The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan

The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan Samir Ghandour, Anton Lebedev, Wei Shao Tung, Konstantin Semianov, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3868289/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Hallux valgus (HV) is a common foot deformity. Traditional detection methods include clinical examination and radiographic imaging, which, although reliable, often remain inaccessible to many due to existing care barriers. This study introduces an innovative approach to computer vision analysis and phone camera-based 3D scanning technology to detect and assess HV severity. We evaluated the accuracy of this method against routine clinical examination as the currently accepted assessment standard. Our study included 120 participants, resulting in 240 foot scans, with a diverse demographic representation. The computer vision algorithm utilized a surrogate angle, automatically derived from the 3D scans, to identify the severity of HV, and its correlation with traditional radiographic measurements for HV. Our findings reveal that computer vision-based detections offer high accuracy, with an Area Under the Curve (AUC) score of 0.947, presenting a promising alternative to conventional methods. This technology offers promise for increasing access to HV detection, potentially aiding in earlier diagnosis as well as non-operative treatment options that may ultimately reduce the need for surgical intervention. Its ease of use and application in telemedicine contexts has the potential, moreover, to significantly benefit patients in remote or underserved areas and expand capacity to promote similar care improvement in other areas of musculoskeletal disease. Biological sciences/Biotechnology Health sciences/Anatomy Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Health sciences/Signs and symptoms Physical sciences/Engineering Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 INTRODUCTION Hallux valgus (HV) is a common anatomic foot deformity that describes an abnormally angulated, rotated, and laterally deviated great toe at the first metatarsophalangeal joint [ 1 ]. The prevalence of HV varies, with different epidemiological studies suggesting estimates ranging from 21–70% globally, depending on the target population [ 2 – 7 ]. HV seems to occur more commonly among women and older individuals [ 3 , 8 ]. Considered a degenerative joint disorder in adults, untreated HV can lead to progressive pain, stiffness, difficulty with footwear, functional limitations, edema, and arthritis in the long term [ 9 – 11 ]. The most reliable and commonly employed method of detecting and diagnosing HV is a thorough clinical and radiographic examination, which includes evaluation of the patient’s foot with observation of great toe position, range of motion, correctability, and determination of radiographic alignment by a healthcare professional. Radiographic imaging is commonly used to confirm the diagnosis and assess severity [ 12 ]. While conservative treatment options exist for those who are primarily in the earlier stages of HV, including orthotic devices, physical therapy, and lifestyle modifications, unaddressed deformity often progresses with associated symptoms and may eventually necessitate a surgical solution that is more costly and exposes the patient to potential complications such as nonunion, malunion, recurrent deformity, infection, nerve injury, and other peri-surgical risks [ 13 , 14 ]. Over the last few decades, three-dimensional (3D) scanning technologies have been rapidly developing in both capacity and accessibility. Structured light and light detection and range (LIDAR) are 3D scanning technologies that actively light the scene to measure distances between target objects and the camera. Utilizing surface points that can be distinguished on the object, ‘depth’ is calculated and used to create highly accurate 3D representations of the target object or the surrounding environment. While the LIDAR technology was mainly used for augmented reality or scanning large objects, such as rooms, structured light technology performed better on smaller objects, such as human anatomical structures [ 15 ]. More recently, both LIDAR and structured light technologies have been successfully incorporated into newer smartphone models for various applications, including foot and ankle deformities [ 16 , 17 ]. Given the 3D nature of HV, this technology combined with computer vision-based analysis may allow patients to detect the deformity in their foot at earlier stages more effectively and subsequently seek treatment options to prevent further deformity progression and its associated morbidity. This study aims to assess the use of computer vision in detecting HV deformity by 3D foot scans obtained using structured light technology in smartphone cameras and determine its diagnostic accuracy with deformity severity when compared to clinical and radiographic examination as the accepted assessment standard. METHODS Study Population The protocol for this prospective study was approved by the Institutional Review Board (IRB no. 2022P001722). The study was performed in accordance with relevant guidelines and regulations granted by the IRB for this study. Informed consent was obtained from participants prior to enrollment. Patients were recruited from two orthopaedic foot and ankle clinics in our institution, which is a tertiary hospital in Boston, Massachusetts. Patients with foot pain for various reasons, including plantar fasciitis, cavovarus foot, pes planus, and pes cavus, were assessed for HV deformity. Feet that were deemed as having HV by a foot and ankle surgeon were labeled as cases, while feet that were not diagnosed with HV were labeled as controls. The inclusion criteria for the subjects recruited in this study were 1) adults aged ≥ 18 years old, 2) experiencing foot pain, 3) being able to ambulate and bear weight on their feet, and 4) having the capacity to provide informed consent for participation. The exclusion criteria were 1) the presence of fractures in the foot, including stress fractures and traumatic fractures, 2) the presence of neuropathy, neuronal injuries, or structural deformities related to chronic diseases (i.e., diabetic neuropathy, Charcot-Marie-Tooth disease, Charcot's foot), 3) unable to stand on one foot, 4) structural deformities in the toes (mainly the first ray). For a random subset of participants who had anteroposterior radiographs of their feet, an expert radiologist measured the hallux valgus angle (HVA) and first intermetatarsal angle (IMA) for the correlation with the surrogate angle derived from the scan. The severity of HV, when present, was labeled according to these radiographic measurements. 3D Scanning and Developing the AI Algorithm An algorithm that estimates the surrogate HVA from the generated point cloud was utilized. To facilitate the use of the algorithm by clinical experts, this algorithm is deployed in an application that could run on smartphones (Neatsy App for iOS, Neatsy, Inc., Menlo Park, CA). The application enables the provider to obtain a 3D scan of the feet using a structured light camera system [ 28 ]. This system generates a 3D point cloud, or an array of points, of the medial view of the patient's foot while the patient maintains a single-footed posture with the camera fixed in a static or near-static position. The application assembles a sequence of Red-Green-Blue-Depth (RGBD) images, subsequently unifying them into a coherent point cloud, achieved through the application of an Iterative Closest Point (ICP) algorithm [ 29 ]. Detailed 3D models created can be found in Fig. 1 . As this novel diagnostic algorithm does not depend on radiographic imaging, the HVA and IMA cannot be measured from the point cloud. Instead, we chose superficial landmarks on the participant’s foot corresponding to the metatarsophalangeal (hallux point) and proximal interphalangeal (toe point) joint of the patient's big toe on the XY plane to develop a surrogate angle. This surrogate angle, which can aid our model in identifying HV and, subsequently, its severity, is measured between two lines: 1) a line passing through the hallux point parallel to the x-axis, and 2) a line passing through the hallux and toe points [ Figure 2 ] . To justify the use of this surrogate angle for our algorithm, this measurement was compared to HVA and IMA taken from participant radiographs of the foot, if already available. Pearson’s correlation coefficient was used to determine the strength of the relationship between the surrogate angle and the validated HVA and IMA. The raw scanner output designates the z-axis as the depth axis (medial to lateral), aligns the x-axis approximately collinear with the vector from the heel to the toes, and positions the y-axis as the vertical axis. Mirroring is exclusively applied to the left-sided scans along the x-axis, which ensures that the x-coordinate of the heel point is less than that of the toe point. The right-sided scans inherently exhibit this characteristic without requiring any additional adjustments. The next step in the algorithm involves rotating the scan for alignment with both the XZ and XY planes. To provide a more intuitive understanding, picture a foot placed flat on the floor, with the heel against a wall. The XZ plane represents the floor, and the XY plane represents the wall. Orientation was achieved by the following: Plane Selection : The XZ or XY plane was chosen based on context and it should pass through the point in the scan with the minimum z or x coordinate, respectively. Angle Exploration : The optimal alignment was found by systematically testing 100 angles within the range of -π/4 to π/4 with stepwise increments of π/200. Angle Selection : The angle that results in the lowest error was selected, effectively aligning the scan with the chosen plane. Rotation Angles : The XZ plane was aligned in relation to the rotation angles around the z-axis. In contrast, the XY plane was aligned through sequentially considering the rotation angles around the X and Y axes. Stability Enhancement : Stability was ensured by repeating the outlined process twice and replacing the minimum point with the value corresponding to the 0.001 quantile. Next, points that are not the hallux or toe points are filtered. To filter these points, the length of the scan, or the distance between the furthest limits on x-axis, is calculated. The points that lie within the first 65% of the total length between the heel and the base of the big toe (the area approximately defining the heel to the distal limit of the foot arch), as well as the points that lie beyond the 95% cutoff of the total length, namely, the phalanges, are removed. Another filtration is performed for the vertical dimension. Here, the ball height is approximated as the distance between the furthest points on the y-axis within the area defined after the index filtration. Then, all the points that are below the second decile of the total ball height are removed to eliminate the outliers near the floor. The resultant scan area is presented in Fig. 3 a. The remaining points are divided into 35 clusters. Each cluster is represented by vertical strips of equal width that break each scan into 35 disjoint sets of points [ Figure 3 b ] . Within each cluster, a 3D point that is equivalent to the first, or shortest, decile in depth on the medial-to-lateral axis (z-axis) is identified and used in the following part of the algorithm. This method of selecting the minimum is preferred over the raw dataset as it removes any possible outliers and improves the overall stability of chosen points in terms of their z-coordinate. Next, the local minima across the clusters are found [ Figure 4 ] . A local minimum is defined in our algorithm as a point whose z-coordinate is less than or equal to the z-coordinates of its two immediately neighboring points to the right and to the left. In most cases, two minima were identified, with the first minimum encountered from posterior to anterior (heel to toe) being of lesser magnitude than the second [ Figure 4 ] . For these instances, the first minimum encountered along the axis is designated as the hallux point, and the second is the toe point. In all remaining cases, the hallux point is ascribed to the sole local minimum characterized by the smallest z-coordinate, with the toe point being unequivocally identified as the last element of the array. By convention, the first point of the array on the z-axis cannot be the hallux point, as there are no neighboring points on its left side. The final step of the algorithm is to calculate the surrogate angle. As previously mentioned, this surrogate angle is not equivalent to either HVA or IMA. This surrogate angle, which can aid our model in identifying HV and its severity, is measured between two lines: 1) a line passing through the hallux point parallel to the x-axis, and 2) a line passing through the hallux and toe points [Figure 3 ]. The entire workflow of the algorithm can be found in Fig. 5 . Statistical Analysis The area under the ROC Curve (AUC) and the area under the Precision-Recall Curve (PR-AUC) was calculated by using output angles without any post-processing and was utilized to assess model performance. For AUC, the DeLong method was applied to acquire the 95% confidence interval to ensure significance [ 18 ]. To obtain binary predictions of a participant's condition, a classification threshold score is required. The F1 score, a machine learning metric used to evaluate model accuracy, for each possible classification threshold was calculated, and the threshold that maximized the F1 score was selected. Whilst algorithm training was not necessary for this methodology, the classification threshold selection process can be considered as a mode of training. Hence, the leave-one-out cross-validation (LOOCV) method was used for metric calculation. The LOOCV method was repeated for all N , where N is the study population. For each participant i , the method calculated the classification threshold using N-1 scans (i.e., without the participant i ’s scan) and compared the angle of the excluded scan with the computed value. After cross-validation, a list of binarized predictions is produced and used for calculating the F1 score, Youden’s index, sensitivity (recall), specificity, accuracy, precision (positive predictive value), and negative predictive value. Pearson’s correlation test was used to determine the association between the surrogate angle and HVA or IMA measured in a subset of participants who had existing weightbearing anterior-posterior radiographs. A cluster analysis was performed to further dissect the correlation between the surrogate angle and HVA [ Figure 7 ] . The initial clusters were formed based on the typical ranges of HVA for each grade of HV in adults: normal (< 15°), mild (15–25°), moderate (25–35°), and severe (≥ 35°). To find the surrogate angle’s classification threshold binary classification problems were created based on the minimum HVA to be categorized in each grade of HV (i.e., < 15° vs. 15–100°, or < 25° vs. 25–100°). The One-vs-All multiclass classification was used to assess the predictive model performance in categorizing HV severity. RESULTS In total, 120 cases and controls(240 feet) were included in the study. Twenty-nine participants had bilateral HV, 12 had unilateral HV, and 79 had normal feet. In total, seventy feet (29.1%) had HV, while 170 (70.8%) were labeled normal following clinical assessment by a foot and ankle surgeon The mean ± standard deviation age of our participants was 37.98 ± 18.23 years with 48 (40%) males and 72 (60%) females. The racial composition of the scanned participants varied, with 61 identifying as Caucasian, 31 as Asian/Pacific Islander, six as African American, six as Middle Eastern, and 16 others as mixed race. Table 1 presents the classification metrics that were calculated on binarized output after LOOCV. Table 1 Classification metrics calculated on the binarized output after LOOCV. Metric Name F1 Score Accuracy Precision (PPV) Recall (Sensitivity) Specificity Youden’s Index Negative Predictive Value Metric Value 0.741 0.845 0.726 0.757 0.882 0.639 0.892 Amongst the 120 patients recruited for the study, 25 random patients had 29 foot radiographs that were obtained as part of their diagnostic workup prescribed by an expert clinician. We measured IMA and HVA on these patients, and Pearson's correlation test showed that the correlation of the surrogate angle with HVA was 0.91 (95% CI: [0.81, 0.96]; p < 0.001), and 0.65 (95% CI: [0.37, 0.82]; p-value < 0.001) for IMA. The optimal classification thresholds found for the surrogate angle using a brute search algorithm maximizing the F1 score for the created clusters were different from the HVA-based clusters. Additional metrics using the union of the two higher-grade clusters were calculated to attempt to improve model performance. Table 2 shows the resultant classification thresholds based on the surrogate angle for the three final clusters, including the combined moderate-severe cluster. AUC and PR-AUC curves for positive and negative classes were calculated on row angles. The AUC score was 0.947 (95% CI: [0.916, 0.971]), calculated via the DeLong method [ 18 ]. PR curves are represented in Fig. 6 . The PR-AUC for the positive class was 0.89 and 0.92 for the negative class. Table 2 Performance metrics of the cluster analysis between the surrogate angle and HVA. Target Degree Interval F1 Score Accuracy Precision (PPV) Recall (Sensitivity) Specificity Youden’s Index Negative Predictive Value 0–15° 0.952 0.966 0.909 1.000 0.947 0.947 1.000 15–25° 0.857 0.931 1.000 0.750 1.000 0.750 0.913 25–35° 0.600 0.862 0.600 0.600 0.917 0.517 0.917 25–100° 0.957 0.966 0.917 1.000 0.944 0.944 1.000 35–100° 0.769 0.897 0.714 0.833 0.913 0.746 0.955 DISCUSSION Using structured light technology, the proposed 3D scanning algorithm showed reliable accuracy for detecting HV and determining the severity of the deformity. With a specificity of 0.882, this algorithm proves to be a reliable and accurate diagnostic tool of HV deformity independent of clinical or radiographic examination. For HV severity assessment, the correlation of the surrogate angle with HVA (r = 0.8915; p < 0.0001) is stronger than that of IMA (r = 0.6598; p < 0.0001). The surrogate angle relies on the extent of the malalignment of the hallux, which corresponds to increased HVA. This strong correlation with HVA is consistent with large-scale data assessment of pre-existing foot scans by Jiao et. al when attempting to calculate HVA automatically from 3D foot scans without referring to radiographic assessment [ 19 ]. Our novel approach with a surrogate angle that is correlated with the radiographic assessment of a random subset of our scanned feet goes one step further in establishing this correlation with the HVA. The IMA does not manifest prominent exterior landmarks that are easily identifiable by our targeted scanning methodology, which is an important prerequisite when using structured light—as co-existent metatarsus adductus might interfere with HV assessment given a decreased IMA despite a notable HV deformity. This may be the reason behind the IMA’s relatively lower correlation to the surrogate angle in relation to severity, albeit still being significant. Developing a phone-based diagnostic tool for identifying HV can vastly improve accessibility and early awareness facilitating prompt treatment for a great proportion of patients with this deformity [ 20 ]. In addition to logistical challenges such as patient mobility and transportation precluding patients from attending clinics in person, a common theme identified during this prospective study deterring patients from seeking expert consultation is embarrassment and shame surrounding the aesthetics of their feet [ 21 ]. Obviating the need for in person visits to assess and diagnose HV deformity at the clinic may alleviate the logistical, financial, and occupational costs these patients may incur. Our 3D scanning algorithm has the potential to provide a convenient solution for these patients in the privacy of their own homes, with even greater potential for expansion to remote monitoring of other orthopaedic pathologies. The use of such accessible technology at home might also help improve health literacy and early identification of HV for those who are not well-versed in foot and ankle pathologies [ 22 ]. Likewise, the use of commonly available devices such as smartphones for this technology may enable patients from low-resource or underserved regions to have access to healthcare by obtaining a diagnosis and treatment recommendations directly and promptly. This expands the options for telemedicine and personalized recommendations for this patient-population. Recently, structured light technology has been employed in many applications within the medical field, showing great potential in various areas of healthcare, wearable devices, and human-machine interfaces. Its ability to create accurate three-dimensional models of human anatomic structures with minimal hardware requirements makes it easily accessible to healthcare providers and researchers alike. For instance, such technology can create digital models that enable their developers to create anatomically conformal prostheses [ 23 ]. Others have explored the non-inferiority of Structured Light in 3D volumetric imaging when compared to other 3D scanning technologies [ 24 ]. Additionally, Structured Light has been shown to be effective in creating conformal microfluidic devices that can interface with the surface of whole organs, allowing researchers to better understand biological tissue and their biomarkers [ 25 ]. Singh et al. utilized smartphone Structured Light facial recognition technology to develop personalized neonatal CPAP masks [ 26 ]. These examples show the reliability of structured light applications in addition to the technology’s potential impact and accessibility, all while utilizing low-cost and commercially available devices such as personal smartphones. While initially used by footwear manufacturers as a tool for their customers to configure shoe size and morphology for fitting, 3D scanning of the foot is becoming a valuable tool for understanding foot deformities and pathologies [ 27 ]. More specifically, Yamashita et al. have utilized a similar 3D scanning technique to explore skeletal structural features and variations in those with HV in different age groups, showing that certain skeletal features of the midfoot influence HV and its progression with age [ 16 ]. In another study, Yamashita et al. used 3D scans of the foot in children to reliably predict the risk of HV development according to analytical parameters abstracted from these scans [ 20 ]. Such an application will not only enable earlier detection but also prompt intervention at earlier stages. Creating and validating a surrogate angle to diagnose HV without relying on radiographic measurements obviates the requirement for radiological examination for diagnosis, reducing the exposure to radiation, time, and overall treatment cost. However, given that this software application was developed with patient independence in mind, it is critical to appreciate the demographic HV affects most. Limitations of this technology may affect specific patient-population groups. Whilst HV does not exclusively affect older patients, a significant proportion exhibits some level of deformity, symptoms, or both. As such, because of the brisk development of technology in the past few decades, many of the patients that fall within the older age category may not own a smartphone, be accustomed to operating advanced software, or have wireless connections with the bandwidth to support large volume data transfers. Another limitation exists in the expectation that these patients would be able to obtain these scans themselves in a safe and viable manner. Certain maneuvers required within the process, such as weightbearing on one foot or propping the smartphone against a wall, may not be achievable for some patients with disability. Lastly, while the algorithm can reliably predict the presence of HV, distinguishing between moderate and severe cases remains a limitation. It is important to note that the current smartphone 3D scanning platform should probably not be used in isolation as a sole diagnostic and treatment tool as final diagnosis and care determination should always be made by a healthcare professional. However, continued development and advancements of this 3D scanning tool will eventually increase its reliability and may even someday obviate the need for early clinical evaluation and intervention by providing personalized digital conservative recommendations. Future directions for research and development include creating alternative ways for patients to obtain these scans independently, as well as expanding this technology to detect other foot deformities or maladies such as pes planus and pes cavus, which saves both patients and clinicians time and operative bandwidth. Declarations DATA AVAILABILITY STATEMENT Some of the datasets generated and/or analyzed during the current study are available in the study’s repository here. This repository contains the minimum information required to substantiate the study’s findings and analysis without compromising participant confidentiality. Further detailed data collected during the current study are available from the corresponding author upon reasonable request. The data that support the methodology of this study are available from Neatsy Inc. (Menlo Park, CA), but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available. Such data are, however, available from the authors upon reasonable request and with permission of Neatsy Inc. (Menlo Park, CA). ACKNOWLEDGEMENTS We want to acknowledge Nour Nassour, MD and Kendal Toy, MS for their contributions in technical suggestions and proofreading the manuscript. AUTHOR CONTRIBUTIONS SG has made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; and draft and substantive revision of the work provided. AL has made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; creation of new software used in the work; and draft of the work provided. WST has made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; and draft and substantive revision of the work provided. KS has made substantial contributions to the conception and design in addition to creation of new software used in the work. AS has made substantial contributions to the conception and design in addition to creation of new software used in the work. CWD has made substantial contributions to the conception and design in addition to draft and substantive revision of the work provided. LBP has made substantial contributions to the conception and design; data acquisition; and draft and substantive revision of the work provided. SAE has made substantial contributions to the conception and design; data acquisition; creation of new software used in the work; and draft and substantive revision of the work provided. ADDITIONAL INFORMATION Competing Interests Statement All the authors in this study declare no potential conflict of interest. References Roddy, E., Zhang, W. & Doherty, M. Prevalence and associations of hallux valgus in a primary care population. Arthritis Care Res. 59, 857–862 (2008). Benvenuti, F., Ferrucci, L., Guralnik, J. M., Gangemi, S. & Baroni, A. Foot Pain and Disability in Older Persons: An Epidemiologic Survey. J. Am. Geriatr. Soc. 43, 479–484 (1995). Dunn, J. E. Prevalence of Foot and Ankle Conditions in a Multiethnic Community Sample of Older Adults. Am. J. Epidemiol. 159, 491–498 (2004). Elton, P. J. & Sanderson, S. P. A chiropodial survey of elderly persons over 65 years in the community. Public Health 100, 219–222 (1986). White, E. G. & Mulley, G. P. Footcare for Very Elderly People: A Community Survey. Age Ageing 18, 275–278 (1989). Leveille, S. G. et al. Foot Pain and Disability in Older Women. Am. J. Epidemiol. 148, 657–665 (1998). Nguyen, U.-S. D. T. et al. Factors associated with hallux valgus in a population-based study of older women and men: the MOBILIZE Boston Study. Osteoarthritis Cartilage 18, 41–46 (2010). Coughlin, M. J. & Jones, C. P. Hallux Valgus: Demographics, Etiology, and Radiographic Assessment. Foot Ankle Int. 28, 759–777 (2007). Cho, N. H., Kim, S., Kwon, D.-J. & Kim, H. A. The prevalence of hallux valgus and its association with foot pain and function in a rural Korean community. J. Bone Joint Surg. Br. 91-B, 494–498 (2009). Barnett, C. H. The normal orientation of the human hallux and the effect of footwear. J. Anat. 96, 489-494.1 (1962). Menz, H. B. & Lord, S. R. The Contribution of Foot Problems to Mobility Impairment and Falls in Community-Dwelling Older People. J. Am. Geriatr. Soc. 49, 1651–1656 (2001). Coughlin, M. J., Saltzman, C. L. & Nunley, J. A. Angular measurements in the evaluation of hallux valgus deformities: a report of the ad hoc committee of the American Orthopaedic Foot & Ankle Society on angular measurements. Foot Ankle Int. 23, 68–74 (2002). Coughlin, M. J. & Jones, C. P. Hallux valgus and first ray mobility. A prospective study. J. Bone Joint Surg. Am. 89, 1887–1898 (2007). Greisberg, J. K. Hallux valgus and first ray mobility. A prospective study. J. Bone Joint Surg. Am. 90, 1166; author reply 1166–1167 (2008). Vogt, M., Rips, A. & Emmelmann, C. Comparison of iPad Pro®’s LiDAR and TrueDepth Capabilities with an Industrial 3D Scanning Solution. Technologies 9, 25 (2021). Yamashita, T., Yamashita, K., Sato, M., Kawasumi, M. & Ata, S. Foot-surface-structure analysis using a smartphone-based 3D foot scanner. Med. Eng. Phys. 95, 90–96 (2021). Kobayashi, T. et al. A Simple 3D Scanning System of the Human Foot Using a Smartphone with a Depth Camera. Proc. 3DBODYTECH 2018 –9th Int. Conf. Exhib. 3D Body Scanning Process. Technol. Lugano Switz. 16–17 Oct 2018 161–169 (2018) doi: 10.15221/18.161 . DeLong, E. R., DeLong, D. M. & Clarke-Pearson, D. L. Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach. Biometrics 44, 837–845 (1988). Jiao, Y., Džeroski, S. & Jurca, A. Analysis of hallux valgus angles automatically extracted from 3D foot scans taken in North America, Europe, and Asia. Ergonomics 66, 1164–1175 (2023). Yamashita, K., Yamashita, T., Sato, M., Kawasumi, M. & Takase, Y. Development of a quantitative measurement system for three-dimensional analysis of foot morphology using a smartphone. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. Annu. Int. Conf. 2019, 3171–3174 (2019). Syed, S. T., Gerber, B. S. & Sharp, L. K. Traveling Towards Disease: Transportation Barriers to Health Care Access. J. Community Health 38, 976–993 (2013). Dunn, P. & Hazzard, E. Technology approaches to digital health literacy. Int. J. Cardiol. 293, 294–296 (2019). Tong, Y. et al. Low-cost sensor-integrated 3D-printed personalized prosthetic hands for children with amniotic band syndrome: A case study in sensing pressure distribution on an anatomical human-machine interface (AHMI) using 3D-printed conformal electrode arrays. PLOS ONE 14, e0214120 (2019). Hoeffelin, H., Jacquemin, D., Defaweux, V. & Nizet, J. L. A Methodological Evaluation of Volumetric Measurement Techniques including Three-Dimensional Imaging in Breast Surgery. BioMed Res. Int. 2014, e573249 (2014). Singh, M. et al. 3D printed conformal microfluidics for isolation and profiling of biomarkers from whole organs. Lab. Chip 17, 2561–2571 (2017). Kamath, A. A. et al. Workflow to develop 3D designed personalized neonatal CPAP masks using iPhone structured light facial scanning. 3D Print. Med. 8, 23 (2022). Telfer, S. & Woodburn, J. The use of 3D surface scanning for the measurement and assessment of the human foot. J. Foot Ankle Res. 3, 19 (2010). Semianov, K., Lebedev, A. & Semyanov, A. System and method for foot scanning via a mobile computing device. (2021). Zampogiannis, K., Fermuller, C. & Aloimonos, Y. cilantro: A Lean, Versatile, and Efficient Library for Point Cloud Data Processing. in Proceedings of the 26th ACM international conference on Multimedia 1364–1367 (ACM, 2018). doi: 10.1145/3240508.3243655 . Additional Declarations Competing interest reported. This study has been funded by Neatsy AI, Inc as an industry research collaboration with Massachusetts General Hospital's Foot and Ankle Research and Innovation Laboratory Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3868289","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":268593410,"identity":"6a64bee9-8bca-4126-84a1-0964301a5e73","order_by":0,"name":"Samir Ghandour","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABDUlEQVRIiWNgGAWjYFACHiBmA2L2BgZmBoYDcEHGBoJaeA6QrEUigUgt8v1rD34uKLNJ3D7zdfLrAoY78uYzkp99eMNgI7vhAHYtBjfeJUvPOJeWOOd27jbrGQzPDOfcSDOeOYchzRinFokzBtK8bYcTZ0jnbjPmYTjMOIPngDEzkJGIS4v8jDPGv8FaJM+CtdjP4Dn+GajlP04tDOd7zCC2SPBufgwyfAZ7D8iWAzi1GNzgMbPmOZdmPIMndxszj8HhZKCWYsY5BsnGM3E5rP+M8W2eMhvZGexnN3/mqThsO4OZfTPDmwo72T5cDgNGBwywSTAYwG3HoRwE+BFmMX/Ao24UjIJRMApGMAAApYpfXRD3kHwAAAAASUVORK5CYII=","orcid":"","institution":"Harvard Medical School, Massachusetts General Hospital","correspondingAuthor":true,"prefix":"","firstName":"Samir","middleName":"","lastName":"Ghandour","suffix":""},{"id":268593411,"identity":"6a0036c0-160e-4ce4-99c9-c78206c3e62b","order_by":1,"name":"Anton Lebedev","email":"","orcid":"","institution":"Neatsy, Inc","correspondingAuthor":false,"prefix":"","firstName":"Anton","middleName":"","lastName":"Lebedev","suffix":""},{"id":268593412,"identity":"73b3cb01-c68f-4234-86ab-cce7dc150524","order_by":2,"name":"Wei Shao Tung","email":"","orcid":"","institution":"Harvard Medical School, Massachusetts General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"Shao","lastName":"Tung","suffix":""},{"id":268593413,"identity":"671ff03a-75fd-43ba-9253-cd4f18e18539","order_by":3,"name":"Konstantin Semianov","email":"","orcid":"","institution":"Neatsy, Inc","correspondingAuthor":false,"prefix":"","firstName":"Konstantin","middleName":"","lastName":"Semianov","suffix":""},{"id":268593414,"identity":"a7051219-2e57-4beb-8dc7-3e24098e1393","order_by":4,"name":"Artem Semyanov","email":"","orcid":"","institution":"Neatsy, Inc","correspondingAuthor":false,"prefix":"","firstName":"Artem","middleName":"","lastName":"Semyanov","suffix":""},{"id":268593415,"identity":"673c4730-81db-4c78-9056-46fed55e334b","order_by":5,"name":"Christopher DiGiovanni","email":"","orcid":"","institution":"Harvard Medical School, Massachusetts General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Christopher","middleName":"","lastName":"DiGiovanni","suffix":""},{"id":268593416,"identity":"e753a1a8-5977-442a-878e-0c2b4e9e4d8b","order_by":6,"name":"Lorena Bejarano-Pineda","email":"","orcid":"","institution":"Harvard Medical School, Massachusetts General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Lorena","middleName":"","lastName":"Bejarano-Pineda","suffix":""},{"id":268593417,"identity":"d9ce4a16-6a2c-4178-b924-d3cfd17e1d00","order_by":7,"name":"Soheil Ashkani-Esfahani","email":"","orcid":"","institution":"Harvard Medical School, Massachusetts General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Soheil","middleName":"","lastName":"Ashkani-Esfahani","suffix":""}],"badges":[],"createdAt":"2024-01-16 01:29:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3868289/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3868289/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":50055156,"identity":"2d6d9136-b093-4c39-a9d0-452081a434c3","added_by":"auto","created_at":"2024-01-23 17:32:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":55641,"visible":true,"origin":"","legend":"\u003cp\u003eExample of a generated 3D model following a patient scan.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/0de9f4ead348ba7e72b9618f.png"},{"id":50054023,"identity":"2691cc01-72f5-4ca0-9cbb-574064d65862","added_by":"auto","created_at":"2024-01-23 17:16:17","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":69918,"visible":true,"origin":"","legend":"\u003cp\u003eTwo unique cases of HV surrogate angle measurements. The red dots represent the hallux point, toe point, and heel point. The heel point and hallux point create a line parallel to the x-axis of the point cloud coordinate system. Another green line is formed between the hallux and toe points. The intersection between both lines creates the surrogate angle denoted by a black angle curve and black arrow.\u003cbr\u003e\nHV, hallux valgus\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/abd09409f771420a5d23aaab.png"},{"id":50054024,"identity":"f81e0f73-80f3-4735-8584-4ba61e0ea041","added_by":"auto","created_at":"2024-01-23 17:16:17","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":43435,"visible":true,"origin":"","legend":"\u003cp\u003eThe filtrations (a) remove all the points that lie within the first 65% of the total length between the heel and base of the big toe (green) and the points that lie beyond the metatarsal bone along the sagittal axis as well as the points that are below the second decile of the total ball height (red). Thirty-five clusters (b) are created within the resultant scan area for further analysis.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/594e59e1abb9a5d5fb769b00.png"},{"id":50054028,"identity":"356c90e7-975f-4059-b33a-7ef10a2c1098","added_by":"auto","created_at":"2024-01-23 17:16:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":31765,"visible":true,"origin":"","legend":"\u003cp\u003eThe local minimum of each cluster is identified and denoted by a green point, while the between-cluster minima are identified and denoted by red points and associated red arrows.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/77b5b1777ab050912cd0d3c4.png"},{"id":50054027,"identity":"9ab75776-3c7c-432f-8d4d-085f8b3854de","added_by":"auto","created_at":"2024-01-23 17:16:17","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":52062,"visible":true,"origin":"","legend":"\u003cp\u003eThe proposed algorithm workflow from patient scanning, model creation, and analysis to HV detection. HV, hallux valgus\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/f3a3b4fdd862d04de8262b96.png"},{"id":50054557,"identity":"0d0986f1-2e1c-40d9-b187-0c12065d9a5f","added_by":"auto","created_at":"2024-01-23 17:24:17","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":21663,"visible":true,"origin":"","legend":"\u003cp\u003ePrecision-recall curves for negative (a) and positive (b) classes, respectively.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/b342ced68c0dc6b2cfd58f7b.png"},{"id":50054555,"identity":"fdbdaa0e-f3f9-4b29-9bc3-375a9e4dd5db","added_by":"auto","created_at":"2024-01-23 17:24:17","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":51725,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between the hallux valgus angle and surrogate angle. The colors distinguish the severity of HV: normal (green), mild (blue), moderate (orange), and severe (red).\u003c/p\u003e","description":"","filename":"Figure7.png","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/27f2ceb9ad56a8f786be0763.png"},{"id":64158752,"identity":"57b66e48-e4b1-4ae8-b515-e3d09b51aba9","added_by":"auto","created_at":"2024-09-09 06:35:31","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":804012,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3868289/v1/49b818b5-ebb4-46cf-ba7a-8233a976cd97.pdf"}],"financialInterests":"Competing interest reported. This study has been funded by Neatsy AI, Inc as an industry research collaboration with Massachusetts General Hospital's Foot and Ankle Research and Innovation Laboratory","formattedTitle":"The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eHallux valgus (HV) is a common anatomic foot deformity that describes an abnormally angulated, rotated, and laterally deviated great toe at the first metatarsophalangeal joint [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The prevalence of HV varies, with different epidemiological studies suggesting estimates ranging from 21\u0026ndash;70% globally, depending on the target population [\u003cspan additionalcitationids=\"CR3 CR4 CR5 CR6\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. HV seems to occur more commonly among women and older individuals [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Considered a degenerative joint disorder in adults, untreated HV can lead to progressive pain, stiffness, difficulty with footwear, functional limitations, edema, and arthritis in the long term [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe most reliable and commonly employed method of detecting and diagnosing HV is a thorough clinical and radiographic examination, which includes evaluation of the patient\u0026rsquo;s foot with observation of great toe position, range of motion, correctability, and determination of radiographic alignment by a healthcare professional. Radiographic imaging is commonly used to confirm the diagnosis and assess severity [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. While conservative treatment options exist for those who are primarily in the earlier stages of HV, including orthotic devices, physical therapy, and lifestyle modifications, unaddressed deformity often progresses with associated symptoms and may eventually necessitate a surgical solution that is more costly and exposes the patient to potential complications such as nonunion, malunion, recurrent deformity, infection, nerve injury, and other peri-surgical risks [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOver the last few decades, three-dimensional (3D) scanning technologies have been rapidly developing in both capacity and accessibility. Structured light and light detection and range (LIDAR) are 3D scanning technologies that actively light the scene to measure distances between target objects and the camera. Utilizing surface points that can be distinguished on the object, \u0026lsquo;depth\u0026rsquo; is calculated and used to create highly accurate 3D representations of the target object or the surrounding environment. While the LIDAR technology was mainly used for augmented reality or scanning large objects, such as rooms, structured light technology performed better on smaller objects, such as human anatomical structures [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. More recently, both LIDAR and structured light technologies have been successfully incorporated into newer smartphone models for various applications, including foot and ankle deformities [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Given the 3D nature of HV, this technology combined with computer vision-based analysis may allow patients to detect the deformity in their foot at earlier stages more effectively and subsequently seek treatment options to prevent further deformity progression and its associated morbidity. This study aims to assess the use of computer vision in detecting HV deformity by 3D foot scans obtained using structured light technology in smartphone cameras and determine its diagnostic accuracy with deformity severity when compared to clinical and radiographic examination as the accepted assessment standard.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eStudy Population\u003c/h2\u003e\n \u003cp\u003eThe protocol for this prospective study was approved by the Institutional Review Board (IRB no. 2022P001722). The study was performed in accordance with relevant guidelines and regulations granted by the IRB for this study. Informed consent was obtained from participants prior to enrollment. Patients were recruited from two orthopaedic foot and ankle clinics in our institution, which is a tertiary hospital in Boston, Massachusetts. Patients with foot pain for various reasons, including plantar fasciitis, cavovarus foot, pes planus, and pes cavus, were assessed for HV deformity. Feet that were deemed as having HV by a foot and ankle surgeon were labeled as cases, while feet that were not diagnosed with HV were labeled as controls. The inclusion criteria for the subjects recruited in this study were 1) adults aged\u0026thinsp;\u0026ge;\u0026thinsp;18 years old, 2) experiencing foot pain, 3) being able to ambulate and bear weight on their feet, and 4) having the capacity to provide informed consent for participation. The exclusion criteria were 1) the presence of fractures in the foot, including stress fractures and traumatic fractures, 2) the presence of neuropathy, neuronal injuries, or structural deformities related to chronic diseases (i.e., diabetic neuropathy, Charcot-Marie-Tooth disease, Charcot\u0026apos;s foot), 3) unable to stand on one foot, 4) structural deformities in the toes (mainly the first ray). For a random subset of participants who had anteroposterior radiographs of their feet, an expert radiologist measured the hallux valgus angle (HVA) and first intermetatarsal angle (IMA) for the correlation with the surrogate angle derived from the scan. The severity of HV, when present, was labeled according to these radiographic measurements.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e3D Scanning and Developing the AI Algorithm\u003c/h2\u003e\n \u003cp\u003eAn algorithm that estimates the surrogate HVA from the generated point cloud was utilized. To facilitate the use of the algorithm by clinical experts, this algorithm is deployed in an application that could run on smartphones (Neatsy App for iOS, Neatsy, Inc., Menlo Park, CA). The application enables the provider to obtain a 3D scan of the feet using a structured light camera system [\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]. This system generates a 3D point cloud, or an array of points, of the medial view of the patient\u0026apos;s foot while the patient maintains a single-footed posture with the camera fixed in a static or near-static position. The application assembles a sequence of Red-Green-Blue-Depth (RGBD) images, subsequently unifying them into a coherent point cloud, achieved through the application of an Iterative Closest Point (ICP) algorithm [\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e]. Detailed 3D models created can be found in Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eAs this novel diagnostic algorithm does not depend on radiographic imaging, the HVA and IMA cannot be measured from the point cloud. Instead, we chose superficial landmarks on the participant\u0026rsquo;s foot corresponding to the metatarsophalangeal (hallux point) and proximal interphalangeal (toe point) joint of the patient\u0026apos;s big toe on the XY plane to develop a surrogate angle. This surrogate angle, which can aid our model in identifying HV and, subsequently, its severity, is measured between two lines: 1) a line passing through the hallux point parallel to the x-axis, and 2) a line passing through the hallux and toe points \u003cstrong\u003e[\u003c/strong\u003eFigure \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cstrong\u003e]\u003c/strong\u003e. To justify the use of this surrogate angle for our algorithm, this measurement was compared to HVA and IMA taken from participant radiographs of the foot, if already available. Pearson\u0026rsquo;s correlation coefficient was used to determine the strength of the relationship between the surrogate angle and the validated HVA and IMA.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n \u003cp\u003eThe raw scanner output designates the z-axis as the depth axis (medial to lateral), aligns the x-axis approximately collinear with the vector from the heel to the toes, and positions the y-axis as the vertical axis. Mirroring is exclusively applied to the left-sided scans along the x-axis, which ensures that the x-coordinate of the heel point is less than that of the toe point. The right-sided scans inherently exhibit this characteristic without requiring any additional adjustments.\u003c/p\u003e\n \u003cp\u003eThe next step in the algorithm involves rotating the scan for alignment with both the XZ and XY planes. To provide a more intuitive understanding, picture a foot placed flat on the floor, with the heel against a wall. The XZ plane represents the floor, and the XY plane represents the wall. Orientation was achieved by the following:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003ePlane Selection\u003c/strong\u003e: The XZ or XY plane was chosen based on context and it should pass through the point in the scan with the minimum z or x coordinate, respectively.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eAngle Exploration\u003c/strong\u003e: The optimal alignment was found by systematically testing 100 angles within the range of -\u0026pi;/4 to \u0026pi;/4 with stepwise increments of \u0026pi;/200.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eAngle Selection\u003c/strong\u003e: The angle that results in the lowest error was selected, effectively aligning the scan with the chosen plane.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eRotation Angles\u003c/strong\u003e: The XZ plane was aligned in relation to the rotation angles around the z-axis. In contrast, the XY plane was aligned through sequentially considering the rotation angles around the X and Y axes.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eStability Enhancement\u003c/strong\u003e: Stability was ensured by repeating the outlined process twice and replacing the minimum point with the value corresponding to the 0.001 quantile.\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eNext, points that are not the hallux or toe points are filtered. To filter these points, the length of the scan, or the distance between the furthest limits on x-axis, is calculated. The points that lie within the first 65% of the total length between the heel and the base of the big toe (the area approximately defining the heel to the distal limit of the foot arch), as well as the points that lie beyond the 95% cutoff of the total length, namely, the phalanges, are removed. Another filtration is performed for the vertical dimension. Here, the ball height is approximated as the distance between the furthest points on the y-axis within the area defined after the index filtration. Then, all the points that are below the second decile of the total ball height are removed to eliminate the outliers near the floor. The resultant scan area is presented in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ea. The remaining points are divided into 35 clusters. Each cluster is represented by vertical strips of equal width that break each scan into 35 disjoint sets of points \u003cstrong\u003e[\u003c/strong\u003eFigure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb\u003cstrong\u003e]\u003c/strong\u003e. Within each cluster, a 3D point that is equivalent to the first, or shortest, decile in depth on the medial-to-lateral axis (z-axis) is identified and used in the following part of the algorithm. This method of selecting the minimum is preferred over the raw dataset as it removes any possible outliers and improves the overall stability of chosen points in terms of their z-coordinate.\u003c/p\u003e\n \u003cp\u003eNext, the local minima across the clusters are found \u003cstrong\u003e[\u003c/strong\u003eFigure \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cstrong\u003e]\u003c/strong\u003e. A local minimum is defined in our algorithm as a point whose z-coordinate is less than or equal to the z-coordinates of its two immediately neighboring points to the right and to the left. In most cases, two minima were identified, with the first minimum encountered from posterior to anterior (heel to toe) being of lesser magnitude than the second \u003cstrong\u003e[\u003c/strong\u003eFigure \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cstrong\u003e]\u003c/strong\u003e. For these instances, the first minimum encountered along the axis is designated as the hallux point, and the second is the toe point. In all remaining cases, the hallux point is ascribed to the sole local minimum characterized by the smallest z-coordinate, with the toe point being unequivocally identified as the last element of the array. By convention, the first point of the array on the z-axis cannot be the hallux point, as there are no neighboring points on its left side.\u003c/p\u003e\n \u003cp\u003eThe final step of the algorithm is to calculate the surrogate angle. As previously mentioned, this surrogate angle is not equivalent to either HVA or IMA. This surrogate angle, which can aid our model in identifying HV and its severity, is measured between two lines: 1) a line passing through the hallux point parallel to the x-axis, and 2) a line passing through the hallux and toe points [Figure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e]. The entire workflow of the algorithm can be found in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical Analysis\u003c/h2\u003e\n \u003cp\u003eThe area under the ROC Curve (AUC) and the area under the Precision-Recall Curve (PR-AUC) was calculated by using output angles without any post-processing and was utilized to assess model performance. For AUC, the DeLong method was applied to acquire the 95% confidence interval to ensure significance [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]. To obtain binary predictions of a participant\u0026apos;s condition, a classification threshold score is required. The F1 score, a machine learning metric used to evaluate model accuracy, for each possible classification threshold was calculated, and the threshold that maximized the F1 score was selected. Whilst algorithm training was not necessary for this methodology, the classification threshold selection process can be considered as a mode of training. Hence, the leave-one-out cross-validation (LOOCV) method was used for metric calculation. The LOOCV method was repeated for all \u003cem\u003eN\u003c/em\u003e, where \u003cem\u003eN\u003c/em\u003e is the study population. For each participant \u003cem\u003ei\u003c/em\u003e, the method calculated the classification threshold using \u003cem\u003eN-1\u003c/em\u003e scans (i.e., without the participant \u003cem\u003ei\u003c/em\u003e\u0026rsquo;s scan) and compared the angle of the excluded scan with the computed value. After cross-validation, a list of binarized predictions is produced and used for calculating the F1 score, Youden\u0026rsquo;s index, sensitivity (recall), specificity, accuracy, precision (positive predictive value), and negative predictive value. Pearson\u0026rsquo;s correlation test was used to determine the association between the surrogate angle and HVA or IMA measured in a subset of participants who had existing weightbearing anterior-posterior radiographs. A cluster analysis was performed to further dissect the correlation between the surrogate angle and HVA \u003cstrong\u003e[\u003c/strong\u003eFigure \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e\u003cstrong\u003e]\u003c/strong\u003e. The initial clusters were formed based on the typical ranges of HVA for each grade of HV in adults: normal (\u0026lt;\u0026thinsp;15\u0026deg;), mild (15\u0026ndash;25\u0026deg;), moderate (25\u0026ndash;35\u0026deg;), and severe (\u0026ge;\u0026thinsp;35\u0026deg;). To find the surrogate angle\u0026rsquo;s classification threshold binary classification problems were created based on the minimum HVA to be categorized in each grade of HV (i.e., \u0026lt;\u0026thinsp;15\u0026deg; vs. 15\u0026ndash;100\u0026deg;, or \u0026lt;\u0026thinsp;25\u0026deg; vs. 25\u0026ndash;100\u0026deg;). The One-vs-All multiclass classification was used to assess the predictive model performance in categorizing HV severity.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"RESULTS","content":"\u003cp\u003eIn total, 120 cases and controls(240 feet) were included in the study. Twenty-nine participants had bilateral HV, 12 had unilateral HV, and 79 had normal feet. In total, seventy feet (29.1%) had HV, while 170 (70.8%) were labeled normal following clinical assessment by a foot and ankle surgeon The mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation age of our participants was 37.98\u0026thinsp;\u0026plusmn;\u0026thinsp;18.23 years with 48 (40%) males and 72 (60%) females. The racial composition of the scanned participants varied, with 61 identifying as Caucasian, 31 as Asian/Pacific Islander, six as African American, six as Middle Eastern, and 16 others as mixed race. Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e presents the classification metrics that were calculated on binarized output after LOOCV.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eClassification metrics calculated on the binarized output after LOOCV.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMetric Name\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1 Score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision (PPV)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall (Sensitivity)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eYouden\u0026rsquo;s\u003c/p\u003e\n \u003cp\u003eIndex\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNegative Predictive Value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eMetric Value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.741\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.845\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.726\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.757\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.882\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.639\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.892\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eAmongst the 120 patients recruited for the study, 25 random patients had 29 foot radiographs that were obtained as part of their diagnostic workup prescribed by an expert clinician. We measured IMA and HVA on these patients, and Pearson\u0026apos;s correlation test showed that the correlation of the surrogate angle with HVA was 0.91 (95% CI: [0.81, 0.96]; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and 0.65 (95% CI: [0.37, 0.82]; p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.001) for IMA. The optimal classification thresholds found for the surrogate angle using a brute search algorithm maximizing the F1 score for the created clusters were different from the HVA-based clusters. Additional metrics using the union of the two higher-grade clusters were calculated to attempt to improve model performance. Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows the resultant classification thresholds based on the surrogate angle for the three final clusters, including the combined moderate-severe cluster. AUC and PR-AUC curves for positive and negative classes were calculated on row angles. The AUC score was 0.947 (95% CI: [0.916, 0.971]), calculated via the DeLong method [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]. PR curves are represented in Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e. The PR-AUC for the positive class was 0.89 and 0.92 for the negative class.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformance metrics of the cluster analysis between the surrogate angle and HVA.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTarget Degree Interval\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1 Score\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrecision (PPV)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecall (Sensitivity)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eYouden\u0026rsquo;s Index\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNegative Predictive Value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u0026ndash;15\u0026deg;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.952\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.966\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.947\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.947\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e15\u0026ndash;25\u0026deg;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.931\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.913\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e25\u0026ndash;35\u0026deg;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.600\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.862\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.600\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.600\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.917\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.517\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.917\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e25\u0026ndash;100\u0026deg;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.957\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.966\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.917\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e35\u0026ndash;100\u0026deg;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.897\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.913\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.746\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.955\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eUsing structured light technology, the proposed 3D scanning algorithm showed reliable accuracy for detecting HV and determining the severity of the deformity. With a specificity of 0.882, this algorithm proves to be a reliable and accurate diagnostic tool of HV deformity independent of clinical or radiographic examination. For HV severity assessment, the correlation of the surrogate angle with HVA (r\u0026thinsp;=\u0026thinsp;0.8915; p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001) is stronger than that of IMA (r\u0026thinsp;=\u0026thinsp;0.6598; p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001). The surrogate angle relies on the extent of the malalignment of the hallux, which corresponds to increased HVA. This strong correlation with HVA is consistent with large-scale data assessment of pre-existing foot scans by Jiao et. al when attempting to calculate HVA automatically from 3D foot scans without referring to radiographic assessment [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Our novel approach with a surrogate angle that is correlated with the radiographic assessment of a random subset of our scanned feet goes one step further in establishing this correlation with the HVA. The IMA does not manifest prominent exterior landmarks that are easily identifiable by our targeted scanning methodology, which is an important prerequisite when using structured light\u0026mdash;as co-existent metatarsus adductus might interfere with HV assessment given a decreased IMA despite a notable HV deformity. This may be the reason behind the IMA\u0026rsquo;s relatively lower correlation to the surrogate angle in relation to severity, albeit still being significant.\u003c/p\u003e \u003cp\u003eDeveloping a phone-based diagnostic tool for identifying HV can vastly improve accessibility and early awareness facilitating prompt treatment for a great proportion of patients with this deformity [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In addition to logistical challenges such as patient mobility and transportation precluding patients from attending clinics in person, a common theme identified during this prospective study deterring patients from seeking expert consultation is embarrassment and shame surrounding the aesthetics of their feet [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Obviating the need for in person visits to assess and diagnose HV deformity at the clinic may alleviate the logistical, financial, and occupational costs these patients may incur. Our 3D scanning algorithm has the potential to provide a convenient solution for these patients in the privacy of their own homes, with even greater potential for expansion to remote monitoring of other orthopaedic pathologies. The use of such accessible technology at home might also help improve health literacy and early identification of HV for those who are not well-versed in foot and ankle pathologies [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Likewise, the use of commonly available devices such as smartphones for this technology may enable patients from low-resource or underserved regions to have access to healthcare by obtaining a diagnosis and treatment recommendations directly and promptly. This expands the options for telemedicine and personalized recommendations for this patient-population.\u003c/p\u003e \u003cp\u003eRecently, structured light technology has been employed in many applications within the medical field, showing great potential in various areas of healthcare, wearable devices, and human-machine interfaces. Its ability to create accurate three-dimensional models of human anatomic structures with minimal hardware requirements makes it easily accessible to healthcare providers and researchers alike. For instance, such technology can create digital models that enable their developers to create anatomically conformal prostheses [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Others have explored the non-inferiority of Structured Light in 3D volumetric imaging when compared to other 3D scanning technologies [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Additionally, Structured Light has been shown to be effective in creating conformal microfluidic devices that can interface with the surface of whole organs, allowing researchers to better understand biological tissue and their biomarkers [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Singh et al. utilized smartphone Structured Light facial recognition technology to develop personalized neonatal CPAP masks [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. These examples show the reliability of structured light applications in addition to the technology\u0026rsquo;s potential impact and accessibility, all while utilizing low-cost and commercially available devices such as personal smartphones.\u003c/p\u003e \u003cp\u003eWhile initially used by footwear manufacturers as a tool for their customers to configure shoe size and morphology for fitting, 3D scanning of the foot is becoming a valuable tool for understanding foot deformities and pathologies [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. More specifically, Yamashita et al. have utilized a similar 3D scanning technique to explore skeletal structural features and variations in those with HV in different age groups, showing that certain skeletal features of the midfoot influence HV and its progression with age [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. In another study, Yamashita et al. used 3D scans of the foot in children to reliably predict the risk of HV development according to analytical parameters abstracted from these scans [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Such an application will not only enable earlier detection but also prompt intervention at earlier stages.\u003c/p\u003e \u003cp\u003eCreating and validating a surrogate angle to diagnose HV without relying on radiographic measurements obviates the requirement for radiological examination for diagnosis, reducing the exposure to radiation, time, and overall treatment cost. However, given that this software application was developed with patient independence in mind, it is critical to appreciate the demographic HV affects most.\u003c/p\u003e \u003cp\u003eLimitations of this technology may affect specific patient-population groups. Whilst HV does not exclusively affect older patients, a significant proportion exhibits some level of deformity, symptoms, or both. As such, because of the brisk development of technology in the past few decades, many of the patients that fall within the older age category may not own a smartphone, be accustomed to operating advanced software, or have wireless connections with the bandwidth to support large volume data transfers. Another limitation exists in the expectation that these patients would be able to obtain these scans themselves in a safe and viable manner. Certain maneuvers required within the process, such as weightbearing on one foot or propping the smartphone against a wall, may not be achievable for some patients with disability. Lastly, while the algorithm can reliably predict the presence of HV, distinguishing between moderate and severe cases remains a limitation. It is important to note that the current smartphone 3D scanning platform should probably not be used in isolation as a sole diagnostic and treatment tool as final diagnosis and care determination should always be made by a healthcare professional. However, continued development and advancements of this 3D scanning tool will eventually increase its reliability and may even someday obviate the need for early clinical evaluation and intervention by providing personalized digital conservative recommendations. Future directions for research and development include creating alternative ways for patients to obtain these scans independently, as well as expanding this technology to detect other foot deformities or maladies such as pes planus and pes cavus, which saves both patients and clinicians time and operative bandwidth.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eDATA AVAILABILITY STATEMENT\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSome of the datasets generated and/or analyzed during the current study are available in the study\u0026rsquo;s repository here. This repository contains the minimum information required to substantiate the study\u0026rsquo;s findings and analysis without compromising participant confidentiality. Further detailed data collected during the current study are available from the corresponding author upon reasonable request. The data that support the methodology of this study are available from Neatsy Inc. (Menlo Park, CA), but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available. Such data are, however, available from the authors upon reasonable request and with permission of Neatsy Inc. (Menlo Park, CA).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eACKNOWLEDGEMENTS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe want to acknowledge Nour Nassour, MD and Kendal Toy, MS for their contributions in technical suggestions and proofreading the manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAUTHOR CONTRIBUTIONS\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSG\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; and draft and substantive revision of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAL\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; creation of new software used in the work; and draft of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWST\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design; acquisition, analysis and interpretation of data; and draft and substantive revision of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eKS\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design in addition to creation of new software used in the work.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAS\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design in addition to creation of new software used in the work.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCWD\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design in addition to draft and substantive revision of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLBP\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design; data acquisition; and draft and substantive revision of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSAE\u0026nbsp;\u003c/strong\u003ehas made substantial contributions to the conception and design; data acquisition; creation of new software used in the work; and draft and substantive revision of the work provided.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eADDITIONAL INFORMATION\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests Statement\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the authors in this study declare no potential conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRoddy, E., Zhang, W. \u0026amp; Doherty, M. Prevalence and associations of hallux valgus in a primary care population. Arthritis Care Res. 59, 857\u0026ndash;862 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenvenuti, F., Ferrucci, L., Guralnik, J. M., Gangemi, S. \u0026amp; Baroni, A. Foot Pain and Disability in Older Persons: An Epidemiologic Survey. J. Am. Geriatr. Soc. 43, 479\u0026ndash;484 (1995).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDunn, J. E. Prevalence of Foot and Ankle Conditions in a Multiethnic Community Sample of Older Adults. Am. J. Epidemiol. 159, 491\u0026ndash;498 (2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElton, P. J. \u0026amp; Sanderson, S. P. A chiropodial survey of elderly persons over 65 years in the community. Public Health 100, 219\u0026ndash;222 (1986).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWhite, E. G. \u0026amp; Mulley, G. P. Footcare for Very Elderly People: A Community Survey. Age Ageing 18, 275\u0026ndash;278 (1989).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeveille, S. G. \u003cem\u003eet al.\u003c/em\u003e Foot Pain and Disability in Older Women. Am. J. Epidemiol. 148, 657\u0026ndash;665 (1998).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen, U.-S. D. T. \u003cem\u003eet al.\u003c/em\u003e Factors associated with hallux valgus in a population-based study of older women and men: the MOBILIZE Boston Study. Osteoarthritis Cartilage 18, 41\u0026ndash;46 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoughlin, M. J. \u0026amp; Jones, C. P. Hallux Valgus: Demographics, Etiology, and Radiographic Assessment. Foot Ankle Int. 28, 759\u0026ndash;777 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCho, N. H., Kim, S., Kwon, D.-J. \u0026amp; Kim, H. A. The prevalence of hallux valgus and its association with foot pain and function in a rural Korean community. J. Bone Joint Surg. Br. 91-B, 494\u0026ndash;498 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarnett, C. H. The normal orientation of the human hallux and the effect of footwear. J. Anat. 96, 489-494.1 (1962).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMenz, H. B. \u0026amp; Lord, S. R. The Contribution of Foot Problems to Mobility Impairment and Falls in Community-Dwelling Older People. J. Am. Geriatr. Soc. 49, 1651\u0026ndash;1656 (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoughlin, M. J., Saltzman, C. L. \u0026amp; Nunley, J. A. Angular measurements in the evaluation of hallux valgus deformities: a report of the ad hoc committee of the American Orthopaedic Foot \u0026amp; Ankle Society on angular measurements. Foot Ankle Int. 23, 68\u0026ndash;74 (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoughlin, M. J. \u0026amp; Jones, C. P. Hallux valgus and first ray mobility. A prospective study. J. Bone Joint Surg. Am. 89, 1887\u0026ndash;1898 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreisberg, J. K. Hallux valgus and first ray mobility. A prospective study. \u003cem\u003eJ. Bone Joint Surg. Am.\u003c/em\u003e 90, 1166; author reply 1166\u0026ndash;1167 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVogt, M., Rips, A. \u0026amp; Emmelmann, C. Comparison of iPad Pro\u0026reg;\u0026rsquo;s LiDAR and TrueDepth Capabilities with an Industrial 3D Scanning Solution. Technologies 9, 25 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYamashita, T., Yamashita, K., Sato, M., Kawasumi, M. \u0026amp; Ata, S. Foot-surface-structure analysis using a smartphone-based 3D foot scanner. Med. Eng. Phys. 95, 90\u0026ndash;96 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKobayashi, T. \u003cem\u003eet al.\u003c/em\u003e A Simple 3D Scanning System of the Human Foot Using a Smartphone with a Depth Camera. \u003cem\u003eProc. 3DBODYTECH\u003c/em\u003e 2018\u003cem\u003e\u0026ndash;9th Int. Conf. Exhib. 3D Body Scanning Process. Technol. Lugano Switz. 16\u0026ndash;17 Oct 2018\u003c/em\u003e 161\u0026ndash;169 (2018) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.15221/18.161\u003c/span\u003e\u003cspan address=\"10.15221/18.161\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeLong, E. R., DeLong, D. M. \u0026amp; Clarke-Pearson, D. L. Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach. Biometrics 44, 837\u0026ndash;845 (1988).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiao, Y., Džeroski, S. \u0026amp; Jurca, A. Analysis of hallux valgus angles automatically extracted from 3D foot scans taken in North America, Europe, and Asia. Ergonomics 66, 1164\u0026ndash;1175 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYamashita, K., Yamashita, T., Sato, M., Kawasumi, M. \u0026amp; Takase, Y. Development of a quantitative measurement system for three-dimensional analysis of foot morphology using a smartphone. \u003cem\u003eAnnu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. Annu. Int. Conf.\u003c/em\u003e 2019, 3171\u0026ndash;3174 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSyed, S. T., Gerber, B. S. \u0026amp; Sharp, L. K. Traveling Towards Disease: Transportation Barriers to Health Care Access. J. Community Health 38, 976\u0026ndash;993 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDunn, P. \u0026amp; Hazzard, E. Technology approaches to digital health literacy. Int. J. Cardiol. 293, 294\u0026ndash;296 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTong, Y. \u003cem\u003eet al.\u003c/em\u003e Low-cost sensor-integrated 3D-printed personalized prosthetic hands for children with amniotic band syndrome: A case study in sensing pressure distribution on an anatomical human-machine interface (AHMI) using 3D-printed conformal electrode arrays. PLOS ONE 14, e0214120 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoeffelin, H., Jacquemin, D., Defaweux, V. \u0026amp; Nizet, J. L. A Methodological Evaluation of Volumetric Measurement Techniques including Three-Dimensional Imaging in Breast Surgery. \u003cem\u003eBioMed Res. Int.\u003c/em\u003e 2014, e573249 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh, M. \u003cem\u003eet al.\u003c/em\u003e 3D printed conformal microfluidics for isolation and profiling of biomarkers from whole organs. Lab. Chip 17, 2561\u0026ndash;2571 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKamath, A. A. \u003cem\u003eet al.\u003c/em\u003e Workflow to develop 3D designed personalized neonatal CPAP masks using iPhone structured light facial scanning. 3D Print. Med. 8, 23 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTelfer, S. \u0026amp; Woodburn, J. The use of 3D surface scanning for the measurement and assessment of the human foot. J. Foot Ankle Res. 3, 19 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSemianov, K., Lebedev, A. \u0026amp; Semyanov, A. System and method for foot scanning via a mobile computing device. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZampogiannis, K., Fermuller, C. \u0026amp; Aloimonos, Y. cilantro: A Lean, Versatile, and Efficient Library for Point Cloud Data Processing. in \u003cem\u003eProceedings of the 26th ACM international conference on Multimedia\u003c/em\u003e 1364\u0026ndash;1367 (ACM, 2018). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/3240508.3243655\u003c/span\u003e\u003cspan address=\"10.1145/3240508.3243655\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-3868289/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3868289/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHallux valgus (HV) is a common foot deformity. Traditional detection methods include clinical examination and radiographic imaging, which, although reliable, often remain inaccessible to many due to existing care barriers. This study introduces an innovative approach to computer vision analysis and phone camera-based 3D scanning technology to detect and assess HV severity. We evaluated the accuracy of this method against routine clinical examination as the currently accepted assessment standard. Our study included 120 participants, resulting in 240 foot scans, with a diverse demographic representation. The computer vision algorithm utilized a surrogate angle, automatically derived from the 3D scans, to identify the severity of HV, and its correlation with traditional radiographic measurements for HV. Our findings reveal that computer vision-based detections offer high accuracy, with an Area Under the Curve (AUC) score of 0.947, presenting a promising alternative to conventional methods. This technology offers promise for increasing access to HV detection, potentially aiding in earlier diagnosis as well as non-operative treatment options that may ultimately reduce the need for surgical intervention. Its ease of use and application in telemedicine contexts has the potential, moreover, to significantly benefit patients in remote or underserved areas and expand capacity to promote similar care improvement in other areas of musculoskeletal disease.\u003c/p\u003e","manuscriptTitle":"The Use of AI For Hallux Valgus Assessment via Mobile Phone-Based 3D Camera Scan","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-23 17:16:13","doi":"10.21203/rs.3.rs-3868289/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"50c07d06-ac46-4e77-8d0d-8aee47f67eb7","owner":[],"postedDate":"January 23rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":28293314,"name":"Biological sciences/Biotechnology"},{"id":28293315,"name":"Health sciences/Anatomy"},{"id":28293316,"name":"Health sciences/Diseases"},{"id":28293317,"name":"Health sciences/Health care"},{"id":28293318,"name":"Health sciences/Medical research"},{"id":28293319,"name":"Health sciences/Signs and symptoms"},{"id":28293320,"name":"Physical sciences/Engineering"}],"tags":[],"updatedAt":"2024-09-09T06:27:23+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-23 17:16:13","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3868289","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3868289","identity":"rs-3868289","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}