Chapter 7. Phase 4 β€” Integration

Hongik Ingan (εΌ˜η›ŠδΊΊι–“)

"Benefit All Humanity"

Emotion AI delivers genuine value when it is integrated into real-world applications that improve human well-being, learning, and experience. The data format, API, and streaming protocol do not by themselves create value; the standard is realised only when those layers reach end users in meaningful ways. This chapter provides integration guidance for five domains β€” healthcare, education, marketing, automotive, and gaming / XR β€” together with the alignment to the principal regulatory regimes (the EU Medical Device Regulation, FDA Software-as-a-Medical-Device guidance, the US HIPAA, GDPR, the EU AI Act, ISO 26262 functional safety, and IEEE 1484 / ADL SCORM education standards).

7.1 Overview

Phase 4 provides integration guidance for deploying emotion AI across multiple domains. Each domain has distinctive requirements, ethical considerations, and best-practice procedures. The guidance in this chapter is more than advisory: it operates as the mandatory checklist for domain-specific certification, so a Phase 4 conformance audit applies different test procedures depending on the deployment domain.

Table 7-1. Five WIA Phase 4 integration domains and their core considerations
DomainPrincipal use caseCore consideration
HealthcareMental-health monitoring, therapy assistanceHIPAA / EU MDR / SaMD; patient privacy; clinical validation
EducationEngagement detection, adaptive learningStudent privacy; parental consent; age-appropriateness
MarketingAdvertising effectiveness, consumer researchConsent; transparency; data minimisation
AutomotiveDriver monitoring, safety alertsSafety-critical; real-time; ISO 26262 / EU GSR
Gaming and XRImmersive experience, NPC reactionPrivacy; user experience; opt-out

Domain-specific certification shares fifty to eighty per cent of its test items with the general certification suite, supplemented by domain-specific items. Healthcare adds clinical-validation and SaMD-classification mapping; automotive adds ISO 26262 ASIL-B reliability tests; education adds parental-consent flow and opt-out tests; marketing adds pseudonymisation and disclosure tests; gaming adds age-rating and minor-protection tests. Multi-domain applications (for example, automotive plus healthcare for an in-cabin mental-state monitor) must pass both domains' tests; where domain requirements conflict, the stricter requirement applies.

7.2 Healthcare Integration

7.2.1 Mental-Health Monitoring

Table 7-2. Mental-health applications, emotion signals, and clinical use
ApplicationEmotion signalClinical use
Depression screeningLow valence, flat affect, reduced AU activityEarly detection; treatment monitoring
Anxiety detectionHigh arousal, fear pattern, voice tremorTherapy-session insight
PTSD assessmentFear response; hyper-vigilance markersTrigger identification
Autism supportEmotion-expression patternsSocial-skills training
Burn-out monitoringSustained low arousal and low valenceRecovery recommendation

Health-grade emotion AI is regulated as Software as a Medical Device (SaMD) in most jurisdictions. The International Medical Device Regulators Forum (IMDRF) framework classifies SaMD by its intended healthcare situation and the seriousness of the condition into four classes (I–IV).[1] An emotion-AI tool that informs treatment decisions for a serious psychiatric condition is typically Class III; one that drives autonomous treatment decisions is Class IV. The corresponding national clearances β€” the FDA 510(k) or De Novo route in the United States, EU MDR (Regulation (EU) 2017/745) Class IIa or higher in the European Union, and the equivalent national medical-device clearance elsewhere β€” are obligations layered on top of WIA conformance, not replaced by it.

7.2.2 Telehealth Integration

Figure 7-1. Telehealth session integration β€” patient camera, local analysis, clinician dashboard
Telehealth session integration:

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚                   Video-call platform                    β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚                                                          β”‚
β”‚   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”    β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚
β”‚   β”‚ Patient │───▢│ WIA Emotion AI  │───▢│ Clinician   β”‚ β”‚
β”‚   β”‚ camera  β”‚    β”‚ analysis (local)β”‚    β”‚ dashboard   β”‚ β”‚
β”‚   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚
β”‚                           β”‚                              β”‚
β”‚                           β–Ό                              β”‚
β”‚                  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                     β”‚
β”‚                  β”‚ Session summary β”‚                     β”‚
β”‚                  β”‚ (clinical note) β”‚                     β”‚
β”‚                  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                     β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Data flow:
1. Patient video is analysed locally (privacy)
2. Only emotion indicators reach the clinician
3. The original video is not retained
4. A session summary is auto-generated for clinical record

Telehealth deployments must respect HIPAA in the United States and the EU MDR in the European Union. The Phase 4 healthcare adapter writes the clinical-note section in a structured form compatible with HL7 FHIR R5 (Health Level Seven Fast Healthcare Interoperability Resources, Release 5).[2] The emotion-AI tool informs but does not replace the clinician; the human-overseer requirement of EU AI Act Article 14 mandates that the clinician retain authority to override the system's output.

7.2.3 FHIR Mapping

WIA Phase 1 emotion records map onto FHIR Observation resources with a code drawn from the LOINC vocabulary for mental-status observations. The recommended mapping uses LOINC codes for mood / affect observation (LOINC 71164-9 et seq.) with the emotion label and confidence carried in the valueCodeableConcept and component fields, respectively.[3]

7.3 Education Integration

7.3.1 Engagement Detection

In educational deployments, the WIA standard requires parental consent for subjects under fourteen years of age (in line with PIPA Article 22-2 and US COPPA), an opt-out path that does not penalise the student, and adherence to the relevant local student-rights regulation. Engagement detection must be presented to teachers as supplementary information rather than as a basis for grading; the EU AI Act prohibits the use of emotion recognition in education to infer emotion for the purpose of evaluation that affects access to opportunity.[4]

7.3.2 SCORM and IEEE 1484 Integration

Adaptive-learning platforms exchange learner records using ADL SCORM 2004 4th Edition and IEEE 1484.20.1 (Reusable Competency Definitions).[5] The WIA Phase 4 education adapter emits emotion-derived engagement signals as SCORM-compatible interaction records and IEEE 1484-compatible competency-evidence records, so that emotion-aware adaptation may be added to existing learning-management infrastructure without rewriting the LMS.

7.3 Marketing Integration

In marketing deployments, three obligations dominate: explicit consent (visual and auditory disclosure that emotion analysis is in operation); pseudonymisation of subject identifiers (Table 4-5); and aggregate-only retention (no individually identifiable emotion data persists beyond the session). GDPR Article 22 limits automated decision-making based on emotional inference, so marketing-domain deployments in the European Union typically operate in advisory rather than decisional mode.

Aspect-based sentiment analysis on social-media text is the lowest-risk marketing use case and is operated under most data-protection regimes without consent-management complexity, provided that the underlying text was lawfully sourced (a public post, a survey response with explicit consent, or aggregated analytics). In-store affect monitoring through cameras is considerably more complex: it almost always falls under biometric-data rules (GDPR Article 9; PIPA Article 23; CCPA / CPRA sensitive personal information) and requires opt-in consent or operates in fully aggregated form with no individual record retention.

7.4 Automotive Integration

7.4.1 Driver Monitoring System

Table 7-3. Driver-monitoring affect targets and intervention mapping
Target stateDetection signalsIntervention
DrowsinessAU43 (eye closure), low arousal, head pose driftAudible alert; lane-keep assist activation
DistractionOff-axis gaze, reduced facial activityVisual alert on instrument cluster
Anger / road-rageAU4 + AU7 + AU23, voice intensity spikeDe-escalation cue (calming audio)
FatigueSustained low arousal across the tripRecommend stop / coffee break

7.4.2 Safety-Critical Reliability

Automotive emotion AI runs under ISO 26262 ASIL-B reliability requirements at minimum and ASIL-C or ASIL-D for higher-autonomy vehicles.[6] The relevant European regulation (EU 2019/2144 β€” General Safety Regulation, GSR) mandates driver-attention warning systems on all new types from 6 July 2022 and on all new registrations from 7 July 2024; emotion AI is one of the underlying technologies expected to satisfy this requirement.[7]

Latency budgets are stricter than in non-safety domains: the closed-loop response from detection to driver alert must remain below one hundred and fifty milliseconds, leaving little headroom for cloud round-trips. WIA-conformant automotive deployments therefore typically perform inference at the edge (in-cabin ECU or roof-mounted compute), with the cloud used only for model-update download and aggregate fleet telemetry. The on-device WIA Phase 1 record is identical to the cloud variant, ensuring portability.

7.5 Gaming and XR Integration

Game and extended-reality applications leverage emotion AI for adaptive difficulty, NPC reaction, and immersive emotional storylines. In Korea and the United States, age-rating systems (the relevant national game-rating board; ESRB in North America; PEGI in Europe) impose disclosure obligations when an entertainment product captures biometric or affect data from minors. The WIA Phase 4 gaming adapter emits a structured "biometric-data-collection" disclosure record that maps onto each rating board's disclosure form, simplifying the rating-submission process.[8]

Privacy-preserving alternatives are particularly important in gaming because the player population skews young. The WIA Phase 4 gaming adapter supports an "ephemeral" mode in which no emotion record is persisted beyond the active game session β€” only aggregate state (e.g. "frustration spike detected") flows to the difficulty-adjustment subsystem.

7.6 Cross-Domain Considerations

Table 7-4. Cross-domain considerations and the standard's response
ConcernCross-domain patternStandard response
ChildrenEducation, gaming, healthcareParental consent (PIPA Art. 22-2; US COPPA); opt-out
Sensitive demographicsHealthcare, marketing, automotiveBias auditing; demographic-decomposed accuracy reporting
Multi-jurisdictional flowAll domainsPer-region endpoint; X-WIA-Region header; adequacy mapping
SafetyAutomotive, healthcareMultimodal mandatory; ISO 26262 / SaMD cross-walk

7.7 Note on Korean Edition Content

The Korean edition of this volume contains additional sections covering the Korean integration environment in detail: the relevant national mental-health legal regime, named domestic mental-health platforms (which operate as advisory-grade rather than diagnostic-grade systems), the Korean telehealth pilot programme (limited initially to psychiatry), the relevant national consumer-protection regulations for biometric data in retail, the relevant national automotive-safety regulations and the named domestic automotive suppliers' driver-monitoring systems, the Korean Game Rating and Administration Committee disclosure forms, and named domestic public-broadcaster digital-textbook platforms.

This English edition deliberately abstracts those passages. References to specific Korean statutes, agencies, broadcasters, hospitals, automotive suppliers, mental-health platforms, or rating boards become "the relevant national mental-health legal regime", "leading domestic mental-health platforms", "the Korean telehealth pilot programme", "the relevant national consumer-protection regulator", "leading domestic automotive suppliers", and "the relevant national rating board". The conformance requirements themselves are identical between the two editions.

7.8 Chapter Summary

Seven key takeaways.

  1. Five domains. Healthcare, education, marketing, automotive, gaming.
  2. Healthcare. SaMD classification, FHIR mapping, telehealth alignment.
  3. Education. Parental consent, COPPA, opt-out without penalty, IEEE 1484 / SCORM.
  4. Marketing. GDPR Art. 22 advisory mode; pseudonymisation; aggregate-only retention.
  5. Automotive. ISO 26262 ASIL-B / GSR; sub-150 ms latency; edge inference.
  6. Gaming. Rating-board disclosure; ephemeral mode for minor protection.
  7. Cross-domain. Multi-jurisdictional applications follow the stricter requirement.

7.9 Review Questions

  1. List the four IMDRF SaMD classes and assign each to a representative emotion-AI use case.
  2. Explain how a Phase 1 emotion record maps onto an HL7 FHIR R5 Observation resource.
  3. Describe why automotive deployments typically perform inference at the edge.
  4. Compare GDPR Article 22 advisory mode and decisional mode in marketing applications.
  5. Explain how the EU GSR and ISO 26262 jointly constrain automotive integration.
  6. Describe the role of the rating-board disclosure record in gaming integration.
  7. Identify a multi-domain application and explain how the "stricter requirement applies" rule resolves conflicts.

7.10 Looking Ahead

Chapter 8 turns to implementation and certification. Where Chapters 4–7 specify what conformance means, Chapter 8 specifies how to obtain it: certification-test procedure, audit checklist, and the cross-walk to NIST AI RMF, EU AI Act, and other regulatory mappings. The simulator's πŸ§ͺ Emotion Test panel (Panel 4) becomes the practical test bench for the procedures described there. The standard's evolution roadmap is recorded in the public GitHub repository.[99]

Chapter 7 Endnotes

  1. International Medical Device Regulators Forum. (2014). "Software as a Medical Device": Possible Framework for Risk Categorization and Corresponding Considerations. IMDRF/SaMD WG/N12 FINAL:2014. IMDRF SaMD framework. ↑
  2. HL7 International. (2023). HL7 FHIR Release 5 (R5). https://hl7.org/fhir/R5/. ↑
  3. LOINC. Logical Observation Identifiers Names and Codes. Regenstrief Institute. https://loinc.org/. ↑
  4. European Union. (2024). Regulation (EU) 2024/1689 β€” AI Act Articles 5(1)(d), 6, and 14. https://eur-lex.europa.eu/eli/reg/2024/1689/oj. ↑
  5. Advanced Distributed Learning Initiative. (2009). SCORM 2004 4th Edition Specification. https://adlnet.gov/projects/scorm-2004-4th-edition/. IEEE 1484.20.1-2007. Reusable Competency Definitions. ↑
  6. ISO 26262 series. Road vehicles β€” Functional safety. International Organization for Standardization. ASIL-B reliability targets. ↑
  7. European Union. (2019). Regulation (EU) 2019/2144 β€” General Safety Regulation (GSR). Driver-attention warning is an obligated subsystem under Annex II. https://eur-lex.europa.eu/eli/reg/2019/2144/oj. ↑
  8. ESRB Privacy Certified Programme. https://www.esrb.org/. PEGI: Pan-European Game Information. https://pegi.info/. ↑
  9. European Union. (2017). Regulation (EU) 2017/745 β€” Medical Device Regulation (MDR). https://eur-lex.europa.eu/eli/reg/2017/745/oj.
  10. FDA. (2022). Clinical Decision Support Software β€” Guidance for Industry and Food and Drug Administration Staff.
  11. US Public Law 104-191. Health Insurance Portability and Accountability Act of 1996 (HIPAA).
  12. US Federal Trade Commission. (1998). Children's Online Privacy Protection Act (COPPA).
  13. European Parliament & Council. (2016). Regulation (EU) 2016/679 (GDPR).
  14. ISO/IEC 27799:2016. Health informatics β€” Information security management in health using ISO/IEC 27002.
  15. IETF RFC 6749. (2012). The OAuth 2.0 Authorization Framework.
  16. WIA Standards public repository (emotion-ai folder), MIT-licensed source for the simulator, specification, API reference, and ebook assets cited throughout this volume: WIA-Official/wia-standards-public/tree/main/emotion-ai. The standard's evolution roadmap, revision history, and SDK source code are maintained openly in this repository, where the WIA standards committee records its formal verification of all primary sources cited in this chapter. ↑