Detection of Children/Youth With Fetal Alcohol Spectrum Disorder Through Eye Movement, Psychometric, and Neuroimaging Data

Chen Zhang, Angelina Paolozza, Po-He Tseng, James N. Reynolds, Douglas P. Munoz, Laurent IttiFrontiers in Neurology 2019 · DOI

TL;DR

This paper introduces a novel, multimodal machine learning approach to screen for Fetal Alcohol Spectrum Disorders (FASD) in children and youth. By combining eye-tracking, psychometric, and neuroimaging data, they developed a high-performing, low-cost screening protocol that significantly improves upon existing diagnostic methods, paving the way for earlier intervention and better long-term outcomes.

What Problem Does This Paper Address?

Fetal Alcohol Spectrum Disorders (FASD) are a leading cause of preventable developmental disabilities, yet diagnosing them is a significant challenge. The current diagnostic process is time-consuming, expensive, and often requires a multidisciplinary team, leading to underdiagnosis and delayed intervention. This gap in early identification means that individuals with FASD may not receive the support they need, potentially exacerbating neurodevelopmental deficits and increasing the risk of secondary comorbidities. The authors aim to bridge this gap by developing an objective, efficient, and cost-effective screening tool.

Methods

The researchers employed a machine learning framework to analyze data from multiple assessment modalities. Their approach involved:

  • Multimodal Data Integration: They collected data from six different assessments:
  • Saccadic Eye Movement Tasks: Prosaccade (ProSac), Antisaccade (AntiSac), and Memory-Guided Saccade (MGSac) tasks to assess sensorimotor control, attention, and inhibition.
  • Natural Viewing Task: Participants watched short video clips while their eye movements were tracked to analyze visual attention and scene processing.
  • Psychometric Tests: Standardized tests (NEPSY-II, Working Memory Test Battery, Woodcock Johnson III, Woodcock Reading Mastery Test) to evaluate cognitive functions like attention, executive function, memory, and academic skills.
  • Diffusion Tensor Imaging (DTI): Neuroimaging of the corpus callosum to assess white matter integrity.
  • Machine Learning Classification: They utilized Support Vector Machine-Recursive Feature Elimination (SVM-RFE) for single-assessment classification and feature selection. For multimodal classification, they employed logistic regression (LR) and a two-layer LR framework for natural viewing data.
  • Feature Engineering: For eye-tracking data, they extracted numerous features related to accuracy, reaction time, velocity, and error types. For DTI, they derived features from different regions of the corpus callosum. Psychometric tests yielded scores from various subtests.
  • Cost-Benefit Analysis: They conducted an analysis to identify a subset of assessments that offered a good balance between classification accuracy, cost, and administration time, leading to a proposed low-cost screening protocol.
  • Value of Information Theory: This theory was used to model the expected benefits of early screening and to propose an optimal annual screening procedure.

Datasets

The study utilized data from participants recruited through a large, multi-site investigation funded by the Kids Brain Health Network.

  • Participants: 91 children/youth diagnosed with FASD and 116 age-matched typically developing controls, aged 5-18 years.
  • Data Collection: Data was collected across five different communities in Canada and at the University of Southern California.
  • New Benchmarks: While not explicitly introducing a new benchmark dataset in the traditional sense, the paper presents a novel multimodal dataset that combines eye-tracking, psychometric, and neuroimaging data for FASD classification. This comprehensive dataset allows for the exploration of inter-assessment relationships and the development of integrated screening tools.

Key Insights and Ideas

  • Multimodality is Key: Combining data from different assessment types (eye movements, cognitive tests, brain imaging) significantly boosts classification accuracy compared to any single modality. This suggests that FASD manifests across a broad spectrum of cognitive and sensorimotor functions.
  • Eye Movements as a Powerful Biomarker: Saccadic eye movement tasks and natural viewing eye-tracking data proved to be highly effective in distinguishing individuals with FASD from controls, even when used alone. This highlights the potential for non-invasive, relatively low-cost eye-tracking technologies in screening.
  • The Corpus Callosum's Role: DTI findings confirmed structural abnormalities in the corpus callosum, particularly in posterior regions, which are consistent with known visuospatial processing deficits in FASD. However, the study also suggested that anterior regions might play a more significant, previously underappreciated role.
  • Optimizing for Practicality: The research moved beyond simply achieving the highest accuracy to developing a practical screening protocol. By analyzing cost-benefit trade-offs, they identified a subset of assessments that offer a strong balance of accuracy, speed, and accessibility.
  • Value of Information for Screening: Applying the "value of information" framework provided a quantitative basis for advocating for annual screening, demonstrating substantial potential cost savings and improved outcomes for individuals at risk.

Key Results

  • Single Assessment Accuracy: The best single assessment classification accuracy was achieved by psychometric tests (78.26%), followed closely by saccadic eye movement tasks (AntiSac: 76.09%) and natural viewing (71.74%). DTI yielded a lower accuracy of 67.39%.
  • Multimodal Accuracy: Combining all six assessments resulted in a classification accuracy of 84.78%, a significant improvement over individual assessments.
  • Optimized Multimodal Protocol: A combination of ProSac, AntiSac, natural viewing, and psychometric tests achieved a high accuracy of 84.78% (after dropping MGSac and DTI), suggesting this subset is sufficient for robust classification.
  • Pairwise Assessment Accuracy: The combination of AntiSac and psychometric assessments yielded the highest pairwise accuracy at 82.61%.
  • Low-Cost Screening Protocol: A screening procedure using ProSac and Natural Viewing achieved a respectable accuracy of 78.26%, with sensitivity of 77.27% and specificity of 79.17%. Adding AntiSac or a short psychometric battery further improved accuracy.
  • Cost-Benefit: The proposed low-cost screening procedure, costing less than $50 per individual, is estimated to save millions of dollars annually for a population of 1,000 individuals by facilitating earlier intervention.

Limitations and Future Work

  • Generalizability: While the study used independent testing sets and cross-validation, the authors acknowledge that a larger dataset with more diverse participants would further enhance the generalizability of their findings.
  • Confounding Factors: The high accuracy of psychometric tests might be partly due to their overlap with clinical diagnostic criteria. The authors note that math ability, a top feature, was not directly part of the diagnostic ground truth for their sample, but inhibition and digit recall (working memory) could be potential confounders.
  • Video Snippet Selection: For the natural viewing task, only about half of the video snippets contributed significantly to classification. Future work could involve refining the selection of video content to further optimize efficiency.
  • Future Work: The authors propose an optimal annual screening procedure for children at risk of FASD, emphasizing its potential for widespread deployment in clinics and schools. They also suggest further evaluation of the "value of information" perspective to refine cost-benefit models for screening. The study also points to potential roles for anterior corpus callosum regions in FASD that warrant further investigation.

Summary updated

Abstract

<b>Background:</b> Fetal alcohol spectrum disorders (FASD) is one of the most common causes of developmental disabilities and neurobehavioral deficits. Despite the high-prevalence of FASD, the current diagnostic process is challenging and time- and money- consuming, with underreported profiles of the neurocognitive and neurobehavioral impairments because of limited clinical capacity. We assessed children/youth with FASD from a multimodal perspective and developed a high-performing, low-cost screening protocol using a machine learning framework. <b>Methods and Findings:</b> Participants with FASD and age-matched typically developing controls completed up to six assessments, including saccadic eye movement tasks (prosaccade, antisaccade, and memory-guided saccade), free viewing of videos, psychometric tests, and neuroimaging of the corpus callosum. We comparatively investigated new machine learning methods applied to these data, toward the acquisition of a quantitative signature of the neurodevelopmental deficits, and the development of an objective, high-throughput screening tool to identify children/youth with FASD. Our method provides a comprehensive profile of distinct measures in domains including sensorimotor and visuospatial control, visual perception, attention, inhibition, working memory, academic functions, and brain structure. We also showed that a combination of four to six assessments yields the best FASD vs. control classification accuracy; however, this protocol is expensive and time consuming. We conducted a cost/benefit analysis of the six assessments and developed a high-performing, low-cost screening protocol based on a subset of eye movement and psychometric tests that approached the best result under a range of constraints (time, cost, participant age, required administration, and access to neuroimaging facility). Using insights from the theory of value of information, we proposed an optimal annual screening procedure for children at risk of FASD. <b>Conclusions:</b> We developed a high-capacity, low-cost screening procedure under constrains, with high expected monetary benefit, substantial impact of the referral and diagnostic process, and expected maximized long-term benefits to the tested individuals and to society. This annual screening procedure for children/youth at risk of FASD can be easily and widely deployed for early identification, potentially leading to earlier intervention and treatment. This is crucial for neurodevelopmental disorders, to mitigate the severity of the disorder and/or frequency of secondary comorbidities.
BibTeX
@inproceedings{zhang2019detection,
  author = {Chen Zhang and Angelina Paolozza and Po-He Tseng and James N. Reynolds and Douglas P. Munoz and Laurent Itti},
  title = {{Detection of Children/Youth With Fetal Alcohol Spectrum Disorder Through Eye Movement, Psychometric, and Neuroimaging Data}},
  booktitle = {Frontiers in Neurology},
  year = {2019},
  doi = {10.3389/fneur.2019.00080},
}
Figure