Prediction of Dyslexia from Eye Movements Using Machine Learning

Prabha, A. Jothi; Bhargavi, R.

doi:10.1080/03772063.2019.1622461

Cited by 34 publications

(21 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Perera et al [47] proposed a cubic SVM model with 78.2% accuracy, 88.2% sensitivity and 66.7% specificity using the brain pattern signal features extracted from the EEG dataset. Prabha and Bhargavi [32] proposed SVM and SVM-PSO models for linear and hybrid kernels, respectively. These models achieved the highest classification accuracy of 90% and 95% respectively.…”

Section: Discussion and Potential Challengesmentioning

confidence: 99%

“…The Brain Extraction Tool (BET), the FMRIB Diffusion Toolbox (FDT) and the Tract-Based Spatial Statistics (TBSS) function software [64] were used to obtain fractional anisotropy (FA) values from DTI dataset collected by Chimeno et al [37]. Statistical measures were implemented on a computer system by Prabha and Bhargavi [32], [33] to extract different fixation events and saccades eye movements from the raw eye tracking dataset. Following the concept of DeepWriter proposed by Xing and Qiao [65], Spoon et al [40], [41] generated 50 randomly patches of handwriting features from each image handwriting dataset used.…”

Section: Data Preparation and Preprocessingmentioning

confidence: 99%

“…For classification of dyslexia and its biomarkers, construction and training of models is achieved using machine learning or deep learning algorithms. While early studies focused on a single traditional machine learning algorithm [3], [42], [43], [47], [50], [59], [60], other studies developed hybrid methods for dyslexia detection [32], [44], [61]. The implementation of deep learning algorithms are the most recent studies in dyslexia and its biomarkers detection researches [57], [58].…”

Section: E Machine Learning Training and Classificationmentioning

confidence: 99%

“…The implementation of deep learning algorithms are the most recent studies in dyslexia and its biomarkers detection researches [57], [58]. Meanwhile, majority of the studies compared different types of machine learning methods using the same dataset [32], [37], [44]- [46], [52], [55], [56], [62]. It is a usual practice to partition datasets into two during the training of machine learning models: training set (containing larger percentage of the entire data, ≥70% in most cases), testing or validation set (containing 30% or lesser of the dataset).…”

Section: E Machine Learning Training and Classificationmentioning

confidence: 99%

“…For eye movement tracking, statistical measures such as dispersion threshold identification, velocity threshold identification algorithms and eye trackers have been utilized to detect learner's eye movement features during cognitive exercise. These features include fixations and saccades, used by recent studies to classify learners into dyslexic and non-dyslexic [32], [33]. Both MRI and EEG scans are used to generate and detect images showing alterations in learner's brain normal morphology with electrical impulses within it.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Advance Machine Learning Methods for Dyslexia Biomarker Detection: A Review of Implementation Details and Challenges

et al. 2021

View full text Add to dashboard Cite

Dyslexia is a neurological disorder that is characterized by imprecise comprehension of words and generally poor reading performance. It affects a significant population of school-age children, with more occurrences in males, thus, putting them at risk of poor academic performance and low selfesteem for a lifetime. The long-term hope is to have a dyslexia diagnostic method that is informed by neural-biomarkers. In this regard, large numbers of machine learning methods and, more recently, deep learning methods have been implemented across various types of dataset with the above-chance classification accuracy. However, attainment of clinical acceptability of these state-of-the-art methods is bedeviled by certain challenges including lack of biologically-interpretable biomarkers, privacy of dataset and classifiers, hyper-parameter selection/optimization, and overfitting problem among others. This review paper critically analyzes recent machine learning methods for detecting dyslexia and its biomarkers and discusses challenges that require proper attentions from the users of deep learning methods in order to enable them to attain clinically relevance and acceptable level. The review is conducted within the premise of implementation and experimental outcomes for each of the 22 selected articles using the Preferred Reporting Items for Systematic review and Meta-Analyses (PRISMA) protocol, with a view to outlining some critical challenges for achieving high accuracy and reliability of the state-of-the-art machine learning methods. As an evidence-based protocol for reporting in systematic reviews and meta-analyses, PRISMA helps to ensure clarity and transparency of this paper by showing a four-phase flow diagram of the selection process for articles used in this review. It is therefore envisaged that higher classification performance of clinical relevance can be achieved using deep learning models for dyslexia and its biomarkers by addressing identified potential challenges.

show abstract

Section: Discussion and Potential Challengesmentioning

confidence: 99%