Recent Innovations Brought about by Weight-Bearing CT Imaging in the Foot and Ankle: A Systematic Review of the Literature

Abstract

The decade from 2010–2020 has seen the development of cone beam weight-bearing CT (WBCT) as a major innovation in the foot and ankle realm, becoming an important modality for bone and joint imaging. The ability to provide three-dimensional images of the naturally loaded skeleton has enabled several subsequent innovations to arise with aims to hasten image processing and to extend the clinical applications of WBCT. The objective of this work was to identify, categorize and explain those emerging techniques. We performed a structured review of the literature according to PRISMA standards, finally including 50 studies. We subsequently proposed a classification of these techniques. Segmentation and distance mapping were identified as key features. We conclude that although WBCT has already been adopted in a number of clinical communities with an immediate improvement in patient workflows, adoption of advanced techniques is yet to come. However, that relies mostly not on the technology itself, but on improvements in AI software allowing practitioners to quickly process images in daily practice and enabling the clinicians to obtain an accurate three-dimensional evaluation of the segment considered. Standardization will be paramount to amass large amounts of comparable data, which will fuel further innovations in a potentially virtuous circle.

1. Introduction

Among computed tomography (CT) techniques, cone beam CT (CBCT) provides three-plane tomography, radiography and 3D reconstructions in a single high-speed, versatile package which can image the entire human skeleton. As an identified technique, it was first published in the journal European Radiology in 1998 for use in the dental arena, following the works of a team led by P. Mozzo from the Department of Medical Physics at the University of Verona, Italy [1]. The dental field has since then become one of the most important uses of the technology, but it was not the first citation of cone beam in the literature, which dates back to 1979 [2], nor the first clinical application, which was in the vascular domain due to its ability to visualize highly contrasted material [3]. The technology is inspired by the mathematical concepts initiated by Hounsfield for parallel fan beam or multi-detector CT (MDCT) [4] using the Fournier and Radon transforms to produce spatially referenced slices of the anatomy. Later, in 1984, Feldkamp [5] described an algorithm based on convolution and back-projection designed to help with reconstruction of acquisitions with incomplete rotations, triggering the possibility of more practical and flexible gantry systems.

In orthopedics, the research around CBCT was initiated by Bab et al. [6] in 2001, for whom intended clinical uses were to be in orthopedic and chest applications. The first dedicated use in orthopedics was described in 2011 by Zbijewski et al. [7]. The first mention of a dedicated extremity CBCT device was by Muhit et al. in 2012 [8], and the first mention of weight-bearing CBCT (WBCT) in the lower limbs was in 2013 by Tuominen et al. [9]. The first mention of WBCT in the foot and ankle concerned the pes planovalgus [10], but that was simulated weight-bearing. The first publications on the use of true (i.e., under body weight) WBCT in the foot and ankle were in 2013 by Collan et al. [11] on the biomechanics of the first ray, followed by Richter in 2014 on the superiority of 3D WBCT measurements as compared to 2D or non-weight-bearing WBCT [12].

A quick look at the literature suggests that WBCT would allow clinicians to optimize acquisition speed, therefore reducing operating costs, with a smaller footprint compared to traditional two-dimensional radiography (2DXR) plus three-dimensional non weight-bearing fan beam MDCT in diverse areas [7,8,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34]. This would enable clinicians to obtain marked improvements in the operative workflow [32,35]. Besides advantages for the clinicians, immediate advantages for patients are also huge time gains in their care pathway [32], more accurate diagnoses (i.e., increased chances of detecting injuries commonly missed through standard imaging) [12,30] and lower radiation doses [32,36,37] (Figure 1).

However, it also seems that the advantages of CBCT are somewhat overshadowed by the weight-bearing side of the equation, since not many authors have focused on non-weight-bearing cone beam imaging so far [29,30,32,38]. Indeed, it is in the foot and ankle field that the industry has chosen to chase initial prospects, likely because this is where 3D combined with weight-bearing have shown the most evident improvements over the conventional 2DXR-MDCT sequence. In turn, this was due to the complex anatomy of the foot, including 28 bones with a variety of shapes and sizes, making it hard to correctly assess without 3D or without bearing weight [12]. A large body of literature preceding the WBCT period has already insisted on the importance of weight-bearing while imaging the foot and ankle, using custom devices to simulate weight-bearing with conventional prone MDCT protocols [10]. The advent of WBCT has since been seen as a step forward for clinicians and researchers specializing in the feet and ankles from the fields of orthopedics, biomechanics and engineering, soon reaching out to scientific societies and computer scientists [15], while new software companies have emerged to capitalize on the possibilities offered by WBCT. The vast majority of the innovations reported in this review are issued from a collaboration between the main stakeholders (WBCT manufacturers, clinicians and engineers) to solve problems regarding musculoskeletal pathology, for the ultimate benefit of patients. They are the result of the development of post-processing visualization (qualitative assessment) or measurement (quantitative assessment) software to make the best use of the available naturally 3D weight-bearing datasets.

In 2016, the International WBCT Study Group was formed on the initiative of pioneering researchers and officially institutionalized as an independent international non-profit scientific association based in Ghent, Belgium in 2017 and re-named the International WBCT Society (https://www.wbctsociety.org, accessed on 5 January 2024). It has since then, amongst other activities, been closely monitoring all relevant scientific publications on the subject, which has since seen an exponential growth. The object of the present review is to report, classify and explain the most recent innovations brought about by WBCT in the foot and ankle field, with a critical discussion of the evidence provided so far in terms of advantages, limitations and future areas of development.

2. Materials and Methods

This is a systematic review of the recent innovations brought by WBCT to the foot and ankle field. In order to specify the scope of the review, we defined ‘recent’ as after 2012, the year Muhit et al. [8] published the first paper on a dedicated extremity CBCT device. We defined innovative techniques as techniques which could not be performed using the conventional 2DXR-MDCT sequence (i.e., absence of concomitant 3D and natural weight-bearing). This would, in theory, include conventional measurements historically performed on weight-bearing 2DXR, performed on 3D WBCT datasets, for instance the first to second intermetatarsal angle [39] in the forefoot, or the Saltzman angle [14] in the hindfoot. In fact, most WBCT devices have the ability to produce digitally reconstructed radiographs (DRR) [40] to help with the user’s learning curve in transitioning from the 2D to the 3D environment. This is possible because WBCT produces a digital clone of the patient’s foot and ankle structure, which can in turn be virtually radiographed from multiple angles (Figure 2).

Classical measurements can therefore be performed in a standard way. However, this technique inevitably produces the same biases as conventional 2DXR. Otherwise, classical angles can be measured within the 3D dataset, using multiplanar reconstruction (MPR) views. In this case, a particular slice or slab must be chosen in order to perform the measurement, which requires multiple iterative changes in the orientation of the MPR, because the points which define the angles of interest are not necessarily on the same plane. For instance, the tibia longitudinal axis, the center of the ankle joint and the lowest weight-bearing point of the calcaneus do not belong to a single vertical plane. For the corresponding angle to be measured manually, a slice has to be found by tuning the orientation of the MPR dataset, but again, it is often not possible to find a slice which includes all three points. Another possibility is to select one segment of the angle (the tibia axis) on one plane and compare it to the other side (the calcaneal axis) using the available software, but measuring a single angle from different slices is not always an available function. A third method consists of manually recording 3D coordinates for the points of interest (in the present example, the tibia extremity and the center of the ankle joint’s lowest calcaneal point) and calculating the angle using standard trigonometry. Whatever the solution or combination of solutions, this method is always time-consuming [41] and introduces a new kind of bias, which we may call ‘slice’ or ‘slab’ bias, materializing as we must choose a particular slice from which to take the measurements, based on surface landmarks which may vary, being dependent on the operator’s habits or knowledge of the anatomy. Although these methods were and are still necessary to perform and describe throughout the process of standardization of measurement methods in WBCT research [15], they inevitably present practical disadvantages which prevent inclusion in the advanced methods review, for we aimed to describe here methods which actually improve the clinical workflow, not slow it down. Hence, articles describing such methods were not included for retrieval.

The PubMed database was used to identify relevant scientific references for the study. We included initially all references regarding CBCT and weight-bearing or WBCT, using the following key words: weightbearing, weight-bearing, standing, extremity, cone beam, CT, computed tomography, multiplanar, foot, ankle, ankle, ankle joint, ankle, joint, ankle joint, ankle. Filters were used to exclude references without an abstract or which were not available in English, French or German. Screening was then performed by two independent reviewers (FL and AB, both senior orthopedic foot and ankle consultants) at different places and times. No automated tool was used for this research. In cases of disagreement between the reviewers, inclusion or exclusion of the concerned references was resolved through discussion, and inclusion was retained only upon agreement by the two reviewers. Articles of interest were then retrieved for analysis.

Since this study describes innovative methods and does not report or meta-analyze numerical data from patients such as demographic or clinical data, no statistical method was applied. A review of the study protocol by our institutional review board was not deemed applicable.

3. Results

3.1. Pre-Screening

A total of 212 studies were initially identified. Five references were excluded at this stage due to returning results clearly outside of the scope of the study such as equine standing CT. At the end of the selection process, 50 studies were included as reported in the dedicated PRISMA-compliant (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) flow chart [42] (Figure 3). The median and mean sample size were 31 and 59 cases, respectively (range from 1 to 500 cases).

3.2. Screening

A total of 207 references were screened. Forty-nine studies were excluded based on the following criteria:

• Studies prior to 2012
• Studies of radiation dosage
• Simulated weight-bearing or absence thereof
• Use of WBCT for evaluation of surgical results without innovative methods

A total of 158 references were sought for retrieval. Seventy-eight further references were excluded based on the following criteria:

• WBCT classifications
• New knowledge but absence of an innovative technique
• Description of normal or pathological anatomy
• Multi-level biomechanical investigation

A total of 80 references were assessed for eligibility. A further 28 references were excluded based on the following criteria:

• Innovative WBCT techniques; study concerning the knee (8 studies)
• Two-dimensional method performed within 3D volume (7 studies)
• Clinical measurement or classification study (5 studies)
• Pre-processing techniques (9 studies)

3.3. Organization of the WBCT Workflow: Proposal for a Classification of Recent Innovations

A critical analysis of the 50 studies included enabled the authors to classify recent innovative techniques brought about by WBCT in the foot and ankle into categories. These categories are based on the nature of the innovations and on their positioning within the patient or image processing workflows. Some of these are identified in italic fonts below as hypotheses from the authors, based on the literature that was excluded from this specific work because it did not directly apply to our object, but which could in theory be applied to the foot and ankle field. For clarity, we included these in the classification, as well as pre-processing techniques in the initial exhaustive screening of the literature summarized and referenced hereunder.

• Techniques for image acquisition

  ◦

  Dynamic/augmented stress techniques

  ▪

  Ankle instability

  • Shod auto-varus [43]
  • Angled jigs/wedges [43]

  ▪

  Syndesmotic instability

  • Torque stress

  ▪

  Coleman block test [19]

  ◦

  Static techniques

  ▪

  Combination with pedography [44]

• Computerized techniques for image processing

  ◦

  Pre-processing (processing of the raw data before image rendering)

  ▪

  Metal artifact reduction

  ▪

  Movement Artifact reduction

  ◦

  Post-processing

  ▪

  3D biometric techniques without segmentation

  • Manual

    ◦

    Linear HU assessment of joint space width [25]

    ◦

    Middle subtalar facet uncoverage [45]

  • Automatic/semi-automatic:

    ◦

    Foot–ankle offset [19,44,46,47,48]

  ▪

  3D biometric techniques with segmentation (manual, semi-auto or auto)

  • Segmentation techniques

    ◦

    Statistical shape modelling (SSM) [49]

    ◦

    Mimics (materialization) [27,50,51]

    ◦

    CurvebeamAI [52]

    ◦

    Disior/Paragon [16,21,26,53,54,55,56]

  • Advanced techniques issued from segmentation (semi-automatic or fully automatic)

    ◦

    3D bone absolute and relative relationships reporting [16,21,22,26,52,53,54,55,56,57,58,59,60,61,62,63,64]

    ◦

    3D Joint space width (3D-JSW) mapping: distance mapping [65,66,67,68,69]

    ◦

    Coverage mapping [70]

    ◦

    Surface measurements

    ▪

    Syndesmosis [17]

    ▪

    Lisfranc [71]

    ◦

    Volumetric measurements

    ▪

    Syndesmosis [49,72,73,74]

    ▪

    Lisfranc [71]

    ◦

    Center of rotation assessment [75]

• Advanced clinical applications derived from computerized techniques

  ◦

  Customized/patient specific surgical jigs [76]

  ▪

  Supra malleolar osteotomy [18]

  ▪

  Total ankle replacement [77]

  ◦

  Robotic surgical protocols

  ◦

  Personalized risk assessment [47]

Following this exhaustive screening of the literature, the authors proposed a systematic classification of the WBCT computerized workflow to illustrate the role of each advanced technique associated with its development (Figure 4).

4. Discussion

The present systematic review found that recent innovations reported in WBCT literature are mostly post-segmentation techniques. The most reported tools aim to provide quantitative 3D bone positions, followed by visualization tools, the most frequently reported being joint space mapping or equivalents. Segmentation is not in itself an innovation, nor is it a clinically applicable method, but its implementation is an indispensable step in the development of innovations bearing the fruits of WBCT dissemination.

4.1. Limits

We acknowledge several limits in the present work. Firstly, the quality of a literature review is only scientifically as good as the quality of reviewed research, which is valid for every ‘secondary’ study. The level of evidence considered here being mostly level III (with the exception of a single level II study [41]), with level IV and V studies as well, our systematic review can be considered level V. However, given the research available on recent innovations in WBCT concerning imaging, most investigations can be performed ex vivo, rendering level II and level I research mostly superfluous. Furthermore, being at the diagnostic level for most (with the exception of patient-specific applications), the impact on treatment outcomes is still very difficult to assess, being a multifactorial problem with implications that spread out through a timescale that far exceeds the development of innovations that we chose to define as recent. Secondly, not all available research databases were assessed, as we limited the scope of this review to the records found on a single scientific database (PubMed). However, this method was chosen after a preliminary search of the most used other databases (Embase, Cochrane, Scopus, Web of Science) which did not lead to find any additional reference related to the development of WBCT. Thirdly, our classification of advanced techniques could be deemed arbitrary. However, we would like to emphasize that, to the best of our knowledge, no author so far has proposed any specific classification system in this area, so the present is intended as a proposal for a flexible benchmark to work from, not the immutable ground truth. The authors would be happy to revise or replace the proposed classification as progress is made, new methods arise and the technical landscape evolves in the future. Finally, we have only considered the foot and ankle field in the present work, but we fully acknowledge that WBCT devices already encompass the knee [20,65,78], with devices already available to image the pelvis and the upper limbs. There is no doubt that soon, devices able to image the full weight-bearing skeleton will be available. In this context, there is no doubt that, pending necessary adaptations locally, the recent innovations reported here will be disseminated to other joints, including the spine and the shoulders. One well-reported example is distance or 3D-JSW mapping, which has been reported on in the knees as well as the feet and ankles and should in the future become the gold standard for the evaluation of degenerative joint disease.

4.2. Literature Analysis

In light of the present systematic review, the most important fact to report is that segmentation is key to enabling the development of innovative techniques from WBCT datasets. Moreover, if these innovations are to be translated into clinically applicable solutions, it is paramount for fast and reproducible automatic segmentation methods to be available so that interpretation times remain possible within the clinical workflow. Furthermore, the push towards standardization of segmentation and measurement methods [15,60] initiated by scientific societies such as the International WBCT Society, the Orthopedic Research Society and the International Society of Biomechanics is of utmost importance if a disseminated acceptance of these innovative techniques and clinical solutions is to be achieved.

4.2.1. Description of Techniques for Image Acquisition

Within the WBCT workflow, techniques for image acquisition are used to dynamically position the patient under physiological load during acquisition to improve detection of designated pathologies. The most investigated one, syndesmotic instability [17,79,80], has cadaveric and clinical studies showing that, due to the configuration of the distal tibiofibular joint and the injury mechanism, external torque potentializes the diagnostic capabilities of WBCT. Without this stress test and if no post-processing is applied, WBCT does not appear to be superior to MDCT in diagnostic terms, although it already is in terms of radiation and time spent in the workflow. This occurs only if manual, one-dimensional measurements are taken. However, post-processing techniques such as distance mapping of the syndesmosis and surface or volumetric measurements described below have re-established this superiority.

The Coleman block test Is reported In a level IV study Investigating hindfoot alignment in cavovarus feet using clinical examination, radiographic views assessing the hindfoot angle and WBCT assessment of the foot–ankle offset [19], showing correlation between the 3 modalities.

In another level V report [43], the use of shod auto-varus is reported in the diagnosis of lateral ankle instability and suggests the use of standardized jigs to induce a normalized amount of varus or valgus depending on the target diagnosis (Figure 5).

Techniques for image acquisition also included static methods for acquiring more data. In a level III study, Richter et al. used a built-in pedography sensor [44] to assess the position of the center of pressure and compared it to the result of the foot–ankle offset (as anatomical foot center) in 90 patients (180 feet). They found an average distance between the two centers at 28.7 mm, being the anatomical foot center distal to the center of pressure in 175 feet and lateral to it in 112 feet. No significant medio-lateral differences were found. It could be anticipated that more sensors investigating weight or bone density could be used in the future to increase semeiology during WBCT acquisition. This could be useful in assessing bone health or pressure points in diabetic patients, for example, or more simply to standardize positioning of patients within the WBCT machine during acquisition. This issue has been taken into account by the industry, using specific gentry to standardize positioning, especially in the knee, but the reproducibility seems to remain questionable, as has been reported recently by a group of radiology researchers [15].

4.2.2. Description of Advanced Computerized Techniques for Image Processing

Pre-Processing

Pre-processing techniques involve the raw file that contains the patient dataset before it is processed to obtain interpretable multiplanar images. WBCT is a computerized tomography technique, meaning that the raw file is treated like a 3D stack of 2D slices. Unlike MDCT, in which it is the case because the acquisition requires ‘slicing’ up the anatomy with a ‘flat’, 0.5–0.8 mm-thick fan-shaped X-ray beam, WBCT acquisition results in a single raw file considered isotropic (the image definition is the same in all dimensions of space), due to the cone-shaped beam. Pre-processing has the single raw file as input and outputs a stack of 2D 0.2–0.4 mm slices in the form of DICOM files.

Other examples of pre-processing steps are metal artifact and movement reduction algorithms. They are based on highly specialized mathematical algorithms with the aim of improving the general quality of images and thus diagnostics.

Post-Processing

• Pre-Segmentation
• Reconstruction

The initial post-processing is performed by the WBCT device manufacturer’s software (https://curvebeamai.com/products/cubevue-software/download-cubevue/, accessed on 5 January 2024) to obtain interpretable multiplanar images, usually presented in multiplanar reconstruction format, with three viewing windows corresponding to the three planes of space (two vertical planes, coronal and sagittal; and one horizontal, axial plane) sometimes including a fourth window with a 3D volume rendering view (Figure 6).

After this process, advanced post-processing techniques are subsequently applied to provide recent innovations.

3.

Manual Measurements

At this stage, classical measurements can be made manually, despite, as described earlier, inducing ‘slice bias’. Innovative solutions for manual measurements include the measurement of middle facet uncoverage in progressive collapsing foot deformities and the measurement of joint space width in ankle osteoarthritis using variations in contrast (measured in Hounsfield units (Hus)) [25,45].

4.

Semi-Automatic 3D Biometrics

The most reported post-processing, pre-segmentation semi-automatic tool is Cubeview Talas^® (Curvebeam AI, Hatfield, PA, USA), which automatically gives the foot–ankle offset (FAO), a 3D biometric hindfoot alignment measurement, after manual identification of four anatomical landmarks: the weight-bearing points of M1 and M5, the calcaneus bones and the center of the ankle joint. Its result is given as a percentage offset of the foot length, which corresponds to the coronal offset between the center of the ankle joint and the bisector of the forefoot passing through the calcaneus weight-bearing point (Figure 7).

The FAO has proven to be an effective hindfoot alignment measurement [48], and from it originated the concept of 3D biometrics, in which a minimum of four points are required to obtain volumetric (3D) rather than surface, angular or linear measurements. Its intra- and interobserver reliability is improved compared to conventional 2D measurements [47,48] by levels nearing 100%, and its correlation with pathologies such as chronic lateral ankle instability [47] and PCFD has proven excellent. This is thought to be because the FAO takes into account the torsional effect of the forefoot on the hindfoot, unlike traditional measurements which only look at the alignment of the tibia versus the hindfoot. It may therefore be more sensitive in its ability to correlate with specific multidimensional pathologies like predicting the need for realignment osteotomies in total ankle replacement [81] or the risk of periprosthetic cysts [82]. However, it remains a measurement of hindfoot alignment and does not assess sub-level type deformities: it is not a diagnostic tool. Other techniques based on the traditional 2DXR-MDCT literature remain for now the only tools to allow for this [14,83].

5.

Segmentation

We identified automatic segmentation techniques as the basis of other recent innovations [84,85]. It appears indispensable because the time required to perform manual segmentation is too much for clinical applications, confining potential innovations to the research area. The important word in ‘automatic segmentation’ is therefore ‘segmentation’, but this also turns out to be the most complicated to achieve.

Segmentation is the process by which individual bones are outlined and labelled. When segmentation is carried out manually, every single bone (28 in a single foot) must be manually outlined on every slice (up to 1000 per dataset), which amounts to 28,000 operations per foot, or 46,000 per bilateral scan. In practice, it generally takes 30 min to segment the hindfoot bones and 2 h to segment the whole foot and ankle [86]. It is therefore paramount that fast and reproducible automatic segmentation software become widely available. In that case, studies report that the same process can take a little as a few seconds [53]. The present review found four reported techniques or software solutions. These initially used automatic contouring based on contrast definition, evolving to more advanced techniques, ultimately aided by proprietary solutions. The two most reported in the literature are Bonelogics^® (Disior Oy, then Paragon28) and Cubeview^® (Curveabeam AI). Although much of their inner work remains undisclosed, the industry advertises that they rely on a combination of algorithms and artificial intelligence (AI) (i.e., to automatically identify bones, to reproduce surgical procedures such as osteotomies or arthroplasties and to run automatic measurements in order to anticipate which gestures are needed to obtain the desired alignment). Throughout the process, it is reported that some of the reasons for the failure of the automatic segmentation are the presence of metalwork; poor bone quality or the presence of arthritic joints leading to areas of contact between bones; and the misinterpretation of two touching bones as a single structure [52]. One important step in the segmentation process is the smoothing of bone contours. This is the equivalent of noise reduction in digital photography: smoothing reduces the risk of two touching bones being misinterpreted as a single one but results in a (marginal) reduction of the quantity of data. The AI training process helps in dealing with the aforementioned contouring issues as well as learning to correctly label the bones. It makes more sense from a practical and industrial standpoint to build a system that can label the anatomy and then produce all relevant measurements, rather than build multiple systems to perform each measurement. It transpires that large amounts of data are necessary to efficiently train these Ais, which will need to be scientifically evaluated by independent entities on a regular basis until satisfactory performance is reached. There is unequivocal agreement on this amongst authors and stakeholders [15,52,53].

Post-Segmentation

Once segmentation is achieved, a digital clone of the foot and ankle is created with all bones correctly labelled. Therefore, any measurement may be performed automatically, since all the spatial coordinates of all the voxels (the 3D equivalent of 2D pixels) are known, with all of them having been grouped into separated voxel clouds corresponding to the individual bones. Using these data and stepping up from classical measurements, new tools have been reported on in the literature to improve visualization and our comprehension of pathological processes. We present these below:

• Absolute and Relative 3D Bone Measurement Reporting

Theoretically, an infinite number of measurements are possible, hence the diversity of measurements described in this area of the literature [16,21,22,26,52,53,54,55,56,57,58,59,60,61,62,63,64]. In more practical terms, considering there are on average 28 bones in the foot and ankle, each with a center of mass that has three spatial coordinates and three inertial vectors each, with each having three spatial coordinates, each bone may be described by a set of 12 ‘absolute’ coordinates, which are calculated within the WBCT frame of reference (Figure 8).

This frame is referenced by the floor plane for the two vertical (sagittal and coronal) planes. There is no consensus on how to set the rotation in the axial plane, but the most commonly used is traditionally the second metatarsal or the bisector of the forefoot in 3D biometrics papers [48]. However, the possibility of relative measurements is remote, as any of the 12 absolute coordinates in each of the 28 bones can be described relative to any of the 12 absolute coordinates of the 27 other bones, which equates to 12²⁷ possibilities. This huge number explains why there has always been a quest to discover new measurements in the realm of musculoskeletal research. Things were easier with 2D radiography, which results in dimensional reduction, thus reducing the possible number of measurements, albeit with a non-negligible loss of information. After the introduction of WBCT, a need has arisen to find and promote a standardized methodology for describing bone orientations in space, which is among the tasks of the International WBCT society mentioned above. However, the industry has anticipated this and is already proposing different off-the-shelf software solutions to provide systematic reporting of bone and joint angles, based on the historical literature, including such widely used angles as the M1-M2 angle for hallux valgus assessment, or the hindfoot alignment angle for deformity assessment. However, even these may be defined differently depending on the software provider, hence the importance and urgency of an international consensus on this matter so that reports may be comparable across all software platforms. There are many reasons for these differences, not the least being the absence of consensus in the existing literature. Other reasons pertain to computerized techniques used to obtain absolute measurements [60].

The 3 main techniques reported are as follows [87].

1—Principal component analysis (PCA) determines the center of mass and the three principal components or inertial axes of the bones through averaging the relative contributions of each spatial dimension, a rather common mathematical tool, but still complex to the layperson. Its advantage is that it can easily be made fully automatic. Its disadvantage is that it provides slightly different results depending on the volumetric shape of considered bones, which can be otherwise interpreted as increased variability in measuring anatomical axes depending on surface landmarks.

2—Statistical shape modelling averages a large number of real-life examples from pre-existing datasets of bones to create a library to which a given bone may then be compared. Once the bone has been recognized, its orientation can be derived from that of the library example. This technique is usually part of the segmentation process itself. Its disadvantages are that it depends on the existence of a large enough patient dataset library and it cannot consider a situation which is not already known within the library.

3—Fitting of geometric primitives averages the different parts of bones by fitting the closest 3D geometrical figure. For example, a long bone diaphysis may be approximated to a cylinder and its base and head to a truncated cone and a barrel shape, respectively. The orientation of the bone is then derived from the known geometry of the fitted primitives. The main disadvantage is that, unlike with a long bone such as a metatarsal, it is more difficult to apply this method to bones with a more complex shape such as the talus or calcaneus.

Depending on the proprietary mix of technologies used to obtain the 3D orientation of each bone, software can provide an automated report of chosen measurements, usually a set of basic metrics (such as the M1–M2 angle, sesamoid rotation angle (SRA) for the forefoot, sagittal and axial talus–M1 angle and hindfoot alignment angle or foot–ankle offset) and a customized set of measurements chosen by the user. Reports in the literature regarding the usefulness and efficiency of these systems agree on their celerity, resulting in important time gains compared to measurements by hand [52,53]. However, they also report that they differ from the latter, raising the question of which is the gold standard [53]. They also report a relevant number of failures or aberrant measurements, which are thought to be due to cases of low bone density, where the density of soft tissues is close to that of the bone, which can ‘blur’ the picture for the software, or the presence of metalwork or traumatic sequala, which are in any case unusual situations which can be misinterpreted by the software. Another explanation may be insufficient training when deep learning-based Ais are used. However, there is generally positive feedback regarding these new generation tools, with good reliability and excellent reproducibility [52,53,55,56,88].

2.

Distance Mapping or 3D-JSW, surface and volume measurements

This is the advanced computerized method which has been the most reported on in the literature, to the best of our knowledge. This tool is based on analyzing the surface-to-surface distance map within any given weight-bearing joint. The sum of all these point-by-point distances adds up to a 3D joint space width map, which explains why it has been dubbed ‘distance mapping’ [66,67,69], ‘3D joint space width’ [65,68,89,90] or a mix of both in the literature. Originally developed as way to visualize regions of bony approximation, authors presented ways to also use it as a quantification tool for such conditions as PCFD or cavovarus deformity, by averaging the distances in reproducible in pre-defined anatomical quadrants within the joints of interest (Figure 9).

Other than classical usage of this tool to quantify the evolution of osteoarthritic joints, authors have reported its ability to detect minute or more specific changes in the onset of hindfoot deformity, in PCFD [66] or cavovarus [67] configurations. This may lead in the future to early detection and preventive intervention in a wide range of foot and ankle pathologies. It is important to take into account that this technique does not enable the visualization of cartilage itself, but only the space in which the cartilage is to be found. It will therefore correctly depict an absence of cartilage, but not an area where cartilage is augmented, and an area where the joint is distracted may incorrectly be considered as cartilage augmentation. However, weight-bearing capacity largely eliminates these downsides, which in any event are also true of conventional 2DXR, while the analysis is not possible at all with conventional non-weight-bearing MDCT.

3.

Coverage Mapping

Coverage mapping has been investigated mostly in PCFD, to evaluate peritalar subluxation. Peritalar subluxation is thought to be the onset of PCFD pathology, while being extremely difficult to define and describe. Sangeorzean at al. [91,92] described it with the production of images showing overlap of the posterior subtalar facet and de Cesar Netto et al. have since then produced similar findings in the subtalar posterior and middle facets and reported a clinically significant correlation of coverage maps with patient reported outcome measures [70,93]. The technique involves digital reconstruction of the articular facets’ borders. This concept in itself may be subject to variability depending on the computerized algorithm used because choices have to be made regarding multiple settings, such as where to define the limit between cartilage and bone when the cartilage cannot be seen. However, it is definitely a step forward in terms of diagnosis, which highlights again the importance of the standardization work [23] initiated by the international WBCT society.

4.

Surface and Volumetric Measurements

As WBCT rose in the mid-2010s, it became apparent that traditional 2D measurements were not fit for the 3D environment, generating, as mentioned above, a new kind of so-called ‘slice’ bias. Also, performing a 2D measurement in a 3D environment is equivalent to a dimensional reduction, thus reducing the quantity of information available. New solutions had to be found, therefore, to make the best of the new technology. The syndesmotic distal tibiofibular joint is the best example of this evolution. In that case, single-dimensional measurements (distances like the medial clear space or angles) have often fallen short of demonstrating reliable diagnostic capabilities and must often be complemented by other modalities such as MRI [94]. Similarly to how the foot–ankle offset describes the use of four landmarks to redefine hindfoot alignment as a volume, authors have described first surface, then volumetric evaluation of the clear space between the distal tibia and fibula to assess syndesmotic instability [72,73,74,95,96]. The latter has offered unequaled diagnostic power, confirming that the 3D environment contains more diagnostic information which should be investigated where possible (i.e., when segmentation has been applied) using 3D measurements [97].

5.

Centre of Rotation Assessment.

In a paper in Scientific Reports, Pena-Fernandez, Goldberg et al. used a method called digital volume correlation (DVC) to identify the center of rotation of the subtalar joint in a series of healthy volunteers. The subjects were placed in a PedCAT bilateral WBCT device (CurvebeamAI, Hatfield, PA, USA) and asked to perform inversion and eversion of their subtalar joints. The authors report that the center of rotation of the subtalar joint was consistently found in the middle of the subtalar joint [75]. The principles of DVC are to reference one of the implicated bones and to measure the displacement using a vector field associated to the other bone, between the two considered positions.

4.2.3. Description of Techniques for Advanced Clinical Applications Derived from WBCT Computerized Techniques

As mentioned before, segmentation is the key to identifying individual bone contours, and once that has been achieved, multiple new techniques can be applied. One of these is the externalized manufacturing of custom 3D-printed surgical guides for osteotomy [18] or total joint replacement [77], in this particular case ankle replacement. However, it is not clear, in the absence of dedicated literature, whether automatic segmentation software would be reliable enough to provide fully automatic surgical guide manufacturing. Indeed, human intervention by biomechanical engineers is still required to ensure the quality of the fit, according to the patient’s specific bone surface characteristics and the quality of the alignment correction on the three planes of space. Therefore, whoever provides the segmentation software, implant manufacturers still must provide human resources to secure these steps, hence an increased cost to include customized bone cut guides, which may not be accessible to all patients, surgeons and healthcare systems. However, in the future, solutions should be found for widespread adoption, taking into account the multiple advantages of customized surgery, such as specific planning of the correction on natural stance WBCT datasets, faster learning curves, operative times and improved accuracy and reliability of axial corrections.

Although customized surgical guides existed before WBCT was implemented, it is important to distinguish what the added value of WBCT is in this case. Here, the segmentation and surface matching of the guide does not require weight-bearing; it is the planning for alignment correction that does. In that case, as mentioned earlier, the rapid acquisition of natural stance 3D-WBCT is definitely an improvement as compared to the traditional 2DXR-MDCT sequence. One often-heard argument here is that the correction is performed surgically on a patient who is lying down, so what is the point of planning on standing images? It is important to debunk this argument: first, planning is carried out standing because the alignment only makes sense standing. Also, even in the traditional sequence, the planning of corrections is made using standing 2D images (for total ankles, anterior–posterior and lateral 2DXR, including the knee), which are less reliable than 3D-WBCT. In that sequence, it is only the surface fitting of the guide which is performed using the MDCT: the cut itself and the corrections are planned using, again, the alignment measurement setup, in the former case, 2DXR.

Similarly, we anticipate that WBCT data will soon be implemented for use in robotic orthopedic surgery, although, to the best of our knowledge, no record of such a procedure has been recorded in the literature at the time this work is being written.

4.2.4. Current Limitations of the Technique

The cost of WBCT devices has often been discussed as potential limiting factor to the spread of the technology in clinical centers. While no cost-effectiveness studies have been published so far, in a population-based study led by Jacques T. et al. in 2021, two periods were compared in emergency setting: a 7-month period during which only a standard multi-detector CT was available and, one year later, an equivalent 7-month period during which a CBCT was also used [32]. The authors found a significant radiation dose and an accelerated turnover (23.6% faster) with CBCT in place. Based on this, and taking into account the need to reduce waiting lists in public hospitals in many Western countries, it could be hypothesized that the initial economical effort to buy the machine would soon be compensated for by the gain in terms of diagnostic workflow, as already discussed above.

On a different note, whether the management of WBCT devices should compete with radiologists (as it should be, given the nature of the device) or with orthopedic surgeons (which it does not, according to the literature discussed above) who, so far, have shown the greatest interest towards the technology given the clinical advantages in diagnosis and, even more, surgical three-dimensional planning, has also been a matter of debate. While, in an ideal setting, both specialties should collaborate and move in the same direction, the fact that this does not always happen and the lack of agreement felt by clinicians in daily practice is advocated as a further limitation to the acquisition and use of WBCT devices.

5. Conclusions

In conclusion, we have found that advanced computerized techniques developed using WBCT in the foot and ankle field can be classified. The most reported currently are standardized absolute and relative 3D relationship and joint distance mapping. Focus has to be made on the development of fast, reliable automatic segmentation software and international academic endeavors to scientifically validate such software. It is important to note that the techniques developed in the foot and ankle will be applicable to other fields, in particular the knees, hips, spine and shoulders and specialties: trauma and emergencies, pediatrics, sports medicine and rheumatology. We anticipate that, beyond measurements, further developments based on deep learning and artificial intelligence will lead to breakthroughs in diagnostics and prognostics.

However, as exciting as these perspectives might be, they remain somewhat distant, and therefore the advanced computer science should not overshadow the immediate clinical advantages of diminished radiation dose, improved diagnostics and slashed patient workflow delays observed with WBCT, which should make practitioners and researchers strive to realize its swift and widespread implementation in the musculoskeletal clinical realm.