Alzheimer disease (AD) is a leading cause of dementia [1]. Symptoms of dementia, including agitation, psychosis, depression, and apathy, impact patients’ quality of life, disrupt relationships, increase caregiver burden, and elevate patients’ risks of institutionalization and mortality [2]. Nonpharmacological interventions such as music therapy (MT) have potential to reduce agitation, support emotional stabilization, and leverage auditory-cognitive pathways to stimulate autobiographical memory and modulate emotional responses [3-5]. The therapeutic effect of MT is enhanced when culturally familiar music is used [6]. However, traditional MT interventions face significant limitations and lack dynamic personalization and real-time adaptation to patients’ emotional and cognitive states, limiting their overall effectiveness.

To solve these critical problems, we developed MusicAlzheimer, an artificial intelligence–powered digital MT platform. It integrates culturally tailored music-based interventions with adaptation to real-time emotional states of patients and personalization measures.

Caregiver testing demonstrated that MusicAlzheimer’s modules functioned cohesively to support personalized dementia care. Findings are organized into 4 domains.

GPT-generated prompts in MusicAlzheimer were intentionally concise and era-specific to support reminiscence and help caregivers sustain engagement during sessions, helping reduce caregivers’ burden and support more consistent therapeutic engagement.

The platform leverages preserved auditory and affective pathways in AD by pairing culturally familiar music with personally relevant cues, which may facilitate emotional regulation and memory recall [7-9]. Such multimodal stimulation can reduce the anxiety or confusion that often interfere with recall, enhancing the therapeutic impact of familiar music interventions. The robust emotional-tracking component captures basic affective states and memory-response levels, offering caregivers simple trend insights during home-based sessions.

Regarding potential longer-term cognitive effects, two theoretical frameworks provide relevant support. First, MT can influence neural plasticity. By stimulating the remaining auditory and emotional neural networks continuously, MT can strengthen the integrity of neural pathways and help regulate the emotions of patients with AD when the disease progresses [10]. Second, music can trigger both direct (music directly evokes specific memories) and indirect retrieval (music can create a mood to facilitate the retrieval of a broader range of memories). MusicAlzheimer combines era-specific music with GPT-generated language to engage both pathways simultaneously [11,12]. We recommend using validated outcome measures, such as the Mini-Mental State Examination and the Alzheimer’s Disease Assessment Scale-Cognitive Subscale, to form longitudinal evaluations before making any claims regarding efficacy.

Study limitations include a small sample size, limited culturally diverse music, reliance on subjective caregiver ratings rather than validated clinical instruments, simplified audio synthesis, dependence on internet connectivity, and the absence of clinical outcome validation. As a prototype-stage feasibility study, the system is not intended to replace professional therapy but to complement dementia care for individuals with preserved musical memory. Future work will expand multimodal sensing, diversify musical content, evaluate clinical impact in controlled studies, and include on-device large language model deployment for offline capability.