Telemedicine system for children's bone age assessment and growth monitoring

Abstract

This invention discloses a remote medical system for children's bone age assessment and growth monitoring, relating to the field of medical diagnostic technology. The system includes: user-end data acquisition compatible with home / hospital equipment and sensors, automatic collection reminders, and transmission identification; image preprocessing with filtering and enhancement; AI-based bone age assessment based on an age-specific model, switching between dual standards, outputting results and anomaly markers; growth analysis comparing with standard curves; a remote interactive encrypted platform, pushing results for annotation and diagnosis; data storage with double encryption, sorted by timeline, supporting retrieval and authentication; anomaly warning with three threshold levels, triggering multi-channel alerts and automatic appointment scheduling; human-computer interaction via touchscreen or APP, providing an entry point, with key operations requiring verification. This invention achieves convenient remote data acquisition and accurate bone age assessment, scientifically monitors growth trends and warns of anomalies, strengthens cross-institutional collaboration and personalized intervention, ensures data security, improves management efficiency and intervention effectiveness, and adapts to the needs of children throughout their growth cycle.

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, and in particular to a telemedicine system for assessing bone age and monitoring growth in children. Background Technology

[0002] Childhood bone age assessment and growth monitoring are core components of pediatric endocrinology diagnosis and treatment, directly impacting early intervention and prognosis for developmental disorders such as growth retardation and precocious puberty. Current clinical practice reveals significant shortcomings in the convenience, accuracy, and synergy of traditional methods, making it difficult to meet the demands of parents for refined management of their children's growth and development.

[0003] Offline assessments suffer from significant limitations in time and space, as well as uneven resource distribution. Bone age assessment relies on hospital X-ray equipment to capture wrist bone images, which are then manually determined by a pediatric endocrinologist according to TW3 or CHN standards. The entire process requires parents to take their children to and from the hospital, which is time-consuming, labor-intensive, and easily affected by local medical resources. Primary healthcare institutions generally lack professional assessment physicians, and most patients need to be referred to tertiary hospitals, leading to prolonged assessment cycles and missed opportunities for optimal intervention. Although home X-ray equipment is becoming increasingly common, the images often cannot be directly used for assessment due to insufficient clarity or positional errors, requiring repeated imaging and further increasing operational costs.

[0004] The technological limitations of AI assessment and growth monitoring urgently need to be overcome. Existing AI bone age assessment models are mostly trained based on a single standard, failing to fully consider age-group and gender differences in children. Assessment accuracy is significantly affected by image quality, and the recognition rate of abnormal images is low. Growth monitoring is limited to simple recording of parameters such as height and weight, lacking dynamic matching analysis between bone age and chronological age. Furthermore, it cannot combine genetic factors, growth rate, and other multi-dimensional data to predict adult height, making it difficult to provide early warnings of developmental abnormalities. Data storage is scattered across hospital systems and manual records by parents, forming "information silos." Physicians cannot obtain continuous growth time-series data, and intervention plans lack comprehensive basis.

[0005] The lack of remote collaboration and closed-loop intervention exacerbates management challenges. Data interfaces between different medical institutions are incompatible; assessment reports and growth data between community health centers and tertiary hospitals cannot be shared in real time. Cross-institutional consultations require manual transmission of images and records, which is inefficient and prone to errors. The intervention process lacks dynamic tracking; nutritional recommendations are mostly generic templates, failing to be personalized based on children's dietary preferences and activity levels, resulting in low parental adherence to intervention plans. Furthermore, children's growth data contains a large amount of sensitive information; existing systems have simple encryption mechanisms, making them susceptible to privacy leaks and failing to meet medical data security standards. In addition, the lack of regular self-checking and calibration mechanisms for equipment means that long-term data collection errors directly affect the reliability of assessment results. These combined problems result in a situation where children's growth and development management is characterized by "difficult assessment, scattered monitoring, weak collaboration, and blind intervention," hindering the development of precise children's health management. Summary of the Invention

[0006] The present invention proposes a telemedicine system for children's bone age assessment and growth monitoring to solve the problems mentioned in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a remote medical system for children's bone age assessment and growth monitoring, comprising: The user-end data acquisition module is used to acquire children's bone age images and physiological growth parameters. It is compatible with home portable X-ray imaging equipment and hospital PACS systems, transmitting bone age images via the DICOM 3.0 protocol. The device resolution is ≥2048×2048, and the exposure dose is ≤0.1mSv. It integrates a smart height and weight scale, a body fat percentage sensor, and a heart rate monitoring module. It automatically reminds users to collect data every 3 months, and supports manual triggering in emergencies. Data includes a timestamp, device ID, and a unique child identifier.

[0008] The image preprocessing module optimizes bone age images and extracts key regions. Adaptive median filtering removes noise, and the Retinex enhancement algorithm improves the contrast between the epiphysis and metaphysis. Based on the U-Net segmentation model, the epiphyseal regions of the radius, ulna, and metacarpophalangeal bones are located, and edge contours, ossification degree, and epiphyseal line clarity features are extracted, outputting a standardized 1024×1024 pixel image.

[0009] The AI-powered bone age assessment module automatically determines and grades bone age. It employs an improved ResNet50 model, taking preprocessed images and the child's gender and actual age as input. Trained on 100,000 clinical images from children aged 0-18 years, it supports switching between TW3 and CHN standards. Outputs bone age values, epiphyseal development grading, and key developmental descriptions. The Kappa value for consistency with physician assessment is ≥0.85, and abnormal images are automatically marked for further review.

[0010] The growth and development analysis module compares growth parameters with bone age. It includes built-in WHO and Chinese children's growth standard curves, calculates height percentiles, BMI, and annual growth rate. It analyzes the deviation between bone age and chronological age (ΔBA), marking developmental abnormalities with ΔBA > 1 year or < -1 year, calculating the standard deviation score (SDS) for height, and generating a comparison chart of growth trends from the last three data points.

[0011] The remote physician interaction module enables AI result review and doctor-patient communication. An encrypted video consultation platform is built, supporting 1080P resolution and latency ≤200ms. AI assessment results are automatically pushed to the pediatric endocrinologist's terminal, marking abnormalities. Physicians can annotate and modify results, upload diagnostic opinions and intervention plans, and transmit them via text, voice, and image attachments. Consultation records are archived with a child's identifier.

[0012] The data storage and traceability module constructs a hierarchical encrypted growth record. It utilizes dual storage via a 32GB SD card and Alibaba Cloud Medical Cloud, with AES-256 encryption for transmission, meeting Level 3 information security standards. The record includes images, parameters, assessment reports, diagnostic opinions, and intervention records, sorted chronologically, supporting multi-dimensional retrieval. Data is retained for up to 5 years after the child reaches adulthood, and access requires authentication.

[0013] The abnormality warning module monitors growth abnormalities and triggers alerts. Three threshold levels are set: Level 1: SDS ∈ [-2,-1) or (1,2], growth rate 3-5 cm / year; Level 2: SDS <-2 or >2, growth rate <3 cm / year; Level 3: Abnormal epiphyseal closure, ΔBA > 2 years or < -2 years. Upon triggering, alerts are sent via APP and SMS, and a doctor's terminal indicator lights up. Level 3 alerts automatically schedule consultations, with a response time ≤ 1 minute.

[0014] The human-computer interaction module supports operation and information display. It features a 10-inch touchscreen or mobile app to display growth curves, assessment reports, and warning information. It provides access to parameter entry, image uploading, and physician appointments, supports bilingual switching, requires fingerprint or password verification for critical operations, and synchronizes operation logs.

[0015] Furthermore, it also includes: The growth trend prediction module, in conjunction with the growth and development analysis module and the data storage module, constructs a dynamic prediction model to achieve accurate growth prediction. Input parameters include current height, bone age, chronological age, annual growth rate, parental genetic height, and gender. The model is trained based on 100,000 cases of clinical follow-up data categorized by gender and age group, and supports TW3-CHN dual-standard adaptation. The prediction formula is HA=Hc×(1+ΔBA×k+GR×m+Hg×n). Where HA is the predicted adult height in cm; Hc is the current height in cm; ΔBA is the difference between bone age and chronological age in years; k is the bone age influence coefficient (0.06 for males in infancy and 0.10 for puberty, 0.07 for females in infancy and 0.11 for puberty); GR is the annual growth rate in cm / year; m is the growth rate coefficient (0.03); Hg is the average genetic height of parents in cm; and n is the genetic influence coefficient (0.25). The prediction results are updated iteratively every 3 months by combining newly collected data, and a 95% confidence interval and key growth nodes are output to provide a time window reference for early intervention. The prediction error is reduced by more than 40% compared with traditional models.

[0016] Furthermore, it also includes: The image quality verification module is connected in series with the user-end acquisition module and image preprocessing module to construct a multi-dimensional quality screening system. It extracts four core indicators: sharpness (C), noise intensity (N), epiphyseal integrity (S), and posture conformity (T). Sharpness is calculated using the Sobel edge gradient operator to determine the average gradient value; noise is judged by the grayscale standard deviation; integrity is assessed by the proportion of key epiphyseal plates displayed; and posture conformity is detected by the angle between the epiphyseal line and the image midline (T ≤ 5° is considered standard). The quality scoring formula is Q = α × C - β × N + γ × S + δ × T. Where Q is the quality score, in points (0-100); α is the sharpness weight (0.4); β is the noise weight (0.3); γ is the integrity weight (0.2); δ is the posture weight (0.1); C, N, S, and T are all from 0-100 points, N decreases as noise increases, and T decreases as the angle increases. A score of ≥80 automatically enters the preprocessing stage; scores of 60-79 receive targeted re-examination guidance; scores <60 are rejected and marked with core issues, while the device ID is simultaneously recorded and associated with quality statistics. A parameter library for different brands of home X-ray equipment is established to automatically match the optimal verification threshold, improving the quality pass rate by 65% ​​compared to the no-verification mode.

[0017] Furthermore, the multi-center data collaboration module, based on the HL7FHIR standard, constructs a cross-institutional data sharing network, connecting pediatric endocrinology departments of tertiary hospitals, community health service centers, and child health care institution terminals. Data encapsulation adopts a structured format, including 32 fields across 5 categories: basic child information, bone age imaging metadata, AI assessment reports, time-series growth parameter data, and intervention program execution records. Transmission is via an IPsec-VPN encrypted channel, employing an "application-review-authorization-revocation" permission mechanism: after a community institution initiates a consultation request, a tertiary hospital physician completes the review within 15 minutes, granting access to limited data; access is automatically revoked 24 hours after the consultation ends. Multidisciplinary collaboration is supported; pediatric endocrinologists can invite nutritionists and orthopedic surgeons to join the consultation, with their respective approvals synchronized in real time, forming a structured multidisciplinary treatment record. A built-in offline synchronization mechanism temporarily stores data in a local encrypted folder during network interruptions, and incrementally uploads it upon recovery, achieving a cross-institutional data consistency error of ≤0.1%, improving consultation efficiency by 80% compared to traditional email transmission.

[0018] Furthermore, it includes an intelligent nutrition intervention suggestion module, which links with the growth and development analysis module, data storage module, and third-party dietary database to generate dynamic personalized plans. Inputting six parameters—BMI, body fat percentage, annual growth rate, bone age, dietary preferences, and activity level—matches the age-specific nutrient requirements standards of the "Chinese Dietary Guidelines for School-Aged Children 2022." The intervention intensity formula is NI=ω×(TW)+θ×(GRs-GR)+φ×(BA-CA). Where NI is the intervention intensity, in levels (1-5); ω is the weight deviation weight (0.4); T is the standard weight for the same age and sex (kg); W is the actual weight (kg); θ is the growth rate weight (0.3); GRs is the standard growth rate (cm / year); GR is the actual growth rate (cm / year); φ is the bone age deviation weight (0.3); BA is the bone age (years); and CA is the actual age (years). Levels 1-2 output daily dietary lists; levels 3-4 add supplement recommendations; level 5 automatically triggers a nutritionist connection to generate weekly menus and ingredient purchase lists. It works in conjunction with a smart refrigerator to identify household food ingredients, make real-time adjustments to recommendations, and update the plan monthly based on growth parameters.

[0019] Furthermore, it includes a tiered parent education guidance module, deeply integrated with the human-computer interaction and data storage modules to achieve precise content delivery. A science popularization content library is built based on children's age segments and abnormality types: short stature is addressed with videos on delayed bone age intervention and growth hormone secretion mechanisms; obesity with articles on BMI management and low-GI diets; and precocious puberty with animations on the risks of premature epiphyseal closure. Content is presented in a "physician on-camera explanation + animation demonstration" format, with videos ≤5 minutes long and text ≤1000 words, accompanied by authoritative literature citations. Intervention plans include implementation reminders; parents can upload photos of their diet and exercise records. The system uses image recognition to calculate the diet adherence rate and exercise duration achievement rate, sending incentive messages if the adherence rate is <60%. A Q&A community with manual review is established, where parents' questions are answered by pediatricians within 24 hours, with sensitive questions automatically filtered. This results in a 70% increase in educational content awareness and intervention compliance compared to traditional methods.

[0020] Furthermore, it includes a full-cycle equipment self-test and calibration module, which works in real-time with the user-end data acquisition module to ensure data acquisition accuracy. Upon power-on, the device initiates a three-level self-test: Level 1 hardware self-test; Level 2 accuracy self-test; and Level 3 compatibility self-test. Self-test anomalies display fault codes and graphical troubleshooting guides, and support QR code scanning to view engineer demonstration videos. The calibration mechanism is designed differently for each equipment type: height and weight scales receive monthly reminders for multi-point calibration, X-ray equipment receives quarterly dose calibration updates, and body fat sensors receive electrode cleaning reminders after every 10 uses. Calibration records include time, deviation values, and operator information, and are synchronized to the cloud to form a device health record, reducing data acquisition errors by 85% compared to the no-calibration mode.

[0021] Furthermore, it includes in-depth privacy protection and identity authentication modules, covering the entire data lifecycle for security. It employs a three-tiered protection system: hardware encryption, transmission encryption, and storage encryption. An SM4 encryption chip is built into the user device, storing the child's unique identifier and parent's identity information. Transmission uses HTTPS with digital signatures, and each data entry is timestamped and signed by the device. Cloud storage uses fragmented encryption, complying with the Personal Information Protection Law and medical data security regulations. Identity authentication employs a tiered mechanism: the parent's side uses facial recognition and dynamic verification codes, with facial recognition integrating liveness detection, and the verification code valid for 5 minutes and bound to the device; the doctor's side uses an NFC chip on the employee badge and fingerprint verification, with the badge containing an encrypted ID and fingerprints extracting 68 feature points with a recognition rate of ≥99.2%; the administrator's side requires triple verification using a password, Ukey, and facial recognition. Access logs are stored on a blockchain, ensuring immutability and automatically initiating an emergency response in the event of a data breach, improving privacy protection compliance by 90% compared to traditional systems.

[0022] Furthermore, it also includes: The closed-loop intervention effectiveness evaluation module is linked with the growth and development analysis module, nutritional intervention module, and telemedicine module. Standardized evaluation periods are set: 1 month, 3 months, and 6 months after intervention. Comparison indicators include four core parameters: growth rate (ΔGR), height SDS change (ΔSDS), BMI deviation (ΔBMI), and bone age progression rate (ΔBA / ΔCA). Effectiveness criteria are: ΔGR ≥ 1 cm / year and ΔSDS ≥ 0.5 for significant effectiveness; ΔGR ≥ 0.5 cm / year or ΔSDS ≥ 0.3 for generally effective effectiveness; failure to meet these criteria indicates ineffectiveness. The effectiveness rate is calculated and a trend curve is generated, marking the onset time and optimal intervention period. For cases with two consecutive ineffective evaluations, dietary records, exercise data, and sleep duration are automatically retrieved to analyze influencing factors, and physician adjustment suggestions are pushed, simultaneously triggering multidisciplinary consultations. The intervention effectiveness rate is 60% higher than traditional empirical adjustments.

[0023] Furthermore, it also includes: The intelligent emergency response and referral module is linked with the abnormal warning module, remote physician module, and regional medical resource database. Upon triggering a Level 3 warning, the system automatically searches for pediatric endocrinology specialist institutions within a 3-kilometer radius, sorting them by distance (nearest to farthest), patient volume (highest to lowest), and patient rating (highest to lowest), displaying information such as institution level, physician qualifications, remaining appointment slots, and average consultation time. Appointments are prioritized: Level 3 warnings have higher priority than Level 2 warnings and can be prioritized for appointments within 24 hours. An encrypted referral data package is generated, containing complete growth records, abnormal indicator annotations, AI assessment heatmaps, and intervention records, and pushed to the receiving institution within 10 minutes. Upon confirmation of receipt, the receiving institution provides appointment information. A 24 / 7 intelligent hotline is provided, using voice recognition to determine the type of abnormality and automatically transferring the patient to the corresponding specialist physician for immediate guidance. A follow-up mechanism is initiated after referral: the receiving physician is reminded to provide initial diagnosis within one week, and growth parameter changes are tracked for one month, forming a closed loop of "warning-referral-follow-up," reducing emergency abnormality handling time by 70% compared to traditional methods.

[0024] Compared with existing technologies, the beneficial effects of this invention are: This invention addresses the pain points of traditional children's bone age assessment and growth monitoring, such as "limited time and space, insufficient accuracy, poor coordination, and inefficient intervention," through deep integration of multiple modules and technological innovation throughout the entire process. It achieves an upgrade from "single offline assessment" to "remote full-cycle management," providing precise, convenient, and safe medical support for children's growth and development.

[0025] The system achieves a dual improvement in remote convenience and assessment accuracy. The user-end data collection module is compatible with both home and hospital devices, and combined with automatic collection reminders, it breaks down the time and space limitations of offline assessments, allowing parents to complete data collection without frequently taking their children to the hospital. The image quality verification module significantly improves the image quality input to the AI ​​model through multi-dimensional indicator screening and targeted re-image guidance. Combined with an improved age-group training model, it significantly enhances the accuracy of bone age assessment and its compatibility with different standards. The device self-test and calibration module enables real-time monitoring and periodic calibration of hardware status, reducing collection errors at the source and laying the foundation for accurate assessments. This effectively solves the problems of traditional assessment methods, such as reliance on offline resources and accuracy being affected by equipment and images.

[0026] The scientific rigor and forward-looking nature of growth monitoring have been significantly enhanced. The growth trend prediction module integrates multi-dimensional parameters such as height, bone age, and genetics to construct a dynamic prediction model that can accurately predict adult height and key growth milestones, providing a clear time window for early intervention. The growth and development analysis module combines dual-standard growth curves to dynamically compare the matching degree between bone age and actual age, automatically marking abnormal deviations and generating continuous growth trend charts, allowing physicians and parents to intuitively grasp the patterns of growth changes. This multi-parameter integrated monitoring mode is more scientific than traditional single-parameter recording, enabling early detection and early warning of developmental abnormalities.

[0027] Cross-institutional collaboration and closed-loop management efficiency have been significantly optimized. The multi-center data collaboration module builds a shared network based on standardized protocols, enabling real-time encrypted transmission and controlled access sharing of cross-institutional data. Cross-institutional consultations no longer require manual data transfer, significantly improving collaboration efficiency. The intelligent nutrition intervention module generates dynamic plans based on children's individual characteristics, enabling real-time adjustments through linkage with third-party devices. Combined with the tiered parent education and guidance module's precise science popularization and implementation reminders, the personalization of intervention plans and parental compliance have been greatly improved. The intervention effect evaluation module promotes dynamic optimization of the plan through multi-period indicator comparison and ineffective intervention analysis, forming a complete closed loop of "assessment-intervention-tracking-adjustment," solving the problems of generic plans, weak implementation, and difficulty in monitoring effects in traditional interventions.

[0028] Data security and privacy protection capabilities have been comprehensively enhanced. The in-depth privacy protection module constructs a three-tiered encryption system encompassing hardware, transmission, and storage. Combined with tiered identity authentication and blockchain-based access logs, security is ensured throughout the entire data lifecycle, from generation to storage and use, fully complying with medical data security standards and effectively mitigating the risk of privacy leaks. This comprehensive security design not only safeguards children's sensitive information but also enhances parents' trust in the system.

[0029] Overall, this invention effectively lowers the assessment threshold and improves monitoring accuracy and intervention effects by integrating remote data collection, precise assessment, scientific prediction, collaborative sharing, personalized intervention, and secure storage technologies. It balances medical professionalism, parental convenience, and data security, perfectly meeting the needs of full-cycle management of children's growth and development, and providing key technical support for the standardized and intelligent development of children's health management. Attached Figure Description

[0030] Figure 1 This is a schematic block diagram of the telemedicine system for children's bone age assessment and growth monitoring proposed in this invention. Figure 2 This is a comparison chart of the core performance of the traditional method and the system of this invention; Figure 3 A comparison chart of the accuracy of bone age assessment in different age groups; Figure 4 A graph showing how the effectiveness of the intervention program changes over time; Figure 5 A comparison chart showing the time taken for each step in cross-organizational data collaboration. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0033] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0034] Reference Figures 1 to 5 A telemedicine system for assessing bone age and monitoring growth in children, comprising: The user-end data acquisition module is used to acquire children's bone age images and physiological growth parameters. It is compatible with home portable X-ray imaging equipment and hospital PACS systems, transmitting bone age images via the DICOM 3.0 protocol. The device resolution is ≥2048×2048, and the exposure dose is ≤0.1mSv. It integrates a smart height and weight scale, a body fat percentage sensor, and a heart rate monitoring module. Height accuracy is ±0.1cm, weight accuracy is ±0.01kg, body fat measurement range is 5%-40%, sampling frequency is 1Hz, and data is transmitted via Bluetooth 5.0. It automatically reminds users to collect data every 3 months, and supports manual triggering in emergencies. Data includes a timestamp, device ID, and a unique child identifier.

[0035] The image preprocessing module optimizes bone age images and extracts key regions. Adaptive median filtering removes noise, and the Retinex enhancement algorithm improves the contrast between the epiphysis and metaphysis. Based on the U-Net segmentation model, the epiphyseal regions of the radius, ulna, and metacarpophalangeal bones are located, and features such as edge contours, ossification degree, and epiphyseal line clarity are extracted. A standardized 1024×1024 pixel image is output, with a processing time of ≤3 seconds and a feature extraction accuracy of ≥98%.

[0036] The AI-powered bone age assessment module automatically determines and grades bone age. It employs an improved ResNet50 model, taking preprocessed images and the child's gender and actual age as input. Trained on 100,000 clinical images from children aged 0-18 years, it supports switching between TW3 and CHN standards. Outputs bone age values ​​(accuracy ±0.1 years), epiphyseal development grades (1-9), and key developmental descriptions. The Kappa value for consistency with physician assessment is ≥0.85, and abnormal images are automatically marked for further review.

[0037] The growth and development analysis module compares growth parameters with bone age. It includes built-in WHO and Chinese children's growth standard curves, calculates height percentiles, BMI, and annual growth rate. It analyzes the deviation between bone age and chronological age (ΔBA), marking developmental abnormalities with ΔBA > 1 year or < -1 year, calculating the standard deviation score (SDS) for height, and generating a comparison chart of growth trends from the last three data points.

[0038] The remote physician interaction module enables AI result review and doctor-patient communication. An encrypted video consultation platform is built, supporting 1080P resolution and latency ≤200ms. AI assessment results are automatically pushed to the pediatric endocrinologist's terminal, marking abnormalities. Physicians can annotate and modify results, upload diagnostic opinions and intervention plans, and transmit them via text, voice, and image attachments. Consultation records are archived with a child's identifier.

[0039] The data storage and traceability module constructs a hierarchical encrypted growth record. It utilizes dual storage via a 32GB SD card and Alibaba Cloud Medical Cloud, with AES-256 encryption for transmission, meeting Level 3 information security standards. The record includes images, parameters, assessment reports, diagnostic opinions, and intervention records, sorted chronologically, supporting multi-dimensional retrieval. Data is retained for up to 5 years after the child reaches adulthood, and access requires authentication.

[0040] The abnormality warning module monitors growth abnormalities and triggers alerts. Three threshold levels are set: Level 1: SDS ∈ [-2,-1) or (1,2], growth rate 3-5 cm / year; Level 2: SDS <-2 or >2, growth rate <3 cm / year; Level 3: Abnormal epiphyseal closure, ΔBA > 2 years or < -2 years. Upon triggering, alerts are sent via APP and SMS, and a doctor's terminal indicator lights up. Level 3 alerts automatically schedule consultations, with a response time ≤ 1 minute.

[0041] The human-computer interaction module supports operation and information display. It features a 10-inch touchscreen or mobile app to display growth curves, assessment reports, and warning information. It provides access to parameter entry, image uploading, and physician appointments, supports bilingual switching, requires fingerprint or password verification for critical operations, and synchronizes operation logs.

[0042] This invention also includes: The growth trend prediction module, in conjunction with the growth and development analysis module and the data storage module, constructs a dynamic prediction model to achieve accurate growth prediction. Input parameters include current height, bone age, chronological age, annual growth rate, parental genetic height, and gender. The model is trained based on 100,000 cases of clinical follow-up data categorized by gender and age group (0-3 years infancy, 4-6 years in preschool, 7-12 years in school age, and 13-18 years in puberty), and supports TW3-CHN dual-standard adaptation. The prediction formula is HA=Hc×(1+ΔBA×k+GR×m+Hg×n). Where HA is the predicted adult height in cm; Hc is the current height in cm; ΔBA is the difference between bone age and actual age in years; k is the bone age influence coefficient, 0.06 for males in infancy and 0.10 for puberty, and 0.07 for females in infancy and 0.11 for puberty; GR is the annual growth rate in cm / year; m is the growth rate coefficient, 0.03; Hg is the average genetic height of parents in cm; and n is the genetic influence coefficient, 0.25. The prediction results are iteratively updated every 3 months based on newly collected data, outputting a 95% confidence interval and key growth nodes (such as the peak growth period during puberty), providing a time window reference for early intervention. The prediction error is reduced by more than 40% compared to traditional models.

[0043] This invention also includes: The image quality verification module, connected in series with the user-end acquisition module and image preprocessing module, constructs a multi-dimensional quality screening system. It extracts four core indicators: sharpness (C), noise intensity (N), epiphyseal integrity (S), and postural conformity (T). Sharpness is calculated using the Sobel edge gradient operator to determine the average gradient value (C≥80 for acceptable results). Noise is assessed using the grayscale standard deviation (N≤20 for home devices, N≤15 for hospital devices). Integrity is evaluated by the proportion of key epiphyses (radius, 3rd metacarpal) shown (S≥90% for completeness). Postural conformity is determined by the angle between the epiphyseal line and the image midline (T≤5° for conformity). The quality scoring formula is Q=α×C-β×N+γ×S+δ×T. Where Q is the quality score, in points (0-100); α is the sharpness weight (0.4); β is the noise weight (0.3); γ is the integrity weight (0.2); δ is the body position weight (0.1); C, N, S, and T are all in points (0-100), with N decreasing as noise increases and T decreasing as the angle increases. A score of Q ≥ 80 automatically enters the preprocessing stage; scores of 60-79 provide targeted re-examination guidance (e.g., high noise prompts adjustment of exposure dose, body position deviation prompts children's palms down); scores < 60 are rejected and the core issue is marked, and the device ID is simultaneously recorded and associated with quality statistics. A parameter library for different brands of home X-ray equipment is established, automatically matching the optimal verification threshold, improving the quality pass rate by 65% ​​compared to the no-verification mode.

[0044] In this invention, a multi-center data collaboration module, based on the HL7FHIR standard, constructs a cross-institutional data sharing network, connecting the pediatric endocrinology departments of tertiary hospitals, community health service centers, and child health care institution terminals. Data encapsulation adopts a structured format, including 32 fields across 5 categories: basic child information (de-sensitized), bone age imaging metadata (equipment model, imaging parameters), AI assessment reports (including feature heatmaps), time-series growth parameter data, and intervention program execution records. Data is transmitted via an IPsec-VPN encrypted channel, employing an "application-review-authorization-revocation" permission mechanism: after a community institution initiates a consultation request, a tertiary hospital physician completes the review within 15 minutes, granting access to limited data (hiding sensitive information such as detailed home address and parents' occupations), and automatically revoking access 24 hours after the consultation ends. Multidisciplinary collaboration is supported; pediatric endocrinologists can invite nutritionists and orthopedic surgeons to join the consultation, with each physician's approvals synchronized in real time, forming a structured multidisciplinary treatment record. With a built-in offline synchronization mechanism, data is temporarily stored in a local encrypted folder when the network is interrupted, and then uploaded incrementally after the network is restored. The cross-institutional data consistency error is ≤0.1%, and the consultation efficiency is improved by 80% compared with traditional email transmission.

[0045] This invention also includes an intelligent nutritional intervention suggestion module, which links with the growth and development analysis module, data storage module, and third-party dietary database to generate dynamic personalized plans. Input parameters include six items: BMI, body fat percentage, annual growth rate, bone age, dietary preferences (vegetarian / meat-based, allergies / restrictions), and activity level (sedentary / light / moderate / high-intensity activity), matching the age-appropriate nutrient requirements standards of the "Chinese Dietary Guidelines for School-Aged Children 2022". The intervention intensity formula is NI=ω×(TW)+θ×(GRs-GR)+φ×(BA-CA). Where NI is the intervention intensity, in levels (1-5); ω is the weight deviation weight (0.4); T is the standard weight for the same age and sex (kg); W is the actual weight (kg); θ is the growth rate weight (0.3); GRs is the standard growth rate (cm / year, 7.5 for males and 7.3 for females in 7-year-old children); GR is the actual growth rate (cm / year); φ is the bone age deviation weight (0.3); BA is the bone age (years); and CA is the actual age (years). Levels 1-2 generate daily meal plans (e.g., breakfast: 250ml milk + 1 egg + 50g whole wheat bread); Levels 3-4 recommend supplements (Vitamin D 400-800 IU / day, Calcium 800-1200mg / day, brand and compatibility range indicated); Level 5 automatically triggers a nutritionist connection to generate weekly meal plans and ingredient shopping lists. It integrates with a smart refrigerator to identify household ingredients, adjust suggestions in real time, and update the plan monthly based on growth parameters, increasing nutrient intake compliance by 50% compared to general plans.

[0046] This invention also includes a tiered parent education guidance module, deeply integrated with the human-computer interaction module and data storage module to achieve precise content delivery. A science popularization content library is built based on children's age segments and abnormality types: for short stature, videos on delayed bone age intervention and growth hormone secretion mechanisms are provided; for obesity, articles on BMI management and low-GI diets are provided; for precocious puberty, animations on the risks of premature epiphyseal closure are provided. Content is presented in a "physician on-camera explanation + animation demonstration" format, with videos ≤5 minutes long and text ≤1000 words, accompanied by authoritative literature citations (such as relevant guidelines from the *Chinese Journal of Pediatrics*). Intervention plans include implementation reminders (such as a daily 8 AM vitamin D dosage reminder). Parents can upload photos of their children's diets and exercise records. The system uses image recognition to calculate the diet adherence rate (ingredient matching degree) and exercise duration achievement rate. If the adherence rate is <60%, an incentive message is sent (such as "Achieving the target for 3 consecutive days can be redeemed for a children's physical examination coupon"). A manually reviewed Q&A community is established, where parents' questions are answered by pediatricians within 24 hours, and sensitive questions are automatically filtered. The awareness rate of educational content and intervention compliance are improved by 70% compared to traditional models.

[0047] This invention also includes a full-cycle device self-test and calibration module, which is linked in real time with the user-end data acquisition module to ensure acquisition accuracy. After the device is powered on, it initiates a three-level self-test: Level 1 hardware self-test (X-ray equipment checks exposure tube voltage / current, sensors check electrode conductivity, time ≤ 2 seconds); Level 2 accuracy self-test (height and weight scales automatically detect zero-point drift, deviation > 0.05 kg / cm prompts calibration; body fat sensors compare with built-in standard resistors, deviation > 2% alarms); Level 3 compatibility self-test (verifies device ID and system compatibility version, pushes firmware updates for older devices). Self-test anomalies display fault codes (e.g., E01 for X-ray dose abnormality, E02 for scale drift) and graphic troubleshooting guidelines, and support for scanning QR codes to watch engineer demonstration videos. The calibration mechanism is designed differently according to device type: height and weight scales remind users to perform multi-point calibration monthly (0kg, 20kg, and 50kg three-point verification using standard weights), X-ray equipment pushes dose calibration procedures quarterly (requires integration with a third-party dosimeter), and body fat sensors remind users to clean the electrodes after every 10 uses. The calibration record includes time, deviation value, and operator information, and is synchronized to the cloud to form a device health record. The data collection error is reduced by 85% compared to the no-calibration mode.

[0048] This invention also includes a privacy-in-depth protection and identity authentication module, covering the entire data lifecycle for security. It employs a three-tiered protection system: hardware encryption, transmission encryption, and storage encryption. The SM4 encryption chip is built into the user terminal device, storing the child's unique identifier and parent's identity information. Transmission uses HTTPS + digital signature (RSA2048 algorithm), with each data entry appended with a timestamp and device signature. Cloud storage uses fragmented encryption (data is split into 10 fragments, encrypted with different keys), complying with the Personal Information Protection Law and medical data security regulations. Identity authentication employs a tiered mechanism: the parent terminal uses "face recognition + dynamic verification code," with face recognition integrating liveness detection (dynamic head nodding and blinking verification to prevent photo forgery), and the verification code is valid for 5 minutes and bound to the device; the doctor terminal uses "employee badge NFC chip + fingerprint verification," with the employee badge containing an encrypted ID, and fingerprints extracting 68 feature points with a recognition rate ≥99.2%; the administrator terminal requires triple verification: "password + Ukey + face recognition." Access logs are stored on a blockchain (with consortium blockchain nodes including hospitals and regulatory agencies), making them tamper-proof. In the event of a data breach, an emergency response is automatically initiated (freezing suspicious accounts and sending alerts to users and regulators). Privacy protection compliance is improved by 90% compared to traditional systems.

[0049] This invention also includes: The closed-loop intervention effectiveness evaluation module is linked with the growth and development analysis module, nutritional intervention module, and telemedicine module. Standardized evaluation cycles are set: 1 month (initial evaluation), 3 months (secondary evaluation), and 6 months (final evaluation) after intervention. Comparison indicators include four core parameters: growth rate (ΔGR), height SDS change (ΔSDS), BMI deviation (ΔBMI), and bone age progression rate (ΔBA / ΔCA). Effectiveness criteria: ΔGR ≥ 1 cm / year and ΔSDS ≥ 0.5 are considered significantly effective; ΔGR ≥ 0.5 cm / year or ΔSDS ≥ 0.3 are considered generally effective; failure to meet these criteria indicates ineffectiveness. The effectiveness rate is calculated and a trend curve is generated, marking the onset time (e.g., increased growth rate 2 months after vitamin D supplementation) and the optimal intervention period. For cases with two consecutive ineffective evaluations, dietary records, exercise data, and sleep duration (connected to smart bracelets) are automatically retrieved to analyze influencing factors (e.g., malabsorption, insufficient exercise), and physician adjustment suggestions are pushed (e.g., increasing calcium supplement dosage, optimizing exercise type). Simultaneously, multidisciplinary consultations are triggered, increasing the intervention effectiveness rate by 60% compared to traditional empirical adjustments.

[0050] This invention also includes: The intelligent emergency response and referral module is linked with the abnormal warning module, the remote physician module, and the regional medical resource database. Upon triggering a Level 3 warning, the system automatically searches for pediatric endocrinology specialist institutions within a 3-kilometer radius, sorting them by distance (nearest to farthest), patient volume (highest to lowest), and patient rating (highest to lowest), displaying information such as institution level, physician qualifications (title, specialty), remaining appointment slots, and average consultation time. Appointments are subject to a priority mechanism: Level 3 warnings (such as abnormal epiphyseal closure) have higher priority than Level 2 warnings and can be prioritized for appointments within 24 hours. An encrypted referral data package (using the SM4 algorithm) is generated, containing a complete growth record, abnormal indicator annotations, an AI assessment heatmap, and intervention records, and is pushed to the receiving institution within 10 minutes. Upon confirmation of receipt, the institution provides appointment information. A 24 / 7 intelligent hotline is provided, using voice recognition to determine the type of abnormality (such as "growth retardation" or "advanced bone age") and automatically transferring the patient to the corresponding specialist physician for immediate guidance. After referral, a follow-up mechanism is initiated: the attending physician is reminded to provide feedback on the initial diagnosis within one week, and changes in growth parameters are tracked for one month, forming a closed loop of "early warning-referral-follow-up". The time for handling emergency abnormalities is shortened by 70% compared to the traditional model.

[0051] Specific examples of telemedicine systems for children's bone age assessment and growth monitoring: Example 1

[0052] Remote bone age assessment and growth monitoring system for large-scale children's hospitals (application scenario at Beijing Children's Hospital) This embodiment addresses the remote diagnosis and treatment needs of the Department of Pediatric Endocrinology at Beijing Children's Hospital. This department receives over 30,000 children annually for bone age assessment and needs to connect with 20 community health service centers and 5 municipal hospitals. The traditional model suffers from overcrowding during offline assessments and data incompatibility across institutions, resulting in an average assessment cycle of 7 days. The solution of this invention enables precise remote management, and the specific implementation process is as follows.

[0053] 1. System hardware setup and parameter configuration User-end data acquisition module: For hospitals, it is compatible with GEDR-X-ray equipment (resolution 3072×3072, exposure dose 0.08mSv) and connects to the PACS system via the DICOM 3.0 protocol; for communities and homes, it is compatible with Yuwell portable X-ray machines (resolution 2048×2048, dose 0.1mSv), and paired with a Huawei smart height and weight scale (height accuracy ±0.1cm, weight ±0.01kg) and an Omron body fat sensor (measurement range 5%-40%). All devices connect via Bluetooth 5.0 or Ethernet, and automatically push data collection reminders every 3 months and 1 day. Data includes the child's unique medical ID (including date of birth and gender code).

[0054] Image preprocessing module: Adaptive median filtering is implemented using Python, with a dynamic window size range of 3-7. The window size is automatically reduced when the noise level is >20. The scale parameter of the Retinex enhancement algorithm is set to 800, improving epiphyseal contrast by 30%. Based on the U-Net segmentation model (input 1024×1024 image, output 512×512 mask), after training on 100,000 images of different genders and age groups (0-3 years, 4-6 years, etc.), the accuracy of radial epiphyseal extraction is 98.5%, and the processing time is 2.2 seconds.

[0055] AI Bone Age Assessment Module: Improved ResNet50 model with added age-segmented convolutional layers. 3×3 small convolutional kernels are used for infants aged 0-3 years, and an attention mechanism is added for adolescents aged 13-18 years. Supports one-click switching between TW3-CHN standards. Input images and child's gender (1 male, 0 females), actual age, output bone age value (accuracy ±0.1 years), epiphyseal grade (1-9), and key feature descriptions (e.g., "3rd metacarpal epiphyseal closure 70%)". Kappa value consistent with chief physician assessment is 0.88. Abnormal images (e.g., premature epiphyseal closure) are marked with a red border.

[0056] Growth Trend Prediction Module: This module stores 5 years of growth time-series data from the data storage module. Inputs include current height (Hc), bone age (BA), chronological age (CA), annual growth rate (GR), and average genetic height (Hg) of parents. For example, a 7-year-old male child has Hc=125cm, BA=7.5 years, CA=7 years (ΔBA=0.5), GR=6cm / year, and Hg=175cm. Substituting these values ​​into HA=Hc×(1+ΔBA×k+GR×m+Hg×n), where k=0.08 (for school-aged males), m=0.03, and n=0.25, we get HA=125×(1+0.5×0.08+6×0.03+175×0.25)=125×1.473=184.125cm, with a 95% confidence interval of 182-186cm.

[0057] Multi-center data collaboration module: Based on the HL7FHIR standard, it encapsulates 32 fields, including image metadata (device model, shooting time), AI assessment heatmap, and growth parameter time series tables. It connects to 20 community terminals via IPsec-VPN. After a community initiates a consultation, hospital physicians review and authorize the consultation within 10 minutes, with access restrictions limited to 24 hours. It supports dual-disciplinary consultations between pediatric endocrinology and nutrition departments, with opinions synchronized to children's records in real time, achieving a synchronization error of 0.05%.

[0058] 2. Implementation of core functions and application of formulas Image quality verification: Images captured by home devices are analyzed by the module, with the following metrics: sharpness C=75, noise N=25, integrity S=90, and position T=85. Substituting these into Q=α×C-β×N+γ×S+δ×T, where α=0.4, β=0.3, γ=0.2, and δ=0.1, we get Q=0.4×75-0.3×25+0.2×90+0.1×85=30-7.5+18+8.5=49 points < 60 points. The system marks this as "noise too high" and sends a notification to "adjust the exposure dose to 0.09mSv and keep the child's wrist flat." After reshooting, Q=82 points and the image enters preprocessing.

[0059] Intelligent Nutritional Intervention: For a 6-year-old female child with a BMI of 18 (overweight), GR of 4 cm / year (below the standard GRs=7.3), BA of 6.5 years, CA of 6 years (ΔBA=0.5), and allergy to milk, the module matches the guidelines. The NI is calculated as ω×(TW)+θ×(GRs-GR)+φ×(BA-CA), where ω=0.4, T=20kg, W=22kg, θ=0.3, and φ=0.3. Therefore, NI=0.4×(20-22)+0.3×(7.3-4)+0.3×0.5=-0.8+0.99+0.15=0.34→Level 2. A milk-free diet list (breakfast: soy milk + egg + whole wheat bread) is provided, along with a low-GI food list, updated monthly.

[0060] Intervention effectiveness evaluation: A child with growth retardation underwent a 3-month intervention. The comparative indicators ΔGR = 1.2 cm / year and ΔSDS = 0.6, indicating significant effectiveness. A trend curve showed an increase in growth rate after 2 months of intervention, marking "vitamin D + calcium supplementation effective." Another child showed no improvement after two consecutive assessments. Reviewing the child's activity records revealed an average daily activity of less than 30 minutes. The physician was advised to adjust the treatment plan to "increase rope skipping exercise to 15 minutes daily," simultaneously triggering a nutrition consultation.

[0061] Privacy protection: Parent login uses facial recognition (liveness detection with head nodding) + 6-digit verification code, while doctor login uses NFC-enabled employee badges + fingerprint verification (extracting 68 feature points). Data transmission is HTTPS + RSA2048 signed, stored in cloud-sharded encrypted form, and access logs are uploaded to the hospital and regulatory authority nodes on the blockchain, ensuring immutability.

[0062] 3. Performance data representation Table 1: Performance Comparison between Traditional Mode and Invention System Table 1 shows data from a one-month operational statistical analysis. The traditional model requires parents to bring children to and from the hospital, with an assessment cycle of up to 7 days, and an image pass rate of only 65% ​​due to quality issues. Inter-institutional consultations rely on manual data transfer, taking 48 hours, and intervention plans are mostly generic templates with poor adaptability. This invention, through remote data acquisition and intelligent verification, reduces the assessment cycle to 2 hours and increases the image pass rate to 92%. Automatic cross-institutional data synchronization completes consultations in 15 minutes, and personalized intervention plans achieve 95% adaptability. Three levels of privacy protection reduce the risk of leakage, perfectly adapting to the needs of multi-center collaboration and precision medicine in large hospitals. Example 2

[0063] Child growth monitoring system in a primary community health service center (application scenario in a community in Shanghai) This embodiment describes the deployment of two systems in a community health service center in Shanghai, serving 3,000 children aged 0-12 in the area. The center lacks specialist physicians for bone age assessment and traditionally relies on referrals to higher-level hospitals every six months, which delays intervention. The solution of this invention enables primary care at the grassroots level and remote review. The specific implementation process is as follows.

[0064] 1. System compatibility configuration and debugging User-side data acquisition module: Utilizing a cost-effective Biolight portable X-ray machine (2048×2048 resolution, 0.1mSv dose), paired with a Xiaomi smart height and weight scale (height ±0.2cm, weight ±0.02kg), the acquisition process is simplified: parents follow the APP instructions to take a wrist bone image, the device automatically uploads the data, and attaches the child's community health record ID. Acquisition reminders are sent every 3 months, and emergency situations can be triggered by phone.

[0065] Image preprocessing module: Simplified algorithm parameters, adaptive median filter window fixed at 5, Retinex scale parameter at 600, processing time 2.8 seconds. U-Net segmentation model retains the identification of the radial and third metacarpal core epiphyses, with an accuracy of 97%, reducing computing power requirements and adapting to low-end terminals in the community.

[0066] AI bone age assessment module: Defaults to CHN standards (conforming to domestic primary care practices), hides the model parameter adjustment entry, and only allows the "Assessment - View Results" function. Abnormal images are automatically pushed to the pediatrician's terminal at a higher-level hospital after being marked, along with the community's preliminary examination record (height, weight, medical history).

[0067] Growth Trend Prediction Module: Simplified input parameters (current height, bone age, actual age, and gender only), with optional parental genetic height. For example, for a 5-year-old female child, Hc=110cm, BA=5.2 years, CA=5 years (ΔBA=0.2), GR=5cm / year, substituting into HA=110×(1+0.2×0.09+5×0.03)=110×1.168=128.48cm, the prediction is pushed to the parent's app with a notification: "Expected growth spurt in puberty is 10 years old."

[0068] Emergency Referral Module: This module links to the pediatric resource databases of two secondary hospitals within the jurisdiction. Upon triggering a Level 3 alert (e.g., ΔBA = 2.5 years), it searches for hospitals within a 3-kilometer radius and displays them sorted by distance: Hospital A (1.2 km, 5 available appointments, specializing in precocious puberty) and Hospital B (2.8 km, 3 available appointments). An encrypted data package (including images and growth curves) is generated and pushed to Hospital A within 5 minutes. One-click appointment booking within 24 hours is supported.

[0069] 2. Implementation of core functions and application of formulas Image quality verification: The image taken by the community parent has C=80, N=22, S=85, T=75. Substituting into Q=0.4×80-0.3×22+0.2×85+0.1×75=32-6.6+17+7.5=50.9 points, the system pushes "posture deviation, wrist rotated inward 5° to retake the shot". After the retake, Q=81 points passed the verification, which is 2 fewer retakes than the traditional manual guidance.

[0070] Intelligent Nutritional Intervention: An 8-year-old male child with a BMI of 15 (underweight), GR of 4 cm / year (standard GRs = 7.5), BA of 7.8 years, CA of 8 years (ΔBA = -0.2), and a vegetarian diet, was input and the NI was calculated as follows: NI = 0.4 × (25-22) + 0.3 × (7.5-4) + 0.3 × (-0.2) = 1.2 + 1.05 - 0.06 = 2.19 → Level 3. A list of high-protein vegetarian foods (tofu, chickpeas) was recommended, along with a vitamin D supplement recommendation (400 IU / day). Discounts on food purchases were offered in conjunction with the community supermarket.

[0071] Parental Education Guidance: The system identifies children with "slow growth rate" and pushes a 3-minute video, "3 Dietary Misconceptions about Children's Growth Delay," along with a link to the guidelines in the *Chinese Journal of Pediatrics*. Daily exercise reminders are set (a "10-minute jump rope" reminder is pushed at 7 PM). Parents upload their exercise check-in records. If the compliance rate is less than 60%, a "redeem a children's toothbrush for 3 consecutive days of check-ins" message is pushed. After one month, the compliance rate increases to 85%.

[0072] Equipment calibration: The height and weight scale is reminded to be calibrated on the 1st of each month. 1kg and 5kg standard weights are provided (provided by the community). The APP displays the calibration steps: "Place the 1kg weight and wait for the value to stabilize before confirming". The calibration record is synchronized to the cloud. The scale drift has been reduced from 0.1kg to 0.02kg.

[0073] 3. Performance data representation Table 2: System performance under different scenarios Table 2 data comes from a two-month test. The traditional model requires community referrals to higher-level hospitals, with a 14-day assessment period. Due to insufficient guidance, the image pass rate is only 70%, the intervention implementation rate is 50%, the referral response time is 72 hours, and parental satisfaction is low. This invention completes routine community monitoring in 3 hours and home self-collection in 1.5 hours, with an image pass rate exceeding 88%. Through education, guidance, and incentive mechanisms, the intervention implementation rate is increased to over 82%. Abnormal referral response time is only 10 minutes, and parental satisfaction reaches over 90%. The system simplifies operation and adapts to grassroots levels, while remote verification solves the problem of insufficient professional resources, perfectly meeting the needs of community-based child growth monitoring.

[0074] Figure 2 This invention visually demonstrates its groundbreaking optimization of traditional child growth monitoring models. Traditional offline methods rely on parents bringing children to and from hospitals, with assessment cycles lasting up to 7 days. Due to a lack of professional guidance, the pass rate for imaging is only 65%. Inter-institutional consultations require manual data transfer, taking 48 hours. Intervention plans are mostly generic templates with less than 60% adaptability, and data encryption is risky due to its singular nature. This invention breaks through time and space limitations with a remote acquisition module, increases the pass rate to 92% with an image quality verification module, enables rapid consultations within 15 minutes with a multi-center collaboration module, and generates personalized plans with 95% adaptability with an intelligent nutrition intervention module. Three levels of privacy protection ensure a risk score of 95, perfectly solving the pain points of traditional models: low efficiency, poor accuracy, and weak collaboration.

[0075] Figure 3 This invention clearly demonstrates the superior accuracy of its AI bone age assessment module. Traditional AI models, failing to consider the differences in children's growth stages and employing a uniform convolutional structure, achieve an accuracy rate of only 80% during adolescence (13-18 years old) due to the complexity of epiphyseal development, with an average of 84.7% across all age groups, falling below clinical standards. The model of this invention optimizes its structure for different age groups: small convolutional kernels are used to capture subtle ossification features during infancy, while an attention mechanism is superimposed during adolescence to focus on the epiphyseal closure area. Accuracy exceeds 95% across all age groups, averaging 97.2% across all age groups, significantly higher than traditional models, fully demonstrating the technical value of age-specific training.

[0076] Figure 4This invention highlights the high efficiency of its closed-loop intervention mechanism. Traditional, generic interventions fail to consider children's dietary preferences and growth characteristics, offering only standardized recommendations. After one month of intervention, the growth rate increases by only 0.1 cm / year, plateauing to 0.3 cm / year after three months, failing to reach the effective threshold. This invention's intelligent nutrition intervention module generates personalized plans, linking food identification with real-time adjustments. Combined with the parent education module's reminders and incentive mechanisms, the growth rate increases by 0.5 cm / year after one month, reaching the effective standard; after three months, it increases to 1.2 cm / year; and after four months, it stabilizes at 1.3 cm / year. The intervention effect evaluation module dynamically optimizes the plan, forming an "evaluation-intervention-tracking" closed loop, resulting in faster onset and longer-lasting effects compared to traditional methods.

[0077] Figure 5 This invention showcases the technological advantages of a multi-center data collaboration module. In the traditional manual file transfer model, physician review takes 4 hours, data transmission via USB drive or email takes 2 hours, and manual summarization of opinions takes 1 hour, resulting in a consultation process that takes 7.5 hours – extremely inefficient. This invention encapsulates data based on the HL7FHIR standard and encrypts transmission via IPsec-VPN. Consultation requests are completed in 5 minutes, physician review is reduced to 10 minutes, data transmission takes only 1 minute, and real-time opinion synchronization takes 2 minutes, completing the entire process in 18 minutes – a 97% efficiency improvement over the traditional model. The breakpoint resume mechanism and automatic permission revocation function ensure data security and avoid human error, perfectly adapting to the collaborative needs of hierarchical medical systems.

[0078] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A telemedicine system for bone age assessment and growth monitoring in children, characterized in that, include: The user-end data acquisition module is used to acquire children's bone age images and physiological growth parameters. It is compatible with home and hospital imaging equipment and transmits images through medical standard protocols. It integrates physiological parameter monitoring components and transmits data wirelessly. It supports regular automatic collection reminders and emergency manual triggering. Each data item is attached with a unique identifier. The image preprocessing module is connected to the user-end data acquisition module. It optimizes image quality through adaptive filtering and enhancement algorithms, uses a deep learning segmentation model to locate key areas of the epiphysis and extract features, and outputs standardized images. The AI ​​bone age assessment module takes pre-processed images and basic information about the child as input, and uses a clinically trained deep learning model to automatically determine and grade bone age. It supports switching between multiple assessment standards and outputs bone age results, developmental descriptions, and marks abnormal images. The growth and development analysis module compares the matching degree of growth parameters with bone age, calculates growth assessment indicators based on standard growth curves, analyzes the deviation between bone age and actual age, and marks developmental abnormalities. The data storage and traceability module adopts dual storage of local and cloud, and uses encryption technology to ensure data security. It builds a growth profile with complete medical records and supports multi-dimensional retrieval and hierarchical access. The device self-test calibration module works in conjunction with the user-end data acquisition module to ensure acquisition accuracy through hardware self-testing, accuracy verification, and compatibility testing, and provides anomaly alerts and calibration guidance. The abnormal warning module sets multiple thresholds based on growth assessment indicators. Once triggered, it pushes reminders through multiple channels. The three-level warning automatically initiates an emergency response. The human-computer interaction module provides a visual operation interface and information display, supports parameter input, report query, and physician appointment, and enables identity verification and operation log storage for key operations.

2. The telemedicine system for assessment of bone age and growth monitoring in children as claimed in claim 1 wherein, Also includes: The growth trend prediction module works in conjunction with the growth and development analysis module and the data storage module to build a dynamic prediction model based on the child's current growth parameters, genetic height, and gender. It adapts to multiple assessment standards and outputs adult height prediction results and key growth nodes. It is iteratively updated every 3 months with new data. The prediction formula is HA=Hc×(1+ΔBA×k+GR×m+Hg×n), where HA is the predicted adult height, Hc is the current height, ΔBA is the difference between bone age and actual age, k is the bone age influence coefficient (0.06 for males in infancy and 0.10 for puberty, 0.07 for females in infancy and 0.11 for puberty), GR is the annual growth rate, m is the growth rate coefficient (0.03), Hg is the average genetic height of the parents, and n is the genetic influence coefficient (0.25).

3. The telemedicine system for assessment of bone age and growth monitoring in children as claimed in claim 1 wherein, Also includes: The image quality verification module evaluates image quality based on four indicators: sharpness, noise intensity, epiphyseal integrity, and postural conformity. It performs "automatic preprocessing / retake guidance / rejection" operations according to the score and establishes a database of compatible parameters for home and hospital equipment. The quality score formula is Q=α×C-β×N+γ×S+δ×T, where Q is the quality score, α is the sharpness weight (0.4), β is the noise weight (0.3), γ is the integrity weight (0.2), δ is the body position weight (0.1), and C, N, S, and T are all between 0 and 100.

4. The telemedicine system for assessment of bone age and growth monitoring in children as claimed in claim 1 wherein, The multi-center data collaboration module builds a cross-institutional sharing network based on medical data interaction standards. It uses encrypted channels to transmit structured data, including desensitized children's information, image metadata, and AI reports. It implements an application-review-authorization-revocation permission mechanism and supports multidisciplinary collaboration and offline data synchronization.

5. The telemedicine system for children's bone age assessment and growth monitoring according to claim 1, characterized in that, It also includes an intelligent nutrition intervention suggestion module, which links with the growth and development analysis module and a third-party dietary database. Based on children's BMI, body fat percentage, dietary preferences, and activity levels, it generates personalized plans and provides dietary lists, supplement recommendations, or nutritionist consultations according to the level of intervention intensity. The plan is updated monthly based on growth data. The intervention intensity formula is NI=ω×(TW)+θ×(GRs-GR)+φ×(BA-CA), where NI is the intervention intensity. ω represents the weight deviation weight, which is 0.

4. T is the standard weight for age and sex, T is the standard weight for age and sex, W is the actual weight, θ is the growth rate weight 0.3; GRs is the standard growth rate; GR is the actual growth rate, φ is the bone age deviation weight 0.3, BA is the bone age, and CA is the actual age.

6. The telemedicine system for assessment of bone age and growth monitoring in children as claimed in claim 1 wherein, It also includes a tiered parent education and guidance module, which is integrated with the human-computer interaction module and data storage module: it pushes popular science content based on the child's age and the type of abnormality, provides intervention implementation reminders, builds a doctor Q&A community, and calculates the intervention implementation rate through image recognition.

7. The telemedicine system for assessment of bone age and growth monitoring in children as claimed in claim 1 wherein, It also includes a full-cycle device self-test and calibration module, with monthly multi-point calibration for height and weight scales, quarterly dose calibration for X-ray equipment, and regular cleaning reminders for body fat sensors. Calibration records are synchronized to the cloud device health records to reduce data collection errors.

8. The telemedicine system for children's bone age assessment and growth monitoring according to claim 1, characterized in that, It also includes a privacy-in-depth and identity authentication module, which employs three levels of protection: hardware encryption, transmission encryption, and storage encryption (fragmented encryption), implements hierarchical identity authentication, stores access logs through blockchain, and supports emergency response to data breaches.

9. The telemedicine system for children's bone age assessment and growth monitoring according to claim 1, characterized in that, Also includes: In conjunction with the growth and development analysis module and the intelligent nutrition intervention module: a standardized assessment cycle of 1 / 3 / 6 months is set. The effectiveness of the intervention is determined by growth rate, height standard deviation score, BMI deviation, and bone age progression rate. For ineffective cases, the influencing factors are analyzed and suggestions for adjusting the plan are pushed, triggering multidisciplinary consultation.

10. The telemedicine system for children's bone age assessment and growth monitoring according to claim 1, characterized in that, Also includes: Intelligent emergency referral submodule: After a level 3 warning is triggered, the system automatically searches for pediatric institutions within the region and sorts them by distance, number of patients, and rating. Priority appointments are made, and an encrypted referral data package is generated and pushed to the receiving institution, establishing a closed loop of warning - referral - 1-week initial consultation feedback - 1-month follow-up.

Images