INSIGHT : Combining Fixation Visualizations and Residual Neural
Networks for Dyslexia Classification from Eye-Tracking Data

J 2025

INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data

ŠVAŘÍČEK, Roman; Nicol DOSTÁLOVÁ; Jan SEDMIDUBSKÝ a Andrej ČERNEK

Základní údaje

Originální název

INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data

Autoři

ŠVAŘÍČEK, Roman (203 Česká republika, garant, domácí); Nicol DOSTÁLOVÁ (203 Česká republika, domácí); Jan SEDMIDUBSKÝ (203 Česká republika, domácí) a Andrej ČERNEK (703 Slovensko, domácí)

Vydání

DYSLEXIA, Chichester, WILEY, 2025, 1076-9242

Další údaje

Jazyk

angličtina

Typ výsledku

Článek v odborném periodiku

Obor

50301 Education, general; including training, pedagogy, didactics [and education systems]

Stát vydavatele

Velká Británie a Severní Irsko

Utajení

není předmětem státního či obchodního tajemství

Odkazy

URL

Impakt faktor

Impact factor: 1.900 v roce 2023

Organizační jednotka

Filozofická fakulta

DOI

http://dx.doi.org/10.1002/dys.1801

UT WoS

001401970700001

EID Scopus

2-s2.0-85215950318

Klíčová slova anglicky

dyslexia; eye tracking; eye movement; fixation data classification; deep learning; ResNet18; AI-based diagnosis

Štítky

rivok

Příznaky

Mezinárodní význam, Recenzováno

Změněno: 24. 6. 2025 12:28, Mgr. et Mgr. Stanislav Hasil

Anotace

V originále

Current diagnostic methods for dyslexia primarily rely on traditional paper-and-pencil tasks. Advanced technological approaches, including eye-tracking and artificial intelligence (AI), offer enhanced diagnostic capabilities. In this paper, we bridge the gap between scientific and diagnostic concepts by proposing a novel dyslexia detection method, called INSIGHT, which combines a visualization phase and a neural network-based classification phase. The first phase involves transforming eye-tracking fixation data into 2D visualizations called Fix-images, which clearly depict reading difficulties. The second phase utilizes the ResNet18 convolutional neural network for classifying these images. The INSIGHT method was tested on 35 child participants (13 dyslexic and 22 control readers) using three text-reading tasks, achieving a highest accuracy of 86.65%. Additionally, we cross-tested the method on an independent dataset of Danish readers, confirming the robustness and generalizability of our approach with a notable accuracy of 86.11%. This innovative approach not only provides detailed insight into eye movement patterns when reading but also offers a robust framework for the early and accurate diagnosis of dyslexia, supporting the potential for more personalized and effective interventions.

Návaznosti

TL05000177, projekt VaV

Název: Diagnostika dyslexie s využitím eye-trackingu a umělé inteligence (Akronym: DYSLEX)

Investor: Technologická agentura ČR, Diagnostika dyslexie s využitím eye-trackingu a umělé inteligence

Citovat

ŠVAŘÍČEK, Roman; Nicol DOSTÁLOVÁ; Jan SEDMIDUBSKÝ a Andrej ČERNEK. INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data. DYSLEXIA. Chichester: WILEY, 2025, roč. 31, č. 1, s. 1-14. ISSN 1076-9242. Dostupné z: https://dx.doi.org/10.1002/dys.1801.

@article{2466326,
   author = {Švaříček, Roman and Dostálová, Nicol and Sedmidubský, Jan and Černek, Andrej},
   article_location = {Chichester},
   article_number = {1},
   doi = {http://dx.doi.org/10.1002/dys.1801},
   keywords = {dyslexia; eye tracking; eye movement; fixation data classification; deep learning; ResNet18; AI-based diagnosis},
   language = {eng},
   issn = {1076-9242},
   journal = {DYSLEXIA},
   title = {INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data},
   url = {https://doi.org/10.1002/dys.1801},
   volume = {31},
   year = {2025}
}

TY  - JOUR
ID  - 2466326
AU  - Švaříček, Roman - Dostálová, Nicol - Sedmidubský, Jan - Černek, Andrej
PY  - 2025
TI  - INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data
JF  - DYSLEXIA
VL  - 31
IS  - 1
SP  - 1-14
EP  - 1-14
PB  - WILEY
SN  - 10769242
KW  - dyslexia
KW  - eye tracking
KW  - eye movement
KW  - fixation data classification
KW  - deep learning
KW  - ResNet18
KW  - AI-based diagnosis
UR  - https://doi.org/10.1002/dys.1801
N2  - Current diagnostic methods for dyslexia primarily rely on traditional paper-and-pencil tasks. Advanced technological approaches, including eye-tracking and artificial intelligence (AI), offer enhanced diagnostic capabilities. In this paper, we bridge the gap between scientific and diagnostic concepts by proposing a novel dyslexia detection method, called INSIGHT, which combines a visualization phase and a neural network-based classification phase. The first phase involves transforming eye-tracking fixation data into 2D visualizations called Fix-images, which clearly depict reading difficulties. The second phase utilizes the ResNet18 convolutional neural network for classifying these images. The INSIGHT method was tested on 35 child participants (13 dyslexic and 22 control readers) using three text-reading tasks, achieving a highest accuracy of 86.65%. Additionally, we cross-tested the method on an independent dataset of Danish readers, confirming the robustness and generalizability of our approach with a notable accuracy of 86.11%. This innovative approach not only provides detailed insight into eye movement patterns when reading but also offers a robust framework for the early and accurate diagnosis of dyslexia, supporting the potential for more personalized and effective interventions.
ER  -

ŠVAŘÍČEK, Roman; Nicol DOSTÁLOVÁ; Jan SEDMIDUBSKÝ a Andrej ČERNEK. INSIGHT : Combining Fixation Visualizations and Residual Neural Networks for Dyslexia Classification from Eye-Tracking Data. \textit{DYSLEXIA}. Chichester: WILEY, 2025, roč.~31, č.~1, s.~1-14. ISSN~1076-9242. Dostupné z: https://dx.doi.org/10.1002/dys.1801.

Přehled o publikaci