# From images to health insight: integrating MLLM, NLP, and objective Q-sorting of nursing-home built environment orientations

## TL;DR

This study uses AI and image analysis to classify nursing home environments into four types that support health and aging, offering a new framework for design and policy.

## Contribution

A novel 'semantic-cognitive' framework combining MLLM, NLP, and Q-sorting to classify nursing home environments and their health implications.

## Key findings

- Four environmental orientation types were identified using Q methodology, explaining 86% of variance.
- Environmental types include interior-centric, layout-oriented, landscape-centered, and rehabilitation-driven.
- The framework links environmental orientations to health-support mechanisms for evidence-based policy.

## Abstract

Population aging has intensified pressure on global healthcare and social security systems, driving a shift in care from treatment-oriented approaches toward functional maintenance and chronic disease rehabilitation. How to design and optimize the built environment of nursing homes to support the physical and mental health of older adults has become an important issue in health policy and architectural design. Existing research lacks comparable types of environmental orientations for nursing homes and an operational guidance framework for environmental design, leaving subjective decision-making unable to align with functional maintenance goals.

This study constructed a “semantic-cognitive” hybrid framework. It treated nursing homes' self-selected built-environment images on eldercare portals as espoused environmental orientation signals, revealing their belief structures and value orientations in convalescent practice. We compiled 3,578 environmental images from 389 nursing homes; used multimodal large language models (MLLMs) to generate structured environmental audit texts; applied natural language processing (NLP) for vectorization, dimensionality reduction, and clustering to refine and standardize the Q statement set; constructed Q-sorting similarity matrices from semantic similarity; and performed factor analysis with rotation to obtain typified belief structures.

Q methodology identified a four-factor solution explaining 86% of the total variance. Four environmental orientation types were identified in chronic disease management settings—Interior-centric type (safe accessibility, low stimulation, uniform lighting); Layout-oriented type (continuous corridors, clear entrances, orderly walking); Landscape-centered type (shaded gardens, good greenery, encouraging outdoor stay and social interaction); and Rehabilitation-driven type (rehabilitation equipment in place, open space, normalized training).

This study provides a comparable and testable research pathway, reveals the linkage pathways between different environmental orientations and health-support mechanisms, and offers clear targets for subsequent longitudinal and mixed-methods evaluations, design, and evidence-based healthy aging policy management, with important theoretical and managerial significance.