Peer-reviewed article
A Systematic Review of Explainable Multimodal Artificial Intelligence for Early Identification of Developmental Potential in Children
Shyamala B, P Shridhara Acharya, Ananth Prabhu G ·
Summary
Written by AI, not yet AICS's view · claude-sonnet-5-5This is a systematic review of 34 studies published between 2016 and 2025 on explainable, multimodal artificial intelligence for developmental screening and assessment in children. The studies combine data such as speech, behaviour, physiology and vision. The review looks at model approaches, data fusion, explainability, and how trust, ethics, bias and generalisation are evaluated. It reports that explanations aimed at clinicians and educators were common and that diagnostic and predictive accuracy stayed high. It also says that data availability, long-term studies, and cultural and demographic bias remain problems.
Key points
- 34 studies were selected under the PRISMA model, covering 2016 to 2025.
- The reviewed work involves data fusion, explainable AI, and stakeholder-driven explainable AI.
- Explanations for clinicians and educators dominated, and high diagnostic and predictive accuracy was maintained.
- Availability of data and long-term (longitudinal) studies is a persistent issue.
- Accountability for cultural and demographic bias is a persistent issue.
Limitations
Only the abstract is available, so details on the individual studies, their quality, the age groups and the settings are not given. The abstract does not state which countries the studies cover, so the India hint could not be confirmed.
Abstract
Children who are developmentally lagging require quick and tailored supports- this is why focusing on potential development is important. AI has advanced by integrating multiple forms of speech, behaviour, physiology, and vision, and emerging AI systems for speech triage and multi-modal systems are demonstrating promise for comprehensive pediatric assessment. This literature review examines the most recent XAI approaches and applications to developmental screening and assessment of learning and development from 2016-2025. Under the PRISMA model, 34 studies were selected that involve data fusion, explainable AI, and stakeholder-driven explainable AI. The review explores various AI model approaches, data fusion, explainable AI, and evaluation with explainable AI trust, ethical, bias, and generalization frameworks. Survey results of the various frameworks show that XAI approaches and selected strikingly high attribution approaches of clinician educator explanation dominated stakeholder-controlled assessment and understanding and maintained high diagnostic and predictive accuracy levels. The issues of availability and sustaining longitudinal studies and accountability for cultural and demographic bias are persistent. This review suggests discussing adaptive and transparent AI systems that are equitable, scalable, and reliable. This synthesis aids the creation of efficient and useful AI technologies for fair access to early childhood assessment and intervention tools.