Evaluating the Reliability of Three-dimensional Models Constructed Photogrammetry Software 3DF Zephyr by Measuring Joint Angles of Fingers: A Comparison to a Conventional Goniometer

Abstract

Objectives:

Photogrammetry is a method of producing three-dimensional (3D) data from multiple photographs and videos. The purpose of this study was to evaluate the reliability of 3D hand models constructed using photogrammetry.

Methods:

The angles of the distal (DIP) and proximal (PIP) interphalangeal joints of the author's index finger were measured using a protractor goniometer. Five different blocks were used to stabilize the angles of the fingers. We collected 50 photographs and one video. Digital 3D models of the hand were constructed using 3DF Zephyr photogrammetry software. The angles of the DIP and PIP joints of the model were measured in virtual space. The angles obtained via the two methods were subjected to correlation analysis.

Results:

Eighty values (40 joints of 20 fingers; goniometer/3D model) were obtained. The maximum value obtained was 84° and the minimum was 33°. The average time required to take the original photographs of the model was 3 minutes, and that for the videos was 51 seconds. The average time required to construct a 3D image was 12 minutes. The greatest difference between the methods was 9° and the mean difference was 0.2°. The intraclass correlation coefficient (2,1) was 0.971.

Conclusions:

A high level of agreement was found between these angles. A 3D hand model constructed using photogrammetry is reproducible.

Introduction

Recent technological advances have made three-dimensional (3D) images easier to construct with increased quality and decreased cost. For plastic surgeons, this is gratifying because they deal with all body surfaces. A 3D model will allow surgeons to observe patients from any angle. Especially in hand surgery, virtual models have potential clinical and research applications, as the shape of the hand reflects its function directly. Furthermore, Gremba¹⁾ reported a high reproducibility of finger length using a 3D model of a hand created using photogrammetry. This technology acquires 3D data by combining photographs or videos taken from various angles. However, the reliability of 3D hand models remains uncertain.

This study aimed to assess the accuracy of a 3D hand model obtained via photogrammetry, where 3D models of a hand were constructed. The angles of the index finger obtained via a conventional goniometer and from the 3D data were compared to evaluate reproducibility.

Materials and Methods

This laboratory-based diagnostic study was conducted using the data collected from the author's hand. The main outcome showed the correlation between the angles of the index finger obtained via a conventional goniometer and those obtained via 3D data. The secondary outcomes were the time required in taking the photographs/videos and constructing the 3D model.

Subject and samples

The subject was the author's marked hand, and a patterned piece of paper was used as a background. Five different blocks were used to stabilize the angles of the fingers (Figure 1).

Figure 1.

The author’s hand was marked, and a patterned piece of paper was used as a background (a). Five different blocks were used to stabilize the angles of the fingers (b-c). The angles of the index finger’s distal and proximal interphalangeal joints were measured using the goniometer (d). 50 photographs and one video were taken (e).

Procedure

The angles of the distal (DIP) and proximal (PIP) interphalangeal joints of the index finger were measured using the goniometer. Then, 50 photographs and one video were taken (Figure 1). Digital 3D hand models were constructed from the photographs or the video. Figure 2 shows a photograph and a screenshot of the 3D hand model. The angles of the DIP and PIP joints of the model were measured and compared in a virtual space.

Figure 2.

(A) An original photograph of the three-dimensional (3D) hand model. The fingers are marked for easy recognition. (B–D) Screenshots of the 3D hand model that enlarge and rotate freely.

Data analysis

The data were studied by determining intraclass correlation coefficients (ICCs) and Pearson product-moment correlation coefficients, drawing scatter diagrams and Bland-Altman plots, and calculating the standard errors of the mean (SEMs). The range of the ICC assessing the degree of agreement between measurements is 0-1; higher values indicate greater reliability. Intraclass correlations can be categorized as slight (0.00-0.200), fair (0.201-0.400), moderate (0.401-0.600), substantial (0.601-0.800), and almost perfect (0.801-1.00). P-values under 0.05 were considered statistically significant. The data were analyzed using R4.2.1 software (https://www.r-project.org).

Instruments

We used a Pixel 6 smartphone (Google LLC, Mountain View, USA) with a 50-megapixel rear camera. The goniometer had a resolution of 2°, with an 82-mm long arm and 20-mm short arm. Five different grip blocks were used to stabilize the angles of the fingers (Figure 1).

Software application and personal computer

I used commercial photogrammetry and 3D modeling software (3DF Zephyr; 3DFLOW, Udine, Italy). The modeling was conducted on a personal computer, running Windows 10 with an i7-7500 U processor, Intel HD Graphics 620 GPU (Intel, Santa Clara, CA, USA), and 16 gigabytes of RAM.

Ethics statement

This study was approved by the Internal Review Board of our institution (S22-19 2023/2/1). Informed consent was obtained from the participant after providing a detailed study description. All clinical investigations were conducted under the principles of the Helsinki Declaration.

Results

Eighty values (40 joints of 20 fingers; goniometer/3D model) were obtained. The maximum value obtained was 84°, and the minimum was 33°. The greatest difference between the two methods was 9°, and the mean difference was 0.2°. When constructing a single model, the average time required for photography was 3 minutes, and the average length of the video was 51 seconds. The average time required to construct the 3D model in digital space was 12 minutes, and the average time required to measure the angle of the index finger was 3 minutes and 7 seconds. The Pearson product-moment correlation coefficient was 0.971 (P < 0.05). The ICC (2,1) was 0.97, indicating very high agreement between the methods. The SEM was 2.47. A scatter diagram and a Bland-Altman plot are shown in Figure 3.

Figure 3.

(Left) Scatter diagram of the measurements obtained using the three-dimensional (3D) hand model versus the goniometer. (Right) Bland–Altman plot of measurements obtained using the 3D hand model and goniometer.

I further compared the results for DIP versus PIP values and values from photographs versus those from videos. The ICC (2,1) was 0.94 for the DIP and 0.96 for the PIP, with SEM being 2.90 and 1.97, respectively. The ICC (2,1) was 0.96 for the photographs and 0.97 for the videos, with SEM being 2.68 and 2.30, respectively. Scatter diagrams are shown in Figure 4.

Figure 4.

Scatter diagram of the measurements obtained using the three-dimensional hand model and goniometer (DIP vs. PIP groups and photograph vs. video groups).

Discussion

This study evaluated the reliability of 3D hand models created by photogrammetry by comparing finger angles measured by a goniometer. The level of agreement was very high. Previous reports have used this technology to measure the shape of the face²⁾ or periocular region³⁾, but not the angles of the fingers. This study was the first to evaluate the accuracy of virtual finger joint angle measurements obtained via photogrammetry.

In hand surgery, measuring the fingers' range of motion (ROM) is crucial for making objective assessments^4,5). The protractor-based goniometer is the gold standard equipment because of its high availability and low cost. However, measuring the ROM in multiple finger joints is often complicated; this depends on the examiner's skill, and recording the fingers' appearance is impossible. Other alternatives have been reported, such as using electrical devices⁶⁾, smartphones⁷⁾, or two-dimensional photogrammetry⁸⁾, but these approaches are rare. In this study, a 3D hand model was built, and its parameters were measured in a virtual space. The high reproducibility of the joint angles shows the precision of our model, leading to a technique that automatically measures all joint angles from 3D models.

Additionally, our hand model had a high degree of color reproducibility because We used photogrammetry for data scanning. A smartphone, an old PC, and free software were used. Other methods include contact scanners, computed tomography, and light detection and ranging. Each method has advantages and disadvantages regarding accuracy, price, simplicity, color reproducibility, and invasiveness. The advantages of photogrammetry are the high color resolution, quick configuration, easy accessibility, noninvasive nature, and low cost. High accuracy was achieved since the amount of time required to take images in this experiment was substantial. New technology has made it possible to create 3D data from a single photograph, but this is not sufficiently precise for the 3D modeling of fingers. Increased precision rapidly expands this technique's application.

A 3D model reproducing the morphology and coloration accurately is useful for preoperative simulations and academic research. In the future, I would like to develop a hand-evaluation program that includes objective esthetics. The Disabilities of the Arm, Shoulder, and Hand index used by the Japanese Society for Surgery of the Hand to evaluate hand functions excludes an esthetic questionnaire, so high color reproducibility would complement the index.

There is ongoing research on motion capture of fingers⁹⁾. I gave up using Google MediaPipe to analyze hand motions because of inaccuracy. The framework is intended for video motion capture and is currently used for sign language video analysis¹⁰⁾. If hand movements could be recorded as a 3D model, their joint angles, linkage, and smoothness would be a new assessment target.

Our study had three main limitations: the measurement target was only the index finger, which is easy to measure; the number of measurements was small; and there was only one examiner. I plan to overcome this problem by increasing the scale of tests in the future.

Conclusion

I used photogrammetry to create a 3D hand model and measured the joint angles of the index finger. I found that the reliability of the model was equivalent to that of a conventional goniometer.

Author Contributions: KM contributed to all of this work.

Conflicts of Interest: There are no conflicts of interest.

Ethical Approval: The study protocol was approved by the Ethics Committee of Saiseikai Matsuyama Hospital.

Consent to Participate: Informed consent was obtained from a participant after a detailed study description was provided.

Consent for Publication: Informed consent was obtained from a participant after a detailed study description was provided.