological processing difficulty, not a visual disorder. In general, individuals with dyslexia struggle to decipher orthographic tasks such as reading graphs, legends, spelling, or labels, so researching eye-tracking during the reading of orthographic-based tasks with 80 adults (29 with dyslexia, 48 without) commenced. Results showed readers with dyslexia had delayed time processing bar graphs. However, subjectts from both groups read graphs that contained no words with similar times.

In Spain, Rello and Bastelaros (2015) performed a controlled trial study with 97 subjects eye-tracking recordings (48 with dyslexia, 49 without) between the ages of 10 and 54. Subjects read 12 Spanish passages with 60 words each. Researchers detected patterns in the differences of eye-tracking behaviors between adults and students with and without dyslexia using Support Vector Machine (SVM), establishing SVM as a reliable tool for predicting dyslexia with 80.2% accuracy using cross-fold identification.

Likewise, Franzen et al. (2021) investigated the differences in visual sampling strategies among two groups of adults with and without dyslexia, using OpenDyslexic software reading assessment, which generated “dyslexia-friendly” font. Research confirmed the OpenDyslexic font is preferred by individuals with dyslexia. Furthermore, they found adults with dyslexia, including abnormal eye movements during reading, had slower average reading speeds, lower text comprehension, and reduced speed in word encoding in comparison.

Merging eye-tracking and functional magnetic resonance imaging (fMRI) offers neurological data to validate these results. Lukasova et al. (2018) tracked eye movements on a computer screen while taking fMRI images with adults (n = 21) (average age 24) and students (n = 15) (average age 11). Brain scans revealed that students learned predictive saccades less effectively than adults. Basic visuomotor circuitry presented in both adults and children, but improving the activation in response to task-related temporal and spatial demands matures later developmentally (Luckasova et al., 2018).

Drawing upon the work of Rayner (1998; 2009) and Clifton et al. (2007), Gran Ekstrand et al., 2021 found statistically significant differences between two groups of elementary students with and without RD. The researchers presented the cognitive characteristics of the students to predict dyslexia with eye-tracking results. Assessment results revealed correlations between performance and cognition among eight students who were eight to nine years old. Researchers detected cognitive limitations and cognitive difficulties among students with RD using a battery of neuropsychological assessments. Students with RD demonstrated poor reading and decoding related to cognition. These research studies support eye-tracking as a tool for monitoring reading during the developmental phases of reading acquisition.

Theoretical frameworks

Several important theoretical frameworks highlight the cognitive and neurological processes of reading and undergird this study. First, Visual Attention Theory (Johnston and Dark, 1982; 1986) brings focus to attention, which acts as a limited resource enabling individuals to selectively process and prioritize certain visual inputs over others (Takano et al., 2020). According to the Visual Attention Theory, attention is not a single process but rather consists of various networks and procedures (Binda and Morrone, 2018; Ciavarelli et al., 2021). According to Logan (1996), Visual Attention Theory offers a theoretical framework for comprehending how attention functions in the visual system and sheds light on the mechanisms that allow individuals to prioritize and interpret visual data during reading (Sims and Conati, 2020; Sluis and Broek, 2023). To study attentional processes, research in Visual Attention Theory uses eye-tracking, behavioral tests, and neuroimaging (Bundesen et al., 2015). Visual Attention Theory investigates how individuals selectively process and allocate attention to specific visual stimuli in the environment, filtering out irrelevant information and focusing on relevant aspects of the visual field (Bundesen et al., 2005). Eye-tracking technology provides a valuable tool for studying visual attention by measuring and analyzing eye movements (Solan et al., 2001). For example, during reading, a reader may focus on key content words while filtering out irrelevant text, illustrating selective attention in practice.

Second, the Magnocellular Deficit Theory proposes that individuals with specific learning disorders, such as dyslexia or RD, likely have impairments within their magnocellular pathway in the visual system (Meng and Schneider, 2022). This pathway is responsible for processing motion, spatial awareness, and other low-level visual features (APA, 2013), illustrating relationships between magnocellular deficits and eye movement patterns that contribute to RD. According to Magnocellular Deficit Theory, RD impacts this pathway as evidenced by longer fixation times. The combination of the Magnocellular Deficit Theory and eye-tracking allowed this study to investigate the underlying mechanisms of visual processing and explore potential links to RD using eye-tracking data with the statistical significance of the longer average fixation times among students with and without RD (Zoccolotti, 2022). By comparing average fixation times and proportions of fixations to total stimuli duration while reading, the current study might shed light on each of these theories.

Visual Attention Theory and Magnocellular Deficit Theory provide complementary explanations for the eye movement variables examined in this study. From a visual attention perspective, longer average fixation times and greater proportions of fixations to total stimuli duration reflect increased cognitive load and inefficient allocation of attentional resources during reading. Similarly, Magnocellular Deficit Theory posits that impairments in low level visual processing disrupt the rapid timing mechanisms necessary for fluent eye movements, resulting in prolonged fixations and altered fixation patterns. Together, these theories explain why elementary students with reading disabilities may demonstrate longer average fixation times and higher proportions of fixations during reading, making these eye tracking variables theoretically grounded indicators of reading difficulty.

The current study

Methodology

This quasi-experimental, comparison-group study of 29 elementary students was approved by the Institutional Review Board (IRB). Information about the current study's context, inclusion criteria, setting, participants, sample size, technological tools, instruments, and protocols follows. Given the exploratory nature and limited sample size, the analytical approach focused on detecting mean group differences rather than training, prediction, or classification of fixations. Independent sample t tests were selected because the research questions examined whether fixation metrics differed between groups rather than whether individuals could be classified into diagnostic categories.

Participant demographics

The participants were in two groups: Group A consisted of elementary students with RD (n = 12), and Group B consisted of students without RD (n = 17). The participants in grades 1, 2, and 3 ranged from 7 to 11 years old, with n = 18 (62%) males and n = 11 (38%) females. Participants’ race ethnicities varied across four categories: n = 14 (48%) were White, n = 9 (31%) were Hispanic, n = 3 (10.5%) were Black, and n = 3 (10.5%) were Asian. Students in the RD group were diagnosed by the school district as having a reading disability for reading fluency on their IEP, which was identified through testing by the district and validated by the teacher and by the parent through a parent survey.

Inclusion/exclusion criteria

Inclusion criteria consisted of fluent English speakers who were 6–11 years old. Participants were excluded who had a prior history of seizures or ultraviolet light sensitivity or blindness, or who refused to remove long hair or masks obstructing the recording of eye movements.

Subjects

The participants were in one group with RD (n = 12) and one group without RD (n = 17). They ranged in age from seven to eleven years old. There were n = 18 (62%) males and n = 11 (38%) females. Participants’ races varied between one of four categories of which n = 14 (48%) were White, n = 9 (31%) were Hispanic, n = 3 (10.5%) were Black, and n = 3 (10.5%) were Asian (seeTable 1).

Setting

Students in this study all attended one local elementary charter school in the southern U.S. Participants completed the reading assessment with the first author and a classroom aide in a quiet school workroom. The eye tracker was attached to an adaptable monitor screen. The first author controlled the Lexplore program from a computer for each participant to ensure the proper screen and sequence were displayed.

Sample size

First, Lexplore assessment offered six reading level outcomes: Emerging Reader, Low, Below Average, Average, Above Average, and High. Prior to running statistical analyses, the eye-tracking data of the 48 original participants were transitioned from Lexplore to the SPSS software program version 29. Once in SPSS, three incomplete data entries and 16 data entries from young readers who received a Lexplore ranking of emerging readers with no eye-tracking data and this message “analysis not possible” were removed. Therefore, a total of 29 participants’ eye-tracking data were utilized for the study as noted in Table 1.

The research team used G*Power to determine the sample size for the study (Faul et al., 2007). To accommodate an independent t-test, according to G*Power, with a large effect size (Cohen's F2) set at 0.35, power set at 80%, and two groups, a minimum total sample size of 29 participants was required. The obtained p-value to a predetermined significance level (alpha) used as a significance level was 0.05 (5%).

Instruments

In this study, AI was embedded within the proprietary Lexplore system and was not manipulated, trained, or retrained by the research team. All analyses conducted by the researchers relied on extracted fixation metrics and independent sample statistical tests rather than machine learning classification. Lexplore, an all-in-one reading assessment software for systematic reading development, incorporates AI and eye-tracking, visual insights, data, and reading activities based on assessment levels (Lexplore, n.d.). For example, Lexplore automatically generates the grade level reading passage. Additionally, we used the Tobii 5 Eye Tracker (Tobii, n.d.) as the primary tool for tracking, monitoring, and recording eye movements. However, without a software program such as Lexplore, the Tobii 5 Eye Tracker cannot aggregate data alone. These two instruments work is tandem, since the data collection required software which are (a) used with elementary students, (b) intended for measurements during the act of reading, as opposed to gaming, driving, or job performance (c) created with built-in grade-appropriate and leveled reading passages, and (d) compatible with an eye tracker. The Tobii 5 Eye Tracker is compatible with Lexplore to track both eyes simultaneously at 200 Hz per second while displaying a three-dimensional image of eye movements. To install the Tobii 5 Eye Tracker device, the first author attached the eye tracker, a discrete, slender bar with two infrared lights that appear when turned on, to the bottom of the computer monitor using two-sided tape included in the eye tracker kit.

Second, Lexplore software accommodates the Tobii to gather eye-tracking data for the portions of the elementary reading assessment. Lexplore provides reading comprehension questions aligned with each reading passage, read aloud to each participant. The assessment did not permit a return to reading passages for answers. Participants read the passage while the Tobii 5 Eye Tracker scanned and recorded their eye movements. The passage was five lines with 4.2 average words per line.

Procedural steps and phases

Participants completed five phases in logical order: (a) calibration, (b) Rapid Automatic Naming (RAN), (c) oral reading, (d) silent reading, and (e) comprehension questions. Each step ensured accurate eye-tracking and reading assessment.

Then, each elementary student, with and without RD, read an automatically generated passage based on their identified reading level on a computer monitor with a white background and black letters in Times New Roman font. The reading passage was five lines with an average of 4.2 words per line. As they read, the eye-tracking software program Lexplore (Lexplore, n.d.) with the built-in reading assessment and eye-tracking technology, tracked and analyzed eye movement data. Based on prior research on fixation times and proportions, the hypothesis was students with RD would have greater average fixation time and larger proportions of fixations to total stimuli duration than those without an identified RD.

To confirm acceptable calibration of the machine, the participant sat within 24–25 inches of the screen with their head positioned in a target zone while following circles with their eyes as the circles appeared on their computer monitor. This process took between 10 and 30 s. The purpose of the second phase, the RAN assessment, was to ensure the accuracy of the calibration. Participants provided the name of each letter aloud from left to right, starting in the top left corner. The RAN assessment consisted of four rows of nine letters each for a total of 36 letters and took one to two minutes. The third and fourth phases of the assessment occurred as participants read Lexplore's automatically generated, appropriate grade-level passages orally and silently. Together with phase five's comprehension questions, the process lasted under 12 min for each student.

Method and criteria to determine RD group

This multi-step classification process reflects school-based identification practices rather than clinical diagnosis. As such, RD status in this study represents the educational identification of reading difficulties which may or may not align perfectly with formal dyslexia diagnoses, yet may provide a beginning path to important RD. Students included in the Reading Disabilities (RD) group were selected through a systematic screening process. Reading disabilities, including dyslexia, are frequently identified within broader disability classifications such as Specific Learning Disability (SLD) rather than as a distinct diagnosis in school records (Lyon, et al., 2003; Peterson and Pennington, 2015).

In many educational contexts, schools focus on identifying students who demonstrate risk for reading difficulties through screening and evaluation rather than providing a formal clinical diagnosis of dyslexia, and therefore students with significant reading challenges may not always have a documented RD diagnosis (Peterson and Pennington, 2015; Snowling and Hulme, 2013). Consequently, RD may be under identified in school documentation, particularly when reading difficulties occur alongside other developmental or learning disabilities. To address this limitation, the initial screening for the present study relied on information beyond the IEP goal in that all parents were asked whether their child experienced reading difficulties and/or whether their child had a broader diagnosis such as a Attention Deficit Disorder (ADD), Attention Deficit Hyperactivity Disorder (ADHD), Obsessive-Compulsive Disorder (OCD), and Autism Spectrum Disorder (ASD) or SLD. This second layer of information ensured a student in the non-RD group was not in the process of evaluation or had issues not yet identified by the school. Students whose parents confirmed reading difficulties were then invited to participate in further screening to ensure they met the inclusion criteria. These students were administered a diagnostic reading assessment to determine whether they exhibited characteristics or symptoms consistent with RD as initially indicated by the parents and current IEP goals. Students who met the established criteria on the diagnostic assessment were subsequently included in the RD group for the study.

Reliability procedures

The following procedures were employed to ensure the accuracy and reliability of the eye-tracking data. The research team drew meaningful conclusions about data by ensuring consistency in (a) instrument calibration, (b) verbal instructions, (c) test administration, (d) and validation checks. Calibration was consistent across participants. For reliability and content validity, the second author reviewed at least 25% of data from Lexplore to SPSS with 100% accuracy.

Data analysis

Using data collected from Lexplore along with the Tobii 5 Eye Tracker, the research team employed an independent t-test to detect between-group differences in the average fixation time and proportions of fixations to total stimuli duration. According to prior research from Prado et al. (2007) and colleagues, a t-test was used to determine if statistically significant differences existed between the average fixation times between students with RD compared to those readers without an identified RD.

Results

Table 2 shows Levene's test for equality of variances that was conducted and indicated that the assumption of homogeneity of variances was not met (F = 78.44, p ≤ .001). Therefore, SPSS used the Welch-Satterthwaite t-test, when two unpaired groups have unequal variances. This violation was corrected by not using the pooled estimate for the error term for the t-statistic and instead making an adjustment to the degrees of freedom with equal variances not assumed.

Therefore, this independent t-test, using data generated by Tobii and Lexplore, indicated significant differences in the average fixation time between the two groups. As expected, the eye-tracking measure was able to identify students with RD, those with longer average fixation times, with veracity. There was a significant difference in average fixation time between students with RD (M = 1,751.25 SD = 759.74) and students without RD (M = 499.24, SD = 134.90); t (27) = 6.696, p = <.001

To perform the independent t-test for RQ2, the total stimuli duration data were converted from seconds to milliseconds (ms), since the average fixation times were reported in ms by Lexplore). The research team converted the seconds to ms, maintaining the average fixation times given in ms. Then, using Microsoft Excel to divide the average fixation time by the total stimuli duration to determine the proportions of fixations to total stimuli duration for each participant. This procedure was conducted in response to RQ2 to measure the visual attention to the total stimuli duration (i.e., exploring how long participants looked at the content on the screen). These data for RQ2 complement and confirm the average fixation time data used for RQ1 (Purves, 2008).

See Table 4 for the results that show the independent t-test did determine mean differences in proportions of fixations to total stimuli duration between students with and without RD. There was a significant similarity in the proportions of fixations to total stimuli duration between students with RD (M = .0166, SD = .01069) and students without RD (M = .0095, SD = .00237); t (27) = 2.689, p = .012. Levene's test for equality of variances was conducted and indicated that the assumption of homogeneity of variances was not met (F = 47.92, p ≤ .001).

The magnitude of the group difference was large in each t-test conducted, consistent with prior eye tracking research. For each question, the research provided effect sizes alongside statistical significance, providing additional context for interpreting the practical importance of these findings for each research question. The difference in proportions of fixations to total stimuli duration was also associated with a large effect size. Cohen's d was approximately 1.08, suggesting a meaningful distinction between groups despite overall low fixation proportions.

Discussion

The purpose of this study was to examine whether eye-tracking measures could distinguish elementary students with and without reading disabilities, and the findings revealed statistically significant group differences in both average fixation time and proportions of fixations to total stimuli duration. These findings should not be interpreted as evidence that eye tracking alone can diagnose lead to important reading disabilities, but rather that fixation-based metrics may differentiate groups under controlled conditions. Reporting effect sizes alongside statistical significance highlights that the observed group differences were not only statistically reliable but also practically meaningful, while remaining exploratory due to sample size constraints. Together, these results suggest that eye-tracking metrics capture meaningful differences in visual and cognitive processing during reading and may offer promise as an early indicator of reading difficulty, while acknowledging the exploratory nature and sample limitations of the study.

Learning sciences: eye-tracking, and AI for the future of early reading screening

Traditional universal screening is often implemented after years of struggles and can take hours and even weeks for teachers to implement, analyze, and consider potential interventions. Fast, effective, efficient, reliable, and valid tools to detect reading difficulties early in the process currently do not exist, but are possible with advancements in learning sciences, machine learning, multi-modal data analytics, advancements in biometrics, and artificial intelligence (Erbeli et al., 2023; Khan et al., 2018). Although technological tools come with their own potential bias by those who create them (Sun et al., 2020), exploring how AI could offset the overburden and demand on teachers by providing eye-tracking reading screenings that are technology-driven is an area of research worth further exploration (Zawacki-Richter et al., 2019).

Diagnosticians and teachers could leverage AI to analyze student performance data, feedback, and assessment results (Marino et al., 2023). These data-driven approaches could help identify effective teaching strategies and inform the development of evidence-based practices and effective teaching strategies. Also, AI mitigates issues typical of traditional universal reading screenings (Benfatto, 2021; Benfatto et al., 2016). Within current diagnostic testing contexts, cultural and linguistic bias, unique learning profiles, and test administration errors persist (Sarker, 2021). Using eye-tracking as a beginning and ongoing screening, teachers could access data they might never see with the potential for immediate implementation of interventions.

Future studies should extend this work by training supervised machine learning classifiers, such as support vector machines, using fixation features to examine classification accuracy and generalizability across samples. The findings from the current study showed that students with RD had longer average fixation times compared with those without RD. Both students with and without RD had statistically significant low proportions of fixations to total stimuli duration. Determining differences in fixation behavior using eye-tracking data is important when detecting RD for several reasons. Eye-tracking tools allow researchers to assess behavioral measures of reading skills at an early stage, even before some students can read fluently (Vajs et al., 2023). Eye-tracking provides an objective and quantitative measurement of eye movements, capturing detailed information about fixation duration, fixation location, saccades (rapid eye movements), and other parameters (Types of Eye Movements, 2022). These measurements can be compared across individuals or groups to identify patterns or deviations indicative of RD. Furthermore, eye-tracking can offer insights into specific aspects of reading behavior (Valtakari et al., 2021). For example, tracking can indicate whether an elementary student has difficulties with word recognition, decoding, or comprehension (Gran Ekstrand et al., 2021).

Future of early intervention and AI-driven technology

Given the findings of the current study, areas of weakness may be identified by examining fixation patterns, allowing for focused interventions for RD-related issues. Eye-tracking, combined with machine learning algorithms, has the potential to serve as a diagnostic tool for RD in the future. With today's burgeoning AI innovations, we look forward to training algorithms on large datasets of eye-tracking data from individuals with and without RD, developing models to accurately classify individuals for intervention based on their eye movement patterns, and aiding in the diagnosis process (Benfatto et al., 2016). Finally, eye-tracking can also be used to monitor the effectiveness of interventions or reading interventions over time. By tracking changes in fixation behavior, researchers and educators can assess the impact of specific interventions and adjust as necessary (Hessel et al., 2021).

Limitations and future research

The current study had similar methodological and analytical limitations found in prior studies (Cui et al., 2020; Kim and Wiseheart, 2017; McChesney and Bond, 2019, Rello and Bastelaros, 2015): (1) drift after calibration, (2) missing data, and (3) distracting environmental conditions. Also, eye-tracking data is complex and difficult to interpret (Sims and Conati, 2020; Sluis and Broek, 2023). The Lexplore program counteracts this complexity by offering streamlined and easy-to-read data outputs. Despite the ease of use of the data, the accuracy of the current study was affected by the unpredictable movement of participants. The research team suggests, in future studies, a chin rest be used to affix the head, ensuring one locked position to increase data accuracy (Valtakari et al., 2021). Although the first author initially incorporated a RAN test, which involves reading isolated letters instead of complete paragraphs, this data could not be included in the analysis, since only second-grade or lower participants could advance to the RAN phase. This automatic feature in Lexplore is likely due to its sole goal of identifying reading levels; therefore, the tool ignores students who cannot yet read and who need additional calibration measures. The question persists surrounding ideation that younger students whose data was not collected may be on track as naturally emergent readers or may be at risk for RD. Also, an important extension to this work would involve comparing researcher-derived statistical results with Lexplore's internal risk classifications to examine convergence, divergence, and interpretability of AI driven outputs.

Several sources of bias should be considered when interpreting these results. The sample was drawn from a single charter school and may not represent broader demographic or instructional variation. Additionally, embedded algorithms within Lexplore are proprietary, limiting transparency regarding how raw eye movement signals are processed prior to exportation.

Additionally, many RDs are comorbid with other conditions (Tiadi et al., 2016), such as Attention Deficit Disorder (ADD), Attention Deficit Hyperactivity Disorder (ADHD), Obsessive-Compulsive Disorder (OCD), and Autism Spectrum Disorder (ASD). These conditions may also affect eye movements, making it difficult to identify specific patterns associated with RDs. Therefore, recruiting and collecting data from larger numbers of participants is needed. Future researchers should access larger sample sizes by teaming up with a school district to administer eye-tracking and machine learning as part of the school’s curriculum (Meisinger and Seimyr, 2019), ensuring a large, diverse sample.

The Lexplore reading passages were generated based on age and grade level. This levelling of passages contradicts current U.S. reading remediation programs where the reading passages are implemented based on reading, rather than grade level. However, the Lexplore assessment was implemented as a diagnostic tool meant to determine whether the reader was proficiently on grade level. Notably, this use appears to have resulted in a possible floor effect. Therefore, the results of this study are interpreted with caution, considering the constraints and limitations.

Certainly, another limitation was the number of participants required for a robust statistical analysis, including an independent t-test with a confidence interval of 90%, which was 48. However, after collecting data, the research team was then made aware that Lexplore would not collect the needed granular level of eye-tracking data from the least proficient or emergent readers. Therefore, only data such as words per minute, comprehension, and rankings were generated by Lexplore for those 16 emergent readers. Fortunately, however, the remaining eye-tracking data of 29 participants did allow statistically significant findings from this exploratory study, which hopefully is the beginning of must needed and continued research in this field.

Educational implications

Further sustained research is needed on the impact of using eye-tracking as an inexpensive, quick, and effective means of screening elementary students to detect visual attention deficits and possible reading difficulties. Anomalies in eye-tracking behaviors while working with elementary students are prevalent across fields such as child development (Franchak et al., 2011), behavioral psychology (Kim et al., 2020; Palama et al., 2022), child neurology and neurological medicine (Bekteshi et al., 2022), pediatric ophthalmology (Wygnanski-Jaffe et al., 2023), special education diagnosis of ASD (Vargas-Cuentas et al., 2017), and speech and language (Christou et al., 2022). Therefore, technological advances, age-appropriate, eye-tracking instruments, and compatible software programs demand further investigation in long-term, large-scale, and diverse settings. Furthermore, additional research to leverage eye movement data toward generating the most effective grade-level passages for assessment data and instruction should be explored to screen and remediate RD.

Using eye-tracking to detect possible reading issues in elementary students could be used on a large scale throughout the U.S. education system as a more efficient method for identifying and remediating reading problems affecting their personal and academic lives. Traditional screening and diagnostic programs offer progress monitoring and have lower costs per student, but higher overall costs from extra fees for program training courses, setup fees, teacher time to administer and analyze outcomes, and additional remediation costs due to the lack of early detection. Eye-tracking screening could offer efficient and economical means of early identification as well as progress monitoring.

Conclusions

For some, learning to read is complicated. Teaching elementary students with disabilities to read is even more complex (Fletcher and Vaughn 2009; Moats, 2020). Teachers may leverage tools, such as eye-tracking, to obtain information on reading speed, fixation duration, and regression patterns (Benfatto et al., 2016). Deviations from typical reading patterns have been used in previous research using machine learning algorithms for RD detection (Cui et al., 2020; Kim and Wiseheart, 2017; McChesney and Bond, 2019). The combination of obtaining eye-tracking data and using machine learning to analyze the data, combined with AI technology, could provide teachers with a faster and potentially more reliable way of detecting reading difficulty early on, addressing reading issues before they persist and worsen over time (National Institute of Child Health and Human Development [NICHD], 2020; Kaye et al., 2022; Rello and Bastelaros, 2015).

The results support eye tracking as a promising research-based indicator of reading difficulty, not as a standalone diagnostic or replacement for existing screening frameworks. By integrating insights from learning sciences using eye-tracking combined with related theoretical frameworks, educators and researchers can potentially further enhance their understanding of reading difficulties. Descriptive data and independent t-tests confirmed the theoretical underpinnings of Visual Attention Theory's claim that impairments are consistent with longer average fixation times. Likewise, the current study's data adds to the body of literature around Magnocellular Deficit Theory, wherein RD appears to impair the key visual pathway as evidenced by longer fixation times and larger proportions of fixations to total stimuli duration. For example, during reading, a student may focus on key content words while filtering out irrelevant text, illustrating selective attention in practice. This theory explains why students with RD often exhibit longer fixation times, as their visual processing speed is compromised.

The long-term impact of undetected RD leads to ongoing reading difficulties throughout their schooling years and into adolescence and adulthood (Fletcher and Vaughn, 2009; Snowling and Hulme, 2013). Reliable tools and assessments are crucial for identifying elementary students at risk of developing reading difficulties (Fletcher et al., 2018; Vaughn and Fletcher, 2020). Eye-tracking tools can assess visual components which contribute heavily to reading skills, such as phonological awareness, decoding, fluency, and comprehension. Early identification using these tools enables targeted interventions to address specific areas of difficulty (Kaye et al., 2022; Petscher et al., 2019). The current study provided strong evidence to support the use of eye-tracking as an early, efficient, and cost-effective measure for early detection of an RD while providing clearer data outcomes and saving teachers compared to traditional reading screeners.

References

• 1

  AdamsB.WilsonN. S. (2022). Investigating student’s during-reading practices through social annotation. Lit. Res. Instr.61 (4), 339–360. 10.1080/19388071.2021.2008560

  • CrossRef
  • Google Scholar

• 2

  AlqahtaniS. S. (2025). Predictive validity of early and mid-year literacy assessments for end-of-year word reading fluency. PLoS One20, e0327242. 10.1371/journal.pone.0327242

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  American Psychiatric Association (2013). Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Arlington, VA: American Psychiatric Publishing. 10.1176/appi.books.9780890425596

  • CrossRef
  • Google Scholar

• 4

  AntúnezM.MilliganS.Hernández-CabreraJ. A.BarberH. A.SchotterE. R. (2022). Semantic parafoveal processing in natural Reading: insight from fixation-related potentials & eye movements. Psychophysiology59 (4), 1–19. 10.1111/psyp.13986

  • CrossRef
  • Google Scholar

• 5

  BekteshiS.KarlssonP.De ReyckL.VermeerbergenK.KoningsM.HellinP.et al (2022). Eye movements and stress during eye-tracking gaming performance in children with dyskinetic cerebral palsy. Dev. Med. Child. Neurol.64 (11), 1402–1415. 10.1111/dmcn.15237

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  BenfattoM. N. (2013). Predicting Reading disability from eye movements [conference session]. Book of Abstracts of the 17th European Conference on Eye Movements. Lund, Sweden.

  • Google Scholar

• 7

  BenfattoM. N. (2021). The Research Behind Lexplore. Helsinki: YouTube. Available online at:https://educationalliancefinland.com/products/lexplore(Accessed March 15, 2026).

  • Google Scholar

• 8

  BenfattoM. N.SeimyrGÖYggeJ.PansellT.RydbergA.JacobsonC. (2016). Screening for dyslexia using eye-tracking during reading. PLoS One11 (12), 1–16. 10.1371/journal.pone.0165508

  • CrossRef
  • Google Scholar

• 9

  BindaP.MorroneM. C. (2018). Vision during saccadic eye movements. Annu. Rev. Vis. Sci.4 (1), 193–213. 10.1146/annurev-vision-091517-034317

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10

  BixlerR.D’MelloS. (2016). Automatic gaze-based user-independent detection of mind wandering during computerized reading. User. Model. User-adapt. Interact.26 (1), 33–68. 10.1007/s11257-015-9167-1

  • CrossRef
  • Google Scholar

• 11

  BundesenC. (1990). A theory of visual attention. Psychol. Rev.97 (4), 523–547. 10.1037/0033-295X.97.4.523

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12

  BundesenC.HabekostT.KyllingsbækS. (2005). A neural theory of visual attention: bridging cognition and neurophysiology. Psychol. Rev.112 (2), 291–328. 10.1037/0033-295X.112.2.291

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13

  BundesenC.VangkildeS.PetersenA. (2015). Recent developments in a computational theory of visual attention. Vision Res.116 (2015), 210–218. 10.1016/j.visres.2014.11.005

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  CaldaniS.GerardC.-L.PeyreH.BucciM. P. (2020). Visual attentional training improves reading capabilities in children with dyslexia: an eye tracker study during a reading task. Brain. Sci.10 (8), 558. 10.3390/brainsci10080558

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  CattsH. W.PetscherY.SchatschneiderC.Sittner BridgesM.MendozaK. (2009). Floor effects associated with universal screening and their impact on the early identification of reading disabilities. J. Learn. Disabil.42 (2), 163–176. 10.1177/0022219408326219

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  ChristouS.ColomaC. J.AndreuL.GuerraE.ArayaC.Rodriguez-FerreiroJ.et al (2022). Online comprehension of verbal number morphology in children with developmental language disorder: an eye-tracking study. J. Speech. Lang. Hear. Res.65 (11), 4181–4204. 10.1044/2022_JSLHR-21-00591

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 17

  CiavarelliA.ContemoriG.BattagliniL.BarolloM.CascoC. (2021). Dyslexia and the magnocellular-parvocellular coactivation hypothesis. Vision Res.179 (2021), 64–74. 10.1016/j.visres.2020.10.008

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  CliftonC.StaubA.RaynerK. (2007). “Eye movements in Reading words and sentences,” in Eye Movements: A Window on Mind and Brain, eds. van GompelR. P. G.FischerM. H.MurrayW. S.HillR. L. (Elsevier), 341–371. 10.1016/B978-008044980-7/50017-3

  • CrossRef
  • Google Scholar

• 19

  CuiX.WangJ.ChangY.SuM.ShermanH. T.WuZ.et al (2020). Visual search in Chinese children with attention-deficit/hyperactivity disorder and comorbid developmental dyslexia: evidence for pathogenesis from eye movements. Front. Psychol.11 (880), 1–8. 10.3389/fpsyg.2020.00880

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  DavisG. N.LindoE. J.ComptonD. L. (2007). Children at risk for Reading failure: constructing an early screening measure. Teach. Except. Child.39 (5), 32–37. 10.1177/004005990703900505

  • CrossRef
  • Google Scholar

• 21

  ErbeliF.HeK.CheekC.RiceM.QianX. (2023). Exploring the machine learning paradigm in determining risk for Reading disability. Sci. Stud. Read.27 (1), 5–20. 10.1080/10888438.2022.2115914

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 22

  FaulF.ErdfelderE.LangA. G.BuchnerA. (2007). G*power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods.39 (2), 175–191. 10.3758/Bf03193146

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  FinnC. E.RotherhamA. J.HokansonC. R. (2001). Rethinking Special Education for a New Century. Thomas B. Fordham Foundation.

  • Google Scholar

• 24

  FletcherJ. M.LyonG. R.FuchsL.BarnesM. A. (2018). Learning Disabilities: From Identification to Intervention. The Guilford Press.

  • Google Scholar

• 25

  FletcherJ. M.VaughnS. (2009). Response to intervention: preventing and remediating academic deficits. Child Development and Perspectives3 (1), 30–37. 10.1111/j.1750-8606.2008.00072.x

  • CrossRef
  • Google Scholar

• 26

  FouganthineA. (2012). Predicting Reading disability from eye movements. European Conference on Eye Movements, Lund, Sweden.

  • Google Scholar

• 27

  FranchakJ. M.KretchK. S.SoskaK. C.AdolphK. E. (2011). Head-mounted eye-tracking: a new method to describe infant looking. Child. Dev.82 (6), 1738–1750. 10.1111/j.1467-8624.2011.01670.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  FranzenL.StarkZ.JohnsonA. P. (2021). Individuals with dyslexia use a different visual Reading strategy to sample text: comprehensive evidence from eye-tracking. Sci. Rep.6449 (11), 1–17. 10.1038/s41598-021-84945-9

  • CrossRef
  • Google Scholar

• 29

  FuchsD.FuchsL. S. (2001). Responsiveness-to-intervention: a blueprint for practitioners, policymakers, and parents. Teach. Except. Child.38 (1), 57–61. 10.1177/004005990503800112

  • CrossRef
  • Google Scholar

• 30

  FuchsD.FuchsL. S.ComptonD. L. (2012). Smart RTI: a next-generation approach to multilevel prevention. Except. Child.78 (3), 263–279. 10.1177/001440291207800301

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31

  García ChimenoY.García ZapirainB.Saralegui PrietoI.Fernandez-RuanovaB. (2014). Automatic classification of dyslexic children by applying machine learning to fMRI images. Bio-Med. Mater. Eng.24 (6), 2995–3002. 10.3233/BME-141120

  • CrossRef
  • Google Scholar

• 32

  GloverT. A.AlbersC. A. (2007). Considerations for evaluating universal screening assessments. J. Sch. Psychol.45 (2), 117–135. 10.1016/j.jsp.2006.05.005

  • CrossRef
  • Google Scholar

• 33

  Gran EkstrandA. C.BenfattoM. N.SeimyrGÖ (2021). Screening for reading difficulties: comparing eye-tracking outcomes to neuropsychological assessments. Front. Educ.6 (643232), 1–13. 10.3389/feduc.2021.643232

  • CrossRef
  • Google Scholar

• 34

  HesselA. K.NationK.MurphyV. A. (2021). Comprehension monitoring during reading: an eye-tracking study with children learning English as an additional language. Sci. Stud. Read.25 (2), 159–178. 10.1080/10888438.2020.1740227

  • CrossRef
  • Google Scholar

• 35

  HueyG. (1908). The Psychology and Pedagogy of Reading. The Macmillan Company.

  • Google Scholar

• 36

  JimenezL. M.MeyerC. K. (2016). First impressions matter: navigating graphic novels utilizing linguistic, visual, and spatial resources. J. Lit. Res.48 (4), 423–447. 10.1177/1086296X16677955

  • CrossRef
  • Google Scholar

• 37

  JohnsonE. S.JenkinsJ. R.PetscherY. (2010). Improving the accuracy of a direct route screening process. Assess. Eff. Interv.35 (3), 131–140. 10.1177/1534508409348375

  • CrossRef
  • Google Scholar

• 38

  JohnstonW. A.DarkV. J. (1982). In defense of interperceptual theories of attention. J. Exp. Psychol. Hum. Percept. Perform.8 (3), 407–421. 10.1037/0096-1523.8.3.407

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39

  JohnstonW. A.DarkV. J. (1986). Selective attention. Annu. Rev. Psychol.37 (1), 43–75. 10.1146/annurev.ps.37.020186.000355

  • CrossRef
  • Google Scholar

• 40

  KaniukaT. S. (2010). Reading achievement, attitude toward reading, and reading self-esteem of historically low achieving students. J. Instr. Psychol37 (2), 184–188.

  • Google Scholar

• 41

  KayeE. L.LozadaV.BriggsC. (2022). Early identification of and intervention for children with and without dyslexia characteristics: a comparison study. Lit. Res. Instr.61 (3), 298–313. 10.1080/19388071.2022.2059418

  • CrossRef
  • Google Scholar

• 42

  KhanR. U.ChengJ. L. A.BeeO. Y. (2018). Machine learning and dyslexia: diagnostic and classification system (DCS) for kids with learning disabilities. Int. J. Eng. Technol7 (3.18), 97–100.

  • Google Scholar

• 43

  KimJ.SinghS.ThiessenE. D.FisherA. V. (2020). A hidden markov model for analyzing eye-tracking of moving objects: case study in a sustained attention paradigm. Behav. Res. Methods.52 (3), 1225–1243. 10.3758/s13428-019-01313-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  KimS.WiseheartR. (2017). Exploring text and icon graph interpretation in students with dyslexia: an eye-tracking study. Dyslexia.23 (1), 24–41. 10.1002/dys.1551

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  Lexplore (n.d.). The Process Step by Step. Lexplore reading assessment [Software]. Lexplore AB, Sweden. Available online at: https://www.lexplore.com(Accessed March 15, 2026).

  • Google Scholar

• 46

  LobierM.DuboisM.ValdoisS. (2013). The role of visual processing speed in Reading speed development. PLoS One8 (4), 1–10. 10.1371/journal.pone.0058097

  • CrossRef
  • Google Scholar

• 47

  LoganG. D. (1996). The CODE theory of visual attention: an integration of space-based and object-based attention. Psychol. Rev.103 (4), 603–649. 10.1037/0033-295X.103.4.603

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48

  LukasovaK.NucciM. P.Machado de Azevedo NetoR.VieiraG.SatoJ. R.AmaroJ. (2018). Predictive saccades in children and adults: a combined fMRI and eye-tracking study. PLoS One13 (5), 1–17. 10.1371/journal.pone.0196000

  • CrossRef
  • Google Scholar

• 49

  LyonG. R.FletcherJ. M.ShaywitzS. E.ShaywitzB. A.TorgesenJ. K.WoodF. B.et al (2003). “Rethinking learning disabilities,” in Rethinking Learning Disabilities, ed. LyonG. R. (New York:Guilford Press), 259–287.

  • Google Scholar

• 50

  MarinoM. T.VasquezE.DiekerL. A.BashamJ. D.BlackorbyJ. (2023). The future of artificial intelligence in special education technology. J. Spec. Educ. Technol.38 (3), 404–416. 10.1177/01626434231165977

  • CrossRef
  • Google Scholar

• 51

  McChesneyI.BondR. (2019). Eye tracking analysis of computer program comprehension in programmers with dyslexia. Empirical Softw. Eng.24 (2019), 1109–1154. 10.1007/s10664-018-9649-y

  • CrossRef
  • Google Scholar

• 52

  McGrathL. M.PenningtonB. F.ShanahanM. A.Santerre-LemmonL. E.BarnardH. D.WillcuttE. G.et al (2010). A multiple deficit model of Reading disability and attention-deficit/hyperactivity disorder: searching for shared cognitive deficits. J. Child Psychol. Psychiatry52 (5), 547–557. 10.1111/j.1469-7610.2010.02346.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53

  MeisingerE. B.SeimyrG. Ö. (2019). Screening for Reading Difficulties Using Lexplore. 10.5281/zenodo.3554903

  • CrossRef
  • Google Scholar

• 54

  MengQ.SchneiderK. A. (2022). Dyslexia Linked to Profound Impairment in the Magnocellular Medial Geniculate Nucleus. Atlanta: Health & Medicine Week, 1822. Available online at:https://link.gale.com/apps/doc/A695893158/AONE?u=orla57816&sid=bookmark-AONE&xid=b706369c(Accessed March 15, 2026).

  • Google Scholar

• 55

  MoatsL. C. (2020). Teaching reading is rocket science what expert teachers of reading should know and be able to do. Am. Educ.44 (2), 4–9.

  • Google Scholar

• 56

  MoatsL. C.DakinK. (2008). Basic Facts About Dyslexia & Other Reading Problems. Baltimore: International Dyslexia Association.

  • Google Scholar

• 57

  MorrisonT. G.WilcoxB.Thomas BillenM.CarrS.WilcoxG.MorrisonD.et al (2011). 50 Years of literacy research and instruction: 1961–2011. Literacy Research and Instruction50 (4), 313–326. 10.1080/19388071.2011.602924

  • CrossRef
  • Google Scholar

• 58

  National Institute of Child Health and Human Development [NICHD]. (2020). NICHD reading and Reading Disorders Research Information. Bethesda, MD: National Institute of Health. Available online at:https://www.nichd.nih.gov/health/topics/reading/researchinfo(Accessed March 15, 2026).

  • Google Scholar

• 59

  OdegardT. N.FarrisE. A.MiddletonA. E.OslundE.Rimrodt-FriersonS. (2020). Characteristics of students identified with dyslexia within the context of state legislation. J. Learn. Disabil.53 (5), 366–379. 10.1177/0022219420914551

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 60

  Ozeri-RotstainA.ShachafI.FarahR.Horowitz-KrausT. (2020). Relationship between eye-movement patterns, cognitive load, and Reading ability in children with Reading difficulties. J. Psycholinguist. Res.49 (3), 491–507. 10.1007/s10936-020-09705-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 61

  PalamaA.MalsertJ.GrandjeanD.SanderD.GentazE. (2022). The cross-modal transfer of emotional information from voices to faces in 5-, 8- and 10-year-old children and adults: an eye-tracking study. Emotion22 (4), 725–739. 10.1037/emo0000758

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62

  PetersonR. L.PenningtonB. F. (2015). Developmental dyslexia. Annu. Rev. Clin. Psychol.11, 283–307. 10.1146/annurev-clinpsy-032814-112842

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 63

  PetscherY.FienH.StanleyC.GearinB.GaabN.FletcherJ. M.et al (2019). Screening for Dyslexia. Bethesda, MD: U.S. Department of Education, Office of Elementary and Secondary Education, Office of Special Education Programs, National Center on Improving Literacy. Available online at:https://www.improvingliteracy.org(Accessed March 15, 2026).

  • Google Scholar

• 64

  PrabhaA. J.BhargaviR. (2020). Predictive model for dyslexia from fixations and saccadic eye movement events. Comput. Methods Programs Biomed.195, 1–13. 10.1016/j.cmpb.2020.105538

  • CrossRef
  • Google Scholar

• 65

  PradoC.DuboisM.ValdoisS. (2007). The eye movements of dyslexic children during Reading and visual search: impact of the visual attention span. Vision Res.47 (19), 2521–2530. 10.1016/j.visres.2007.06.001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 66

  PughK. R.MenclW. E.JennerA. R.KatzL.FrostS. J.LeeJ. R.et al (2001). Neurobiological studies of Reading and Reading disability. J. Commun. Disord.34 (6), 479–492. 10.1016/S0021-9924(01)00060-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67

  PurvesD. E. (2008). Neuroscience. 4th ed. Sunderland, MA: Sinauer.

  • Google Scholar

• 68

  RaatikainenP.HautalaJ.LobergO.KärkkäinenT.LeppänenP.NieminenP. (2021). Detection of developmental dyslexia with machine learning using eye movement data. Array12 (100087), 1–7. 10.1016/j.array.2021.100087

  • CrossRef
  • Google Scholar

• 69

  RaneyG. E.CampbellS. J.BoveeJ. C. (2014). Using eye movements to evaluate the cognitive processes involved in text comprehension. J. Visualized Exp.83, 1–7. 10.3791/50780

  • CrossRef
  • Google Scholar

• 70

  RaynerK. (1998). Eye movements in Reading and information processing: 20 years of research. Psychol. Bull.124 (3), 372–422. 10.1037/0033-2909.124.3.372

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 71

  RaynerK. (2009). Eye movements and attention in Reading, scene perception and visual search. Q. J. Exp. Psychol.62, 1457–1506. 10.1080/17470210902816461

  • CrossRef
  • Google Scholar

• 72

  RaynerK.ChaceK. H.SlatteryT. J.AshbyJ. (2006). Eye movements as reflections of comprehension processes in Reading. Sci. Stud. Read.10, 241–255. 10.1207/s1532799xssr1003_3

  • CrossRef
  • Google Scholar

• 73

  RelloL.BallesterosM. (2015). Detecting readers with dyslexia using machine learning with measures. Proceedings of the 12th Web for All Conference, 1–8. 10.1145/2745555.2746644

  • CrossRef
  • Google Scholar

• 74

  RiedlM. O. (2019). Human-centered artificial intelligence and machine learning. Hum. Behav. Emerg. Technol.1 (1), 33–36. 10.1002/hbe2.117

  • CrossRef
  • Google Scholar

• 75

  RitchieK. D.SpeeceD. L. (2004). Early identification of reading disabilities: current status and new directions. Assess. Eff. Interv.29 (4), 13–24. 10.1177/073724770402900404

  • CrossRef
  • Google Scholar

• 76

  SarkerI. H. (2021). Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions. SN Comput. Sci.2 (6), 1–20. 10.1007/s42979-021-00815-1

  • CrossRef
  • Google Scholar

• 77

  SchoenbachC.GreenleafC.MurphyL.HoganN. (2023). Reading for Understanding how Reading Apprenticeship Improves Disciplinary Learning in Secondary and College Classrooms. 3rd ed. San Francisco: Jossey-Bass.

  • Google Scholar

• 78

  SeimyrGÖBenfattoM. N. (2017). Lexplore: Eye-tracking-based Reading assessment.

  • Google Scholar

• 79

  ShaywitzB. A.LyonG. R.ShaywitzS. E. (2006). The role of functional magnetic resonance imaging in understanding reading and dyslexia. Dev. Neuropsychol.30 (1), 613–632. 10.1207/s15326942dn3001_5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 80

  ShaywitzS. E.ShaywitzB. A. (2008). Paying attention to Reading: the neurobiology of Reading and dyslexia. Dev. Psychopathol.20 (4), 1329–1349. 10.1017/S0954579408000631

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 81

  SimsS. D.ConatiC. (2020). A neural architecture for detecting user confusion in eye-tracking data. Proceedings of the 2020 International Conference on Multimodal Interaction, 15–23. 10.1145/3382507.3418828

  • CrossRef
  • Google Scholar

• 82

  SluisL.BroekE. L. (2023). Feedback beyond accuracy: using eye-tracking to detect comprehensibility and interest during reading. J. Am. Soc. Inf. Sci. Technol.74 (1), 3–16. 10.1002/asi.24657

  • CrossRef
  • Google Scholar

• 83

  SnowlingM. J.HulmeC. (2013). The Science of Reading: A Handbook. 1st ed. Malden, MA: Blackwell Publishing. 10.1604/9780470757635

  • CrossRef
  • Google Scholar

• 84

  SnowlingM. J.MuterV.CarrollJ. (2007). Children at family risk of dyslexia: a follow-up in early adolescence. J. Child Psychol. Psychiatry48 (6), 609–618. 10.1111/j.1469-7610.2006.01725.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 85

  SolanH. A.LarsonS.Shelley-TremblayJ.FicarraA.SilvermanM. (2001). Role of visual attention in cognitive control of oculomotor readiness in students with reading disabilities. J. Learn. Disabil.34 (2), 107–118. 10.1177/002221940103400202

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 86

  SteinJ. (2014). Dyslexia: the role of vision and visual attention. Curr. Dev. Disord. Rep.1 (4), 267–280. 10.1007/s40474-014-0030-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 87

  SunW.NasraouiO.ShaftoP. (2020). Evolution and impact of bias in human and machine learning algorithm interaction. PLoS One15 (8), e0235502–e0235502. 10.1371/journal.pone.0235502

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 88

  TakanoS.MatsumiyaK.TsengC. H.KurikiI.DeubelH.ShioiriS. (2020). Displacement detection is suppressed by the post-saccadic stimulus. Sci. Rep.10 (1), 9273–9273. 10.1038/s41598-020-66216-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 89

  TiadiA.SeassauM.GerardC. L.BucciM. P. (2016). Differences between dyslexic and non-dyslexic children in the performance of phonological visual-auditory recognition tasks: an eye-tracking study. PLoS One11 (7), 1–16. 10.1371/journal.pone.0159190Tobii

  • CrossRef
  • Google Scholar

• 90

  Tobii Eye Tracker 5. (n.d.). Available online at:https://gaming.tobii.com/product/eye-tracker-5/(Accessed March 15, 2026).

  • Google Scholar

• 91

  Tobii Connect (2022). Types of eye Movements. Stockholm: Tobbi Connect. Available online at:https://connect.tobii.com/s/article/types-of-eye-movements?language=en_US#:∼:text=Fixations%20are%20the%20most%20common,what%20is%20being%20looked%20at(Accessed March 15, 2026).

  • Google Scholar

• 92

  TorgesenJ. K.WagnerR. K.RashotteC. A.RoseE.LindamoodP.ConwayT.et al (1999). Preventing reading failure in young children with phonological processing disabilities: group and individual responses to instruction. J. Educ. Psychol.91 (4), 579–593. 10.1037/0022-0663.91.4.579

  • CrossRef
  • Google Scholar

• 93

  Types of Eye Movements. (2022). Tobii Connect. Available online at:https://connect.tobii.com

  • Google Scholar

• 94

  VajsI.PapićT.KovićV.SavićA. M.JankovićM. M. (2023). Accessible dyslexia detection with real-time Reading feedback through robust interpretable eye-tracking features. Brain. Sci.13 (3), 405–417. 10.3390/brainsci13030405

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 95

  ValtakariN. V.HoogeI. T. C.ViktorssonC.NyströmP.Falck-YtterT.HesselsR. S. (2021). Eye tracking in human interaction: possibilities and limitations. Behav. Res. Methods.53 (4), 1592–1608. 10.3758/s13428-020-01517-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 96

  Van Der KleijS. W.SegersE.GroenM. A.VerhoevenL. (2019). Post-treatment reading development in children with dyslexia: the challenge remains. Ann. Dyslexia.69 (3), 279–296. 10.1007/s11881-019-00186-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 97

  Vargas-CuentasN. I.Roman-GonzalezA.GilmanR. H.BarrientosF.TingJ.HidalgoD.et al (2017). Developing an eye-tracking algorithm as a potential tool for early diagnosis of autism spectrum disorder in children. PLoS One12 (11), e0188826–e0188826. 10.1371/journal.pone.0188826

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 98

  VaughnS.FletcherJ. M. (2020). Identifying and teaching students with significant Reading problems. Am. Educ.44 (1), 4–11.

  • Google Scholar

• 99

  VaughnS.WanzekJ. (2014). Intensive interventions in Reading for students with Reading disabilities: meaningful impacts. Learn. Disabil. Res. Pract.29 (2), 46–53. 10.1111/ldrp.12031

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 100

  VidyasagarT. R. (2013). Reading into neuronal oscillations in the visual system: implications for developmental dyslexia. Front. Hum. Neurosci.7 (811), 1–10. 10.3389/fnhum.2013.00811

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 101

  WanzekJ.StevensE. A.WilliamsK. J.ScammaccaN.VaughnS.SargentK. (2018). Current evidence on the effects of intensive early Reading interventions. J. Learn. Disabil.51 (6), 612–624. 10.1177/0022219418775110

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 102

  Wygnanski-JaffeT.KushnerB. J.MoshkovitzA.BelkinM.YehezkelO. (2023). An eye-tracking-based dichoptic home treatment for amblyopia: a multicenter randomized clinical trial. Ophthalmology130 (3), 274–285. 10.1016/j.ophtha.2022.10.020

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 103

  Zawacki-RichterO.MarínV. I.BondM.GouverneurF. (2019). Systematic review of research on artificial intelligence applications in higher education: where are the educators?Int. J. Educ. Technol. High. Educ.16, 39. 10.1186/s41239-019-0171-0

  • CrossRef
  • Google Scholar

• 104

  ZhangK.AslanA. B. (2021). AI Technologies for education: recent research and future directions. Comput. Educ.: Artif. Intell.2, 100025. 10.1016/j.caeai.2021.100025

  • CrossRef
  • Google Scholar

• 105

  ZoccolottiP. (2022). Success is not the entire story for a scientific theory: the case of the phonological deficit theory of dyslexia. Brain. Sci.12 (425), 1–11. 10.3390/brainsci12040425

  • CrossRef
  • Google Scholar