🤖 AI

Quantitative Movement Testing: Measuring Patient Movements from a Single Smartphone Video

This paper presents and validates Quantitative Movement Testing (QMT), a scalable computer vision pipeline that extracts accurate 3D kinematic biomarkers from single smartphone videos, offering an accessible alternative to costly laboratory motion capture for objectively tracking chronic pain progression and treatment response in real-world settings.

Original authors: Pranav Mahajan, Amanda Wall, Eleonora Maria Camerone, Julie Stebbins, Eoin Kelleher, Shuangyi Tong, Annina Schmid, Katja Wiech, Anushka Irani, Ben Seymour

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Pranav Mahajan, Amanda Wall, Eleonora Maria Camerone, Julie Stebbins, Eoin Kelleher, Shuangyi Tong, Annina Schmid, Katja Wiech, Anushka Irani, Ben Seymour

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: A "Fitness Tracker" for Your Joints

Imagine you have a doctor who needs to know exactly how your body moves when you bend down to pick up a sock or squat to tie your shoe. Usually, to get this information, you'd have to visit a high-tech lab, wear a suit covered in reflective stickers, and be filmed by a bank of expensive cameras. It's like trying to film a movie in a Hollywood studio: perfect, but expensive and hard to schedule.

The researchers in this paper wanted to build a "mobile studio." They created a system called Quantitative Movement Testing (QMT). Their goal was to see if a regular smartphone camera could act like that expensive Hollywood camera, turning a simple video of you moving into precise, scientific data about your joints.

How It Works: The "Magic Translator"

Think of the smartphone video as a flat, 2D drawing. The problem is that a flat drawing doesn't tell you how deep your knee bent or how far your spine twisted.

The QMT system uses a piece of computer software (a "deep learning" model) that acts like a magic translator. It looks at the flat video and "lifts" the image into 3D space. It builds a digital skeleton of the person in the video, calculating exactly where their hips, knees, and spine are in three-dimensional space, just by looking at a single video clip.

The Three-Step Test Drive

The team didn't just guess that this would work; they put it through three rigorous tests, like a new car model being tested on a track, in a city, and on a dirt road.

1. The "Track Day" (Lab Validation)
First, they tested the system in a perfect, controlled environment (a lab). They had 13 healthy people wear the expensive "gold standard" motion capture suits and record videos on smartphones at the same time.

The Result: The smartphone system was surprisingly accurate. It matched the expensive lab suits very closely (over 85% agreement).
The Trick: They found that looking at the movement as a flat, 2D angle (like a shadow on a wall) was actually more accurate for the spine than trying to calculate complex 3D twists. It's like trying to measure the angle of a door opening: looking at it from the side is often clearer than trying to guess the angle from a weird diagonal view.

2. The "City Commute" (Fibromyalgia Trial)
Next, they took the system to a real clinical trial involving patients with fibromyalgia (a condition causing widespread pain and fatigue). They wanted to see if the system could detect if a sleep treatment made people move better.

The Result: The system worked reliably. It could measure the same person's movement on different days without getting confused (high reliability).
The Catch: The treatment didn't actually change how the patients moved, so the system didn't find a difference between the treated group and the control group. The paper notes this doesn't mean the system failed; it just means the treatment didn't change the movement, or the system wasn't sensitive enough to see tiny changes. It proved the system could be used in a real trial, even if the specific result was "no change."

3. The "Dirt Road" (At-Home Monitoring)
Finally, they sent the system home with 97 people (some with chronic sciatica, some healthy) for 30 days. These people had to film themselves bending over every day in their own living rooms, with their own lighting, furniture, and angles.

The Result: This was the hardest test. The "noise" (errors) was higher because living rooms aren't perfect labs. However, the system still worked well enough to spot the difference between the healthy group and the pain group when looking at the data as a whole.
The Limitation: While the system could tell the groups apart, it couldn't perfectly predict how a single person was feeling on a specific day just by looking at one video. The daily ups and downs of home life made the data a bit "fuzzy."

The Bottom Line

The paper concludes that this smartphone method is a scalable, accessible alternative to the expensive lab suits.

What it does well: It turns a simple video into scientific data that is accurate enough to compare groups of people and track general trends. It's like having a high-quality map that is good enough to navigate a whole country, even if it's not perfect for finding a single specific house number.
What it struggles with: It's not yet perfect for measuring tiny, day-to-day changes in a messy home environment. The "fuzziness" of home videos means you need to look at the big picture (averages over time) rather than a single snapshot.

In short: The researchers built a tool that lets doctors use a smartphone to get a "3D X-ray" of your movement. It's not quite as perfect as the million-dollar lab equipment, but it's free, easy to use, and good enough to help us understand how chronic pain affects the way we move in the real world.

Technical Summary: Quantitative Movement Testing (QMT)

Problem Statement

Chronic pain and musculoskeletal conditions significantly impair functional ability, yet objectively measuring this impact in real-world settings remains a major challenge. Current clinical practice relies heavily on subjective patient reporting or static visual appraisal, neither of which captures the dynamic quality of movement required to monitor disease progression or treatment response. While optical motion capture (MoCap) serves as the gold standard for high-precision biomechanical assessment, it is restricted to laboratory environments due to high costs, the need for dedicated infrastructure, and the physical discomfort of attaching reflective markers to patients with sensory hypersensitivity. Furthermore, the logistical burden of laboratory visits hinders systematic longitudinal tracking. Although recent computer vision advances have lowered barriers, existing solutions either require multiple synchronized cameras (limiting home use) or rely on end-to-end deep learning models that lack geometric interpretability and generalizability beyond specific training datasets. There is a critical need for a scalable, accessible, and interpretable method to extract 3D kinematic biomarkers from standard monocular smartphone video.

Methodology

The authors developed Quantitative Movement Testing (QMT), a computer vision pipeline designed to extract 3D kinematic biomarkers from single-camera smartphone videos. The framework balances clinical accessibility with biomechanical accuracy by utilizing a "lifting" architecture rather than triangulation.

Pipeline Architecture

Pose Estimation: Standard monocular video is processed using a pre-trained VideoPose3D model to estimate 3D skeletal poses from 17 joints (including pelvis, hips, knees, ankles, spine, and upper torso).
Metric Derivation: From the 3D joint trajectories, the system computes specific planar (2D) angles and spatial distances to maximize geometric stability. Key metrics include:
- Lumbar flexion: Angle of the spine-pelvis vector relative to a vertical axis.
- Knee/Hip flexion: Angles between limb vectors.
- Spatial reach: Vertical distance between wrist and ankle (forward bend) or pelvis and ankle (minisquat).
Temporal Alignment: To handle variable recording speeds and durations, trajectories are interpolated onto a normalized time axis. Dynamic Time Warping (DTW) is employed against a sinusoidal template to identify and align repeating movement motifs (peaks and troughs), allowing for the extraction of functional ranges (peak minus trough) across repetitions.
Calibration: To correct systematic biases inherent in monocular depth estimation, a leave-one-subject-out cross-validation (LOOCV) calibration is applied. An additive offset is estimated from the remaining subjects and applied to the held-out subject's measurements to eliminate bias.

Validation and Clinical Deployment

The QMT pipeline was evaluated across three distinct datasets:

Laboratory Validation: 13 healthy controls performed forward bending, marching on the spot, and minisquats in a controlled gait lab equipped with gold-standard optical MoCap.
PainLESS Trial: A prospective randomized controlled trial (N=80 enrolled, N=57 with QMT data) involving fibromyalgia patients undergoing a digital cognitive behavioral therapy for insomnia (dCBT-I) intervention. QMT was used to assess pre- and post-intervention changes.
BeADS Study: A 30-day longitudinal at-home monitoring study (N=97 enrolled, N=83 analyzed) involving chronic sciatica patients and healthy controls performing daily forward-bending tasks.

Key Results

Laboratory Validation

High Agreement: After LOOCV calibration, QMT demonstrated strong agreement with optical MoCap, with correlations exceeding r > 0.85 for most metrics.
Planar vs. Euler Angles: QMT outputs for planar (2D) surrogate angles (e.g., planar lumbar flexion) showed higher correlation with MoCap than traditional 3D Euler derivations (e.g., planar lumbar flexion r = 0.97 vs. Euler r = 0.92 during minisquats).
Error Margins: Mean Absolute Errors (MAE) for planar angles ranged from 2.7° to 7.5°, performing comparably to standard clinical goniometry (~5° error) and significantly better than clinical visual estimation (9°–12° error).
Robustness: Diagnostic accuracy remained stable even when reducing the recording burden from 10 trials to a single trial.

Clinical Utility (PainLESS Trial)

Reliability: QMT demonstrated high test-retest reliability in fibromyalgia patients (r = 0.78–0.99).
Treatment Sensitivity: The study found no statistically significant differences in kinematic change scores between the dCBT-I and control groups. Bayes Factors provided moderate to anecdotal evidence for the null hypothesis. The authors note this may indicate either a true lack of treatment effect on movement or insufficient sensitivity of QMT to detect subtle kinematic changes without a positive control.

Longitudinal Monitoring (BeADS Study)

Home Feasibility: The pipeline successfully captured movement fluctuations in unconstrained home environments. While absolute measurement noise was 2–3 times higher than in the lab, the system tracked day-to-day changes with a mean $\Delta$ MAE of 3.10 cm.
Group Differentiation: Despite increased noise, aggregating longitudinal data successfully differentiated healthy controls from sciatica patients. Healthy controls exhibited significantly greater reach distances and higher lumbar flexion angles (p < 0.001).
Classification Limits: Single-day classification accuracy between groups was moderate (68.5%), highlighting the high intra-individual variance inherent in home monitoring.

Significance and Claims

The paper positions QMT as a step toward making movement assessment equitable and accessible for diverse chronic pain populations. Its primary significance lies in:

Scalability: Offering a low-cost alternative to optical motion capture that removes the logistical friction of laboratory visits, enabling large-scale clinical trials and longitudinal monitoring.
Validation: Establishing a validated, open-source protocol that benchmarks monocular 3D pose estimation against gold-standard MoCap, demonstrating that planar (2D) surrogates can effectively mitigate monocular depth ambiguities for axial kinematics.
Operational Feasibility: Proving that remote, smartphone-based kinematic capture is viable in clinical workflows, as evidenced by successful deployment in both a randomized controlled trial and a 30-day home study.

The authors maintain a modest stance, acknowledging that while QMT provides objective biomarkers for tracking disease progression, further research is required to optimize reliability in home-based environments (e.g., via domain-specific fine-tuning or intelligent user interfaces for framing feedback) and to validate responsiveness to specific interventions. The work does not claim to replace clinical judgment but rather to provide an accessible tool for objective functional profiling.