Introduction

Postural defects are among the most commonly diagnosed health problems of the developmental age. Determining the scale of the phenomenon is difficult, however: estimates depend strongly on the criteria adopted and the assessment method, and published rates vary widely, from a dozen or so to several dozen percent of the school population. The consequences of uncorrected asymmetries go beyond aesthetics. They can affect gait biomechanics, load distribution in the joints of the lower limbs, and the severity of musculoskeletal complaints later in life. Traditional postural assessment, although well established in practice, has substantial limitations. The result of visual and palpatory examination depends on the examiner’s experience, and agreement between assessments made by different specialists can be low. Partly to improve reproducibility, software tools for postural analysis have been developed, such as the PAS/SAPO system, for which satisfactory measurement reliability has been demonstrated [1]. Reference methods, in turn, such as surface topography or standing radiographs, are either hard to access outside specialist centres or involve exposure to ionising radiation, which raises legitimate ethical concerns in a growing child.

These abnormalities are clinically relevant to school and community nurses. Organised school screening detects scoliosis and related deviations early and reduces the number of advanced cases that reach treatment; its coverage depends closely on whether a nurse is on staff and whether a defined referral pathway exists [2, 3]. A review addressed to a nurse-practitioner readership sets out what the clinician should recognise and when to refer [4]. The entry of artificial intelligence into nursing has so far produced mainly conditional benefits, dependent on validated tools and clear governance [5], which is the cautious position adopted throughout this paper.

Against this background, the development of real-time body-pose estimation systems opens a new avenue for non-invasive and economically accessible postural assessment [6, 7]. Because such models can run on widely available hardware, their use is no longer confined to motion laboratories; it can enter office practice and, potentially, school-based postural screening as well.

The aim of this paper is to present the principles of the RTMPose (Real-Time Multi-person Pose Estimation) model and to critically assess its usefulness and limitations in screening postural assessment of children and adolescents, with particular attention to the role of the school and community nurse. The discussion is illustrated by the analysis of a single case.

RTMPose: technical assumptions, capabilities, and limitations

RTMPose is a deep-learning model developed within the MMPose library (OpenMMLab) and released in 2023 [6]. It belongs to the top-down family of architectures and combines high precision in detecting key points with low latency, so that analysis can run in real time even on devices with limited computing power.

The features of the model relevant to clinical applications include: detection of 17 to 26 anatomical points (up to 133 in the COCO-WholeBody variant), covering mainly the limb joints and the approximate positions of the shoulder and pelvic girdles; a favourable trade-off between accuracy and speed (the RTMPose-m variant reaches an average precision (AP) of about 75.8% on the COCO dataset at more than 90 frames per second) [6]; and the possibility of integration with clinical applications thanks to support for the ONNX and TensorRT standards. The analysis is based on an ordinary colour (RGB) image, without markers attached to the body, which simplifies the examination and makes it less burdensome for the child [8, 9].

The limitations of the method should be stated clearly at the outset. RTMPose locates two-dimensional skeleton points on the image plane. The standard point sets (COCO, COCO-WholeBody) do not include the bony landmarks of the pelvis, such as the anterior and posterior superior iliac spines (ASIS, PSIS), nor the course of the spinal curvatures in the sagittal plane; reliable assessment of these structures therefore requires dedicated models and separate validation. Single-camera analysis is also sensitive to perspective, to the patient’s position relative to the lens, and to out-of-plane rotations. For this reason, the angular values obtained should be treated as approximate screening indicators rather than as measurements equivalent to reference methods [10, 11].

Postural asymmetry: parameters assessed by image-based systems

Postural asymmetry comprises deviations from body symmetry in the frontal, sagittal, and transverse planes [12]. In practice, the parameters assessed are primarily: the position of the head relative to the midline, shoulder height, the symmetry of the waist triangles, pelvic position, the alignment of the knees and feet, and the spinal curvatures in the sagittal plane.

Pose-estimation systems make it possible to determine reference lines automatically and to compute angles for those parameters that lie in the image plane, for example the tilt of the eye and shoulder lines or the alignment of the lower limbs in the anterior and posterior projections [8, 9]. This gives a quantitative, reproducible record of deviations that visual assessment alone does not capture. It must be remembered, however, that the accuracy of this record depends on image quality and on the standardisation of conditions, and that the sagittal parameters and the position of the bony pelvic landmarks remain, for the reasons described above, less reliable when assessed from a single 2D image.

Clinical illustration: postural analysis in three projections

To illustrate how the tool works, a single case of a school-aged girl was analysed using a system based on the RTMPose model. Images were taken in three standard projections: anterior (AP), lateral (L), and posterior (PA). The girl stood in a relaxed standing position, in her underwear, in a room with even lighting, about 2.5 m from the camera; these conditions limited perspective distortion. On the recorded images, the system placed key points and reference lines and then determined angular values for the individual segments.

Anterior projection

In the anterior projection (Fig. 1), what stands out is the asymmetry of the upper body: an eye-line tilt of 4.57° and a shoulder-line tilt of 3.58°, with a symmetrical pelvic position (0.05°) and knee alignment at the adopted threshold (2.86°) (Table 1). This distribution of deviations suggests that the source of the problem should be sought in the shoulder girdle and the cervical segment rather than in the pelvic structures or lower limbs.

Lateral projection

In the lateral projection (Fig. 2), the value of 5.93° for the knee joint stands out, slightly exceeding the adopted threshold for physiological knee hyperextension (Table 2). The remaining parameters, namely forward head posture and trunk inclination, fall within the adopted range.

Posterior projection

The posterior projection (Fig. 3) revealed a lower-limb alignment deviation of 5.46° with a symmetrical position of the pelvis, shoulders, and ear line (Table 3). This localises the problem below the pelvis, in the knee joints or in foot alignment, rather than centrally. The corresponding anterior-projection value was 2.86°; such differences between projections illustrate the sensitivity of single-camera 2D analysis to perspective and rotation.

Discussion

Early postural assessment in nursing practice

The analysis presented here is a reminder of the value of an accurate, global postural assessment as an element that, although traditionally assigned to physiotherapy, also belongs within the whole-person nursing assessment of the child. The school or community nurse is often the first and sometimes the only health-care professional with whom a pupil has regular contact over many years of schooling, which makes the nurse a natural link in the early detection of musculoskeletal abnormalities.

The image from the anterior projection, with its clear shoulder asymmetry against a symmetrical pelvis, has a concrete bearing on educational work. A nurse who has numerical data can show the family exactly where the problem sits, mainly in the upper spine and the shoulder girdle, and a measured angle gives parents something concrete to act on. That specificity helps them follow the recommendations.

The borderline knee hyperextension (5.93° in the lateral projection) is also notable. In a child during a period of intensive growth, such a position is often transient; combined with prolonged sitting and weakness of the muscles that stabilise the joint, however, it may predispose to overload of the anterior knee compartment. This relationship is biomechanically plausible, although data for the developmental population are limited, so it justifies observation and ergonomic education rather than immediate intervention.

The lower-limb alignment deviation in the posterior projection (5.46°) with a symmetrical pelvis localises the problem peripherally. Deviations of this kind, if they persist for years, may gradually load the hip joints and the lumbar segment, and their clinical consequences not infrequently become apparent only in adulthood. In such a case, early recognition and prompt referral to a physiotherapist or orthopaedist is an important contribution by the nurse to the patient’s long-term health.

Objectifying clinical observations

For preventive care of children and adolescents, the main advantage of image-based tools is that they replace a subjective impression with a measurable parameter. Descriptions such as “one shoulder higher” have been hard to document and to compare over time. Angular values entered into the records create an objective reference point that can be revisited at subsequent visits, and an examination repeated every 6 to 12 months makes it possible to judge whether the recommended exercises and ergonomic correction are working.

The visualisation of the results, with key points and angular values marked on the image, may also make conversations with parents easier and support adherence to recommendations. This is a plausible hypothesis, but one that needs confirmation in studies involving parents; in the present paper we treat it as a direction for further analysis rather than as an established fact.

In systematic reviews, markerless systems show, in many applications, accuracy close to that of reference methods, at considerably lower cost and without the need to prepare the patient, although with substantial variability depending on the parameter assessed and the measurement conditions [8, 9]. In practical terms, this means that implementing such a tool requires neither expensive equipment nor lengthy training, provided the procedure is reliably standardised. These estimates can be qualified in practical terms. The hardware requirement is limited to a standard RGB camera and a computer of the kind already available in most school health offices and outpatient rooms, so the additional equipment outlay is minimal. The analytical software is offered on a subscription basis, at a recurring cost that is an order of magnitude lower than the capital outlay for a professional marker-based laboratory; precise figures depend on vendor and licence. Operator training is correspondingly brief, measured in hours rather than days, compared with the considerably longer instruction and repeated calibration required by advanced marker-based systems.

The nurse’s role in screening the school population

The prevalence of postural defects justifies treating screening postural assessment as an important public-health task. In everyday school practice, however, this assessment is often omitted or reduced to a cursory observation, owing to time pressure and the lack of simple, standardised tools. Image-analysis systems can partly fill this gap, because they require only a standard-resolution camera and a computer with appropriate software. Deep-learning analysis of back photographs has been developed and validated for scoliosis screening [13], and a study based on a single smartphone photograph confirmed the usefulness of such screening in adolescents compared with clinical examination [14]. In a review of artificial intelligence in orthopaedics, Trofa et al. [15] note that the progressive miniaturisation of motion-analysis systems favours their wider use, including in school medicine.

Safe implementation, however, requires a standard examination procedure that specifies the measurement conditions (distance, lighting, and the position and clothing of the person examined), the criteria for referral for further diagnostics, and the rules for protecting the pupil’s personal data and image in accordance with the General Data Protection Regulation (GDPR). Meeting these requirements assumes cooperation between nurses, physiotherapists, medical-informatics specialists, and school management. Without such an organisational framework, even an efficient tool cannot be used responsibly.

Limitations and directions for further research

The results presented here should be interpreted with caution. The analysis was observational and concerned a single case, which precludes statistical inference and generalisation to the population; the case serves here only an illustrative function. Despite the high accuracy of RTMPose under laboratory conditions, its validation under real screening conditions, taking into account variation in lighting, diversity of clothing, and the dynamics of group examination, requires further prospective studies [10, 14].

A serious practical limitation is the absence of Polish normative data for the postural parameters assessed with this method in school-aged children. For this reason, the thresholds adopted in Tables 1-3 are operational rather than normative. Developing norms that take into account sex, age, and stage of maturation is a precondition for the routine use of the system in screening. Czaprowski et al. [12] remind us that the interpretation of postural deviations in children must account for the physiological variability of posture across successive growth phases, which further complicates normalisation.

From a nursing perspective, intervention studies would be especially valuable, assessing the extent to which the use of image-based tools in communication with parents translates into lasting family health behaviours: adherence to ergonomic recommendations, regularity of exercise, and timeliness of follow-up visits.

Implications for nursing practice

In the school setting, the nurse of the teaching and care environment is usually the only health professional who sees a whole cohort year after year, so postural screening falls naturally within the nurse’s remit. The literature reviewed here translates that position into specific, evidence-based tasks.

The evidence regarding school screening itself should be considered first. Organised school programmes detect scoliosis and related deviations early and reduce the number of advanced cases that reach treatment, with fewer radiographs and lower cost when a structured, two-step procedure is used [2, 16].

Whether screening happens at all is closely associated with the presence of a nurse. In a survey of 291 schools, programmes were considerably more common when a nurse was employed, and the single biggest barrier to screening was the absence of a referral pathway for a positive result [3]. These findings suggest that establishing school nursing provision and a clearly defined referral pathway may represent key conditions for effective screening implementation.

The image-based tool fits this role because its accuracy is approaching the level required for screening use. A convolutional neural network (CNN)-based markerless system matched marker-based 3D motion capture on most gait parameters in recent validation work [17], and deep-learning screening for scoliosis from photographs has performed well in adolescents [13, 14]; confirmation under real school conditions is still pending.

Read against the present case, the evidence translates into concrete actions for the school nurse:

1. Build a quantitative posture check into the periodic health examination. Rather than a free-text note such as “one shoulder higher”, the nurse records the angular values the system produces, which gives a reproducible baseline for the next visit, by analogy with follow-up documented in young adults [18].

2. Write the referral pathway first. Following the logic of scoliosis screening, the nurse refers the child to a physiotherapist or orthopaedist when measured values cross the pre-agreed operational thresholds (Tables 1-3) or when a scoliometer reading exceeds the accepted cut-off [2, 4, 16]. Without that route in place, screening stalls [3].

3. Use the annotated image to educate parents. The marked key points and angles let the nurse show the family which segment is affected and why follow-up matters, which may support adherence; this benefit still needs testing in families [13, 14].

4. Track change over time. A screen repeated every 6 to 12 months tells the nurse whether ergonomic advice and exercise are working and flags progression early enough for timely referral [2, 18].

5. Treat the tool as one part of a governed pathway. Safe use assumes a written examination protocol (distance, lighting, position, clothing), data protection rules under the GDPR (Regulation (EU) 2016/679), structured training for the nurse, and attention to the child’s privacy [19-21].

Where artificial intelligence enters nursing, the benefits reported so far are mostly potential, conditional on validated tools and clear governance [5]. That caution sits comfortably with the screening role: the nurse gains a measurable, repeatable baseline and a documented first point of contact in a referral chain, which is where the reviewed evidence enters everyday school practice.

Conclusions

Pose-estimation systems make it possible to obtain quantitative, reproducible postural indicators in real time, without expensive marker equipment, which makes them potentially accessible in school and outpatient settings.

The values obtained from a single 2D image should be treated as approximate screening indicators; assessment of the sagittal curvatures and the bony pelvic landmarks requires dedicated models and separate validation.

Visualisation of the results may make communication between the nurse and parents easier and support health education, but this benefit should be verified in studies involving families.

Implementing the tool in schools requires a standard examination procedure, criteria for referral for further diagnostics, and protection of the pupils’ personal data and image in accordance with the GDPR.

Further validation studies in the paediatric population are needed, together with the development of Polish postural norms for school-aged children.

Disclosures

This research received no external funding.

Institutional review board statement: Not applicable. The informed consent of the child’s parent was obtained.

The authors declare no conflict of interest.

References