Skip to content

Foundational three-dimensional gait model predicts disorders of 18 bodily systems

Share
Foundational three-dimensional gait model predicts disorders of 18 bodily systems

Listen to this article

Read by Anchor

Traditional medical practice treats human gait disorders as a localized manifestation of isolated motor or neurological causes, but a recent study published in Nature Medicine redefined body movement as a comprehensive biomarker that reveals the biological state of multiple organ systems. Researchers developed a foundational model based on machine learning to analyse three-dimensional skeletal motion, using visual recordings captured by a single depth camera while participants performed five motor tasks. The study sample comprised 3,328 adult volunteers who underwent deep phenotyping within the Human Phenotype Project.

Previous attempts to assess movement relied on pre-defined manual geometric metrics such as step speed, length and frequency, which are numeric measures that fail to capture the fine-grained muscular and neural coordination across the whole body. In contrast,The foundational model re-represents motion as learned numeric vectors that describe how skeletal joints move together over time, surpassing separate computational readings in favor of an integrated dynamic fingerprint that reflects functional interaction between limbs and trunk.

Test results showed that the digital vectors derived from gait outperformed traditional geometric metrics in predicting complex physiological variables; they achieved a strong statistical correlation in estimating chronological age and body mass, and reached a Pearson correlation of 0.83 for measuring visceral fat tissue area. The model successfully predicted about 1,887 physiological and disease traits out of 3,210 tested, including precise measurements of hand-grip strength and sleep quality, confirming that skeletal dynamics carry vital signals beyond the visible bones and joints.

When the performance of the digital vectors was compared with a baseline model that relied solely on age, body-mass index, visceral fat and height, the gait vectors added independent predictive power covering all eighteen vital systems in men and seventeen in women.These vectors went beyond morphological indicators to predict actual clinical diagnoses and patterns of therapeutic drug use, demonstrating that an individual’s movement pattern constitutes a self-standing biological detector rather than merely reflecting weight and age factors.

Anatomical exclusion analyses performed by the research team revealed a specific functional distribution of health information within motion components; lower-limb dynamics and leg movement dominated predictions of metabolic indices and frailty disorders, while trunk motions carried cues linked to sleep patterns and daily lifestyle. This close association explains the AI’s ability to read complex interactions such as motor-balance deterioration associated with sleep deprivation or muscle-balance disruption accompanying visceral-fat accumulation and liver fibrosis, based on code and data the researchers made available for non-commercial academic purposes through open platforms including GitHub and Zenodo.

This improvement in motion modelling accuracy using a single camera has a direct practical impact on healthcare plans and preventive screening projects in medical facilities across the Gulf and Egypt; it enables operators to replace costly, complex motion sensors with low-cost passive visual monitoring solutions that can be integrated into hospital corridors and primary-care clinics. The approach allows clinical teams to flag frailty and metabolic-disorder indicators before advanced symptoms appear, reducing routine-screening costs, and requires health-IT managers in the region to prepare secure pipelines for three-dimensional video data without compromising patient-privacy policies.

Don't miss the next story

Subscribe for updates