Treffer: A smart community interactive art therapy platform based on multimodal computer graphics and resilient artificial intelligence for home-based elderly care.

This research presents an innovative smart community interactive art therapy platform that integrates multimodal computer graphics with resilient artificial intelligence adaptation mechanisms to address the growing challenges of home-based elderly care. The platform employs a four-layered hierarchical architecture encompassing perception, network, platform, and application layers to deliver personalized therapeutic interventions. The system utilizes multimodal data fusion algorithms to process visual, auditory, and haptic inputs while implementing adaptive learning mechanisms that continuously optimize user experiences based on individual preferences and capabilities. Experimental validation demonstrates superior performance with response times averaging 387 ms under 100 concurrent users, therapeutic recommendation accuracy of 87.3%, and user satisfaction scores of 4.2/5.0 across multiple evaluation dimensions. The resilient adaptation mechanisms achieved 99.7% service availability and 34% improvement in CPU utilization compared to conventional systems. Long-term usage tracking revealed sustained engagement patterns with minimal dropout rates over 6-month evaluation periods. The platform successfully addresses key limitations of traditional elderly care models by providing comprehensive support that encompasses cognitive stimulation, emotional well-being, and social connection while maintaining cost-effectiveness and scalability for large-scale deployment in smart community environments.
(© 2025. The Author(s).)

Declarations. Competing interests: The authors declare no competing interests. Ethical approval and data privacy protection: This research was conducted in accordance with ethical guidelines and received approval from the Institutional Review Board of Silla University (Protocol Number: SU-IRB-2024-037, approved on March 15, 2024, annual continuing review required). The study involved computational validation using synthetic user models and system performance testing without direct human subject participation during primary validation phase, with pilot real-world deployment (n = 5, 4 weeks) requiring full informed consent procedures described below. All data collection and analysis procedures adhered to institutional research ethics standards, Korean Personal Information Protection Act (PIPA) requirements, and privacy protection requirements outlined in international guidelines. Informed consent and vulnerable population protections: Cognitive impairment in elderly participants complicates capacity assessment for informed consent. We implemented tiered consent process: (1) Simplified consent document written at 8th-grade reading level (Flesch-Kincaid score 8.2) with large 16pt font, avoiding technical jargon, highlighting key information (purpose, duration, risks, benefits, voluntary nature) in colored boxes, and limiting length to 3 pages versus typical 12-page consent forms; (2) Multimedia consent presentation using 5-minute video with voice narration and visual illustrations explaining study procedures, showing sample interface interactions, and emphasizing withdrawal rights without penalty; (3) Teach-back method where participants explained study back to researcher in own words, with questions like “Can you tell me what you’ll be asked to do in this study?” and “What happens if you want to stop participating?”, with comprehension scoring requiring 80% accuracy (4/5 key points correct) before proceeding; (4) Surrogate consent pathway for participants lacking decision-making capacity (MMSE < 18 or failed teach-back assessment), requiring legally authorized representative approval plus participant assent obtained by asking “Are you willing to try the art activities on the tablet?” and observing for verbal agreement or behavioral assent (nodding, reaching for tablet). Capacity reassessment occurred every 4 weeks given potential cognitive fluctuation in elderly populations. Data privacy risks from multimodal biometric collection: The platform’s multimodal data collection creates four primary privacy risks requiring specific mitigation strategies: Risk 1—Re-identification from facial data: Facial emotion recognition captures detailed facial geometry that could re-identify participants even after removing explicit identifiers. Mitigation: Facial landmarks (68 3D coordinate points) are extracted and original images deleted within 24 h of capture per automated data lifecycle policy, differential privacy with ε = 1.0 (epsilon parameter controlling privacy-utility tradeoff) applied to aggregate facial expression statistics ensuring individual-level facial patterns cannot be reverse-engineered from aggregated data, all visual data stored on encrypted servers (AES-256 encryption standard with 256-bit keys), and quarterly access audits review all 847 researcher queries to facial feature database with anomalous access patterns (e.g., downloading > 50 individual records) triggering automatic investigation. Risk 2—Voice print sensitivity: Voice recordings contain health indicators including Parkinsonian tremor, respiratory conditions, and emotional distress not explicitly consented for analysis. Mitigation: Voice features (40 Mel-Frequency Cepstral Coefficients per frame) extracted and raw audio deleted immediately after processing, speaker de-identification applies pitch-shift normalization (± 20% random shift) rendering voices unrecognizable while preserving emotional content, and opt-out option allows users to disable voice input entirely using text-based alternatives with no functionality loss (85% of features accessible via text, 15% voice-specific features like music rhythm games appropriately excluded). Risk 3—Behavioral pattern inference: Interaction logs (timestamped actions, error patterns, session durations) could reveal daily routines enabling unauthorized surveillance or identifying cognitive decline patterns participants may wish to keep private from family members. Mitigation: Data minimization principle implemented with only 23 essential metrics logged (vs. initial design of 67 metrics, reduced after privacy impact assessment), temporal aggregation replaces timestamped individual events with daily summary statistics (e.g., “completed 3 painting sessions on Tuesday” vs. “started session 9:23am, ended 9:47am”), and right to deletion enables users to request full data erasure within 30 days with automated verification confirming removal from all systems including backups. Risk 4—Data breach consequences: Server compromise would expose vulnerable elderly population to identity theft, embarrassment from emotional expression data, and potential discrimination based on cognitive status. Mitigation: Defense-in-depth security architecture implements encryption at rest (AES-256 for all databases) and in transit (TLS 1.3 with perfect forward secrecy for all network traffic), multi-factor authentication required for all researcher access (password + hardware token + biometric), intrinsic incident response plan mandates < 24-hour notification to participants if breach occurs (contact list maintained with 95% updated phone numbers verified quarterly), cyber insurance policy ($2 million coverage) provides financial protection and breach remediation services, and annual penetration testing by independent security firm (last conducted April 2024, identified and resolved 3 medium-severity vulnerabilities). Ethical oversight and continuous monitoring: Independent Data Safety Monitoring Board (DSMB) comprising three members (geriatrician, bioethicist, biostatistician) unaffiliated with research team convenes quarterly reviewing adverse events (0 serious adverse events, 3 mild adverse events in pilot study: temporary frustration, eye strain, finger fatigue), privacy incidents (0 breaches, 2 participant requests for data deletion honored within 23 days), and recruitment/retention metrics, with authority to suspend study if concerns arise (stopping rules: >10% serious adverse events, > 3 privacy breaches, > 50% dropout rate). Institutional Review Board annual continuing review requires submission of updated consent documents, cumulative adverse event reports, and protocol modifications, with most recent approval February 2025 extending approval through February 2026. Participant advisory board comprising 8 elderly users (selected from pilot study volunteers) consulted bi-monthly on design decisions including interface modifications (e.g., recommended increasing font size from 18pt to 24pt), acceptable data collection practices (unanimously approved anonymous usage statistics sharing but rejected location tracking proposals), and research dissemination preferences (requested plain-language summary of findings sent to all participants). Vulnerable population specific protections: No coercion safeguards emphasized voluntary participation through repeated reminders that “your decision will not affect your access to community center services” and “there is no penalty for refusing or stopping”, with recruitment materials reviewed by elderly advisory board confirming clarity. Continuous consent protocol implements monthly check-ins asking “Do you still want to continue participating?” rather than assuming ongoing consent, recognizing that elderly individuals may feel obligated to continue once committed. Burden minimization limits sessions to 30 min maximum duration preventing fatigue (adaptive system automatically prompts break after 25 min if user continues), and distress protocol triggers when system detects severe emotional distress through facial expression analysis (crying detected via facial action units 1 + 4 + 15 sustained > 10 s, anger via units 4 + 5 + 7 + 23): pause activity immediately displaying calming blue screen, show support resource information (crisis hotline numbers, on-site staff contact), automatic notification sent to study coordinator’s phone, and mandatory follow-up call within 24 h (occurred 0 times in pilot study, system tested via simulated distress scenarios). Data sharing and open science principles: De-identified dataset will be shared via Open Science Framework (OSF) repository after 2-year embargo period protecting competitive advantage for publications, including 300 synthetic user interaction logs (csv format, 847 variables, comprehensive data dictionary with variable definitions and coding schemes), aggregated performance metrics (group-level statistics with n ≥ 10 per cell preventing individual identification), and analytic code (R and Python scripts with documentation enabling reproduction of all reported analyses). Synthetic data generator software will be released immediately via GitHub repository (MIT open-source license) enabling other researchers to create synthetic elderly user populations for validation studies, with 47-page technical documentation, example configuration files, and validation scripts comparing synthetic versus real population statistics. Exceptions to data sharing: facial landmark coordinates, voice feature vectors, and video recordings never shared due to re-identification risk and participant consent limitations (consent forms explicitly stated “facial and voice data will not be shared with other researchers”). Research materials (consent forms in Korean with English translation, assessment instruments, interface screenshots) will be shared upon reasonable request to corresponding author, with 72-hour typical response time during active study periods. This comprehensive ethical framework prioritizing participant protection, transparency, and scientific integrity establishes standards for responsible research with vulnerable elderly populations.