Application of Deep Learning-Based Artificial Intelligence Model in Lung Ultrasound for Pediatric Lobar Pneumonia

Application of Deep Learning-Based Artificial Intelligence Model in Lung Ultrasound for Pediatric Lobar Pneumonia | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Application of Deep Learning-Based Artificial Intelligence Model in Lung Ultrasound for Pediatric Lobar Pneumonia Tong Su, Sipeng Tang, Yang Li, Lei Zhong, Yajun Wang, Yingtao Zhang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6410448/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background Pneumonia remains a major contributor to global childhood morbidity and mortality, posing significant public health challenges. Lung ultrasound (LUS) serves as a critical tool for phased assessment of pneumonia progression and guidance of clinical management. This study developed a deep learning artificial intelligence (AI) model (LunNet) to automatically identify and precisely segment lesion characteristics in LUS images, aiming to assist ultrasound physicians in accurate lesion measurement for longitudinal disease monitoring and treatment guidance. Methods We retrospectively analyzed 419 pediatric patients diagnosed with lobar pneumonia (male : female, 199:220; mean age 7.1 ± 3.0 years) who underwent LUS examinations between May 2024 and December 2024. A total of 1,383 images from this cohort were used for LunNet (modified U-Net) development and validation. The model's lesion segmentation performance was evaluated using the dice coefficient and compared with the performance of ultrasound physicians. Results LunNet demonstrated robust performance in automatically identifying and segmenting lung consolidation, B-lines, and pleural effusion, achieving mean dice coefficients of 0.8401 (95% CI: 0.8191–0.8610), 0.8274 (95% CI: 0.7874–0.8673), and 0.8140 (95% CI: 0.7808–0.8472), respectively. The segmentation performance for lung consolidation exhibited marked disparity between junior and senior ultrasound physicians, with mean dice coefficients of 0.6946 (95% CI: 0.6312–0.7581) and 0.9441 (95% CI: 0.9352–0.9530), respectively. Notably, when assisted by LunNet, junior ultrasound physicians exhibited substantial improvement in lung consolidation segmentation, attaining a mean dice coefficient of 0.9221 (95% CI: 0.8191–0.8610), (P < 0.001). In the generalizability validation experiment, LunNet maintained competent performance for lung consolidation segmentation with a Dice coefficient of 0.7773 (95% CI: 0.9108–0.9335). Conclusion The LunNet AI model demonstrates excellent segmentation capabilities for pediatric lobar pneumonia lesions in ultrasound imaging. It effectively assists ultrasound physicians in precise quantification of pathological findings and significantly enhances diagnostic efficiency for novice practitioners. These results underscore LunNet's potential clinical value in supporting diagnosis, longitudinal monitoring, and therapeutic decision-making for lobar pneumonia. Lung ultrasound Artificial intelligence Pediatric pneumonia Deep learning Lobar pneumonia Figures Figure 1 Figure 2 Figure 3 Figure 4 Background Pediatric pneumonia remains a major global threat to child health, characterized by high morbidity and mortality rates [ 1 , 2 ], particularly in resource-limited regions [ 3 ]. Current primary imaging modalities for pneumonia diagnosis include chest X-ray and CT. However, both techniques carry potential radiation risks, making them less suitable for pediatric populations [ 4 ]. In contrast, ultrasound offers a radiation-free alternative with advantages in safety and sensitivity for diagnosing pediatric pneumonia, Its non-invasive nature permits serial examinations, making it particularly suitable for longitudinal disease monitoring and therapeutic guidance in children [ 5 ]. While sonographic diagnosis of pediatric pneumonia currently focuses on identifying consolidation patterns, B-lines, and potential pleural effusion complications, the clinical implementation of LUS remains limited. This disparity primarily stems from its relative novelty as an imaging modality, resulting in insufficient operator expertise and standardized diagnostic criteria. AI offers a promising solution to bridge this clinical gap by assisting image interpretation, enhancing diagnostic accuracy, and accelerating the learning curve for practitioners [ 6 ]. This study therefore aims to develop an enhanced U-Net-based AI model using ultrasound imaging features of lobar pneumonia in children, with the ultimate goals of improving diagnostic efficacy, facilitating clinical adoption of pulmonary ultrasound, and optimizing its therapeutic monitoring capabilities. Methods Patient Selection The inclusion criteria were: (1) pediatric patients who underwent LUS at the The Sixth Affiliated Hospital of Harbin Medical University between May 2024 and December 2024; (2) clinically diagnosed with pneumonia; and (3) radiologically confirmed as having lobar pneumonia. Patients with lesions not reaching the subpleural region were excluded. General clinical data, including age, sex, and imaging findings, were recorded for all enrolled subjects. Image Selection and Quality Control Clinician expertise levels were self-assessed using two predefined categories: (1) junior ultrasound physicians, defined as those with less than six months of LUS experience, and (2) senior ultrasound physicians, defined as those with over five years of LUS experience. A total of 1,383 ultrasound images from 419 pediatric patients meeting the inclusion criteria were retrospectively collected. For LunNet development and evaluation, ultrasound images were acquired by two senior ultrasound physicians using a Philips EPIQ7 ultrasound system equipped with a high-frequency linear probe (eL18-4). These images were divided into non-overlapping training and validation sets at a ratio of 4:1. Test Set A, used to evaluate physician performance, was also acquired by the same senior physicians using identical equipment. To assess model generalizability, Test Set B was collected by different operators using diverse ultrasound systems and probes. The composition of the dataset is detailed in Table 1 . Table 1 Composition of the dataset Dateset Number of Patients Total Images Consolidation Images B-line Images Pleural Effusion Images Training Set 236 867 534 168 165 Validation Set 58 219 134 43 42 Test Set A 25 58 58 0 0 Test Set B 100 239 239 0 0 All datasets, including the training set, were annotated by two senior ultrasound physicians based on medical knowledge and clinical experience. Lesion boundaries were delineated using LabelMe [ 7 ], a robust deep learning image annotation tool, and validated against CT scans to ensure consistency. Model Architecture LunNet adopts a modified U-Net [ 8 , 9 ] encoder-decoder architecture, which incorporates three innovative modules: a BottleNeck for multi-scale feature fusion, a ResNet34-based encoder, and a transposed convolution-driven decoder. The BottleNeck integrates a dual-path feature enhancement mechanism:(1) Dense Atrous Convolution (DAC): This module progressively increases convolutional depth to capture multi-scale receptive field features. (2) Residual Multi-kernel Pooling (RMP): A residual multi-scale pooling mechanism enables parallel attention to lesion features of varying sizes. The synergistic interaction of DAC and RMP significantly enhances the sensitivity and specificity of lesion localization. The encoder inherits the ResNet34 network architecture [ 10 ], utilizing ResBlock modules with shortcut connections to facilitate deep feature extraction. This design not only effectively mitigates the vanishing gradient problem but also allows the network to deepen while approaching optimal performance, thereby significantly accelerating model convergence. The decoder employs transposed convolution for feature map upsampling. Compared to traditional bilinear interpolation, this approach leverages learnable adaptive mappings to better restore fine lesion details. By integrating low-level semantic features from the encoder with high-level semantic information from the decoder via skip connections, the model enhances the characterization of subtle lesions while preserving spatial localization accuracy. Ultimately, a progressive upsampling process generates high-resolution mask maps, demonstrating superior detail reconstruction performance in ultrasound image analysis. Model Training The training set was input into the model for development and training, enabling the model to learn annotated features from the dataset. By integrating the DAC and RMP modules, the model assigns higher weights to lesion regions and their corresponding feature maps [ 11 ], ultimately yielding the AI model LunNet capable of segmenting lung consolidation, B-lines, and pleural effusion. Binary Cross-Entropy Loss (BCELoss) was employed as the loss function, and the model was trained end-to-end. The technical workflow is illustrated in Fig. 1 . Model Validation In our experiment, the dice coefficient was employed as the primary evaluation metric to quantify the similarity between segmentation outputs and reference standards [ 12 ]. The validation set was first processed through LunNet for lesion segmentation on ultrasound images, followed by quantitative validation using the Dice coefficient. To compare LunNet's performance with clinical practitioners, Test Set A was independently analyzed by both a junior ultrasound physician and a senior ultrasound physician for pathological area delineation. When their Dice coefficients for pulmonary consolidation segmentation underperformed compared to LunNet's benchmarks, the ultrasound physicians subsequently utilized LunNet's outputs as reference guidance to reassess lesion boundaries. This iterative process, validated through repeated Dice coefficient measurements, was designed to investigate LunNet's potential clinical utility in enhancing diagnostic workflows. The experimental paradigm specifically examined how AI-driven segmentation could complement and refine human interpretation in real-world ultrasound practice. Additionally, the storage quality of ultrasound images may vary due to differences in examination techniques among ultrasound physicians and the use of diverse ultrasound devices. Such variability could introduce errors in LunNet’s analysis of lesion images, thereby compromising the generalizability of results. To validate this hypothesis and prepare for model optimization, Test Set B—collected under heterogeneous imaging conditions (different operators and equipment)—was processed by LunNet using the same experimental protocol to evaluate its performance. Results Patient Characteristics From May 2024 to December 2024, a total of 450 pediatric patients with lobar pneumonia underwent LUS examinations at our institution. Among these, 31 patients were excluded due to lesions not reaching the subpleural region. Consequently, 419 patients were enrolled in this study. The detailed characteristics are summarized in Table 2 . Table 2 General clinical data of 419 pediatric patients Characteristics Training Group (n = 236) Validation Group (n = 58) Test Group A (n = 25) Test Group B (n = 100) Total (n = 419) Mean Age 7.3±3.1 7.2±2.6 6.7±3.0 6.7±2.8 7.1±3.0 Sex Male 114(27.2%) 27(6.4%) 11(2.6%) 47(11.2%) 199(47.5%) Female 122(29.1%) 31(7.4%) 14(3.3%) 53(12.7%) 220(52.5%) Imaging Results Lobar Pneumonia 236 58 25 100 419 Lesion Segmentation Performance of the Model After sufficient training on 534 ultrasound images of lung consolidation, 168 images of B-lines, and 165 images of pleural effusion from 236 pediatric patients (training set), and validation using 134 images of lung consolidation, 43 images of B-lines, and 42 images of pleural effusion from 58 patients (validation set), LunNet achieved mean dice coefficients of 0.8401 (95% CI: 0.8191–0.8610), 0.8274 (95% CI: 0.7874–0.8673), and 0.8140 (95% CI: 0.7808–0.8472) for lung consolidation, B-lines, and pleural effusion segmentation, respectively. As shown in Fig. 2 , LunNet successfully localized lesion regions with high concordance between algorithmic outputs and ground truth annotations. Performance of Consolidation Segmentation by Ultrasound Physicians and AI-Assisted Physicians As shown in Fig. 3 , the segmentation results of lung consolidation were evaluated using the dice coefficient for ultrasound physicians of different expertise levels and those assisted by LunNet, based on 58 ultrasound images from 25 pediatric patients (Test Set A). The senior ultrasound physicians achieved a mean dice coefficient of 0.9441 (95% CI: 0.9352–0.9530), while junior physicians attained 0.6946 (95% CI: 0.6312–0.7581). LunNet’s prediction accuracy for lung consolidation fell between these two groups, outperforming junior physicians but remaining below senior physicians. Notably, when junior physicians were assisted by LunNet, their mean dice coefficient significantly improved to 0.9221 (95% CI: 0.8191–0.8610), (P < 0.001). Generalization Performance of the Model To validate the external applicability and enhance the credibility of the model’s generalization capability, Test Set B comprising 239 ultrasound images of lung consolidation from 100 pediatric patients which was collected by different operators using diverse imaging devices. As illustrated in Fig. 4 , the segmentation results for Test Set B were evaluated using dice coefficient. LunNet achieved a mean dice coefficient of 0.7773 (95% CI: 0.9108–0.9335), indicating a slight decline in generalization performance compared to internal validation sets but maintaining a high level of accuracy. Discussion Pneumonia remains the leading single infectious cause of mortality among children worldwide [ 13 ]. Current diagnostic approaches for pneumonia primarily rely on physical examinations, imaging studies, and laboratory tests [ 14 – 17 ], among which imaging plays a pivotal role. Chest X-ray and CT are widely used modalities for pneumonia diagnosis. While chest X-ray is rapid and convenient, and CT offers superior resolution and sensitivity, both involve radiation exposure risks, Magnetic Resonance Imaging (MRI), although free of ionizing radiation, is time-consuming and costly [ 4 ]. Ultrasound, in contrast, is radiation-free and repeatable but remains underutilized in clinical practice [ 18 ]. In practice, LUS has demonstrated sufficient diagnostic accuracy for pulmonary diseases through the analysis of various artifacts [ 19 , 20 ]. Specifically, LUS aids in pneumonia diagnosis by detecting subpleural lesions, primarily lung consolidation, in pediatric patients. Experienced ultrasound physicians can monitor disease progression, assess ventilation status, and evaluate therapeutic outcomes by observing dynamic changes in consolidation and air bronchogram signs [ 21 , 22 ]. Furthermore, LUS sensitively identifies pneumonia complications such as pleural effusion and atelectasis [ 23 , 24 ]. Its portability also facilitates follow-up assessments in pneumonia management [ 25 , 26 ]. Accumulated clinical experience in pediatric populations underscores the significance of LUS across different stages of pediatric lobar pneumonia [ 27 – 30 ], establishing it as a reliable diagnostic tool for pediatricians [ 31 , 32 ]. With the increasing application of LUS in pediatric respiratory diseases, the realization of its clinical value faces two critical challenges: the subjective interpretation of ultrasound images, which heavily relies on operator experience, and the efficiency limitations of traditional manual analysis. In this context, AI offers a transformative approach by translating diagnostic expertise into deep learning models for objective segmentation of pathological features. This study not only pioneers an innovative pathway to enhance the standardization and efficiency of LUS diagnosis for pediatric lobar pneumonia but, more importantly, enables precise assessment of lesion extent through objective segmentation. Such advancements provide an objective foundation for evaluating disease progression across stages, guiding therapeutic strategies, and facilitating post-discharge follow-up, thereby demonstrating significant clinical relevance. Previous studies have confirmed that AI-based medical image analysis matches or surpasses human expert performance [ 33 ], with successful implementations in ultrasound-based anatomical structure analysis, including breast and thyroid imaging [ 34 ]. Based on an in-depth analysis of existing lesion classification research, this study explores AI applications in LUS from a novel perspective, with a primary focus on developing a novel deep learning model for objective pathological segmentation. Our investigation specifically targets pediatric patients with lobar pneumonia, as the consolidation morphology during the consolidation phase typically aligns with lobar or segmental pulmonary patterns [ 35 – 37 ], which predominantly occur subpleurally and exhibit excellent visualization in ultrasound imaging [ 38 – 40 ]. The developed model enables precise segmentation of subpleural abnormalities, including tissue-like hepatization of lung consolidation, B-lines, and associated complications such as pleural effusion [ 41 ]. While conventional classification approaches merely serve diagnostic purposes for pneumonia identification, the real-time and repeatable nature of ultrasound highlights its critical clinical value in evaluating pneumonia severity through quantitative measurement of consolidation size. Furthermore, serial comparisons of ultrasonographic measurements during disease progression and treatment provide vital evidence for therapeutic efficacy assessment and clinical decision-making. This underscores the essential requirement for high-precision segmentation in our model to support dynamic monitoring and evidence-based clinical management. Both this study and the automated LUS classification research by Fatima et al. (2024) aim to optimize pediatric respiratory disease diagnosis through AI-enhanced ultrasound analysis, employing deep learning models for medical image processing. However, our investigation presents distinct methodological and clinical advancements. While Fatima et al. primarily focused on multi-category scoring of dynamic LUS video sequences, leveraging temporal information to capture pathological motion patterns [ 42 ], our work addresses the critical challenge of static image segmentation in pediatric lobar pneumonia cases. This technical distinction enables superior spatial resolution in lesion boundary delineation, effectively addressing the limitations of conventional video-based scoring systems in detecting subtle parenchymal abnormalities. The proposed framework provides clinicians with quantitatively precise measurements of consolidation extent through pixel-wise characterization of pulmonary infiltrates, a significant improvement over the regional scoring paradigm. Furthermore, we have empirically validated the clinical utility of this AI system in enhancing diagnostic accuracy among junior clinicians, demonstrating substantially improved operational practicality compared to previous video-dependent approaches in real-world clinical workflows. The accurate identification of pulmonary consolidation plays a pivotal role in the management of pediatric lobar pneumonia. Precise sonographic assessment of consolidation extent by practitioners is crucial for providing clinicians with diagnostically relevant information and therapeutic guidance. In this study, dedicated efforts were directed toward training LunNet for pulmonary consolidation segmentation, thereby capitalizing on ultrasonography's inherent advantages including excellent reproducibility and serial examination comparability. This technical enhancement facilitates objective longitudinal monitoring of pathological changes. To our knowledge, this is the first study to employ AI for segmenting consolidation in pediatric ultrasound images. Our results demonstrate that LunNet achieves precise localization of subpleural lesions, effectively distinguishing rib shadows and other anatomical structures (e.g., heart, liver, and blood vessels) without misregistration. Traditionally, convex probes with larger near- and far-field views, optimized for adults, were used for LUS. In contrast, LunNet was trained on images acquired using a high-frequency linear probe with enhanced near-field resolution, which improves differentiation between the pleural line and adjacent structures (e.g., gastric wall) in pediatric patients. Given children’s thinner chest walls, this approach yields clearer images, minimizing potential misdiagnosis [ 43 , 44 ] while enhancing diagnostic accuracy and efficiency for pediatric pneumonia. LunNet not only localizes lesions accurately but also excels in segmentation. Its outputs reliably identify the pleural line and shred sign at lesion margins. The pleural line, a key diagnostic marker, typically requires subjective evaluation by experienced clinicians, posing challenges for novices [ 45 ]. The shred sign, characterized by irregular hyperechoic interfaces between normal and consolidated lung tissue [ 46 ], serves as a critical boundary marker for measuring consolidation extent. Based on these two aspects, we implemented a series of innovative experimental designs by specifically annotating pleural lines and shred signs in the training set of lung consolidation images. The results demonstrated that nearly all LunNet outputs achieved precise identification of pleural lines and shred signs, with the segmentation performance reaching a Dice coefficient of 0.8401. This procedural innovation significantly enhanced the model's segmentation capability, which not only assists ultrasound physicians in more accurately delineating pathological regions in real clinical scenarios but also provides a valuable methodological framework for subsequent studies. Specifically, our technical approach establishes a reference paradigm for improving computer-aided diagnosis systems in pulmonary ultrasound imaging, while offering clinically interpretable features that align with ultrasound physicians' diagnostic reasoning processes. LunNet demonstrated high accuracy in identifying B-lines and pleural effusion. B-lines manifest as hyperechoic vertical artifacts arising from the pleural line and extending to the bottom of the screen without attenuation in ultrasound imaging. Their clinical significance lies in the positive correlation between their quantity, density, and tendency for coalescence with the severity of pulmonary involvement. Precise segmentation of B-lines enables quantitative assessment of regional lung pathology. Validation studies revealed LunNet achieved a Dice coefficient of 0.8274 for B-line segmentation. Pleural effusion, a critical pneumonia complication resulting from inflammatory pleural involvement, requires prompt diagnosis due to its association with dyspnea and chest tightness in pediatric patients [ 47 ]. Ultrasound, as the primary imaging modality for detecting pleural effusion [ 48 ], is indispensable from diagnosis to clinical management [ 49 ]. Sonographically characterized by anechoic or hypoechoic spaces, accurate delineation of effusion volume and distribution significantly impacts clinical decision-making. LunNet attained a Dice coefficient of 0.8140 in pleural effusion segmentation, demonstrating sufficient reliability for assisting ultrasound physicians in localization and quantification. Notably, LunNet exhibited promising diagnostic potential even with limited training data for these features. The integration of three diagnostic models (lung consolidation, B-lines, pleural effusion) achieved anticipated performance in segmenting fundamental pathological features of pneumonia. Previous studies have demonstrated that certain results obtained through deep learning algorithms for medical image analysis surpass those achieved by professional clinicians [ 50 , 51 ]. To validate this, we compared the segmentation performance between ultrasound physicians and LunNet using Test Set A. The senior ultrasound physician achieved a dice coefficient of 0.9441, while the junior ultrasound physician obtained 0.6946. While LunNet underperformed compared to senior ultrasound physicians, it demonstrated superior diagnostic capabilities over junior practitioners in pulmonary consolidation assessment. To investigate LunNet's potential for enhancing junior ultrasound physicians' performance, a validation study was conducted where these practitioners re-evaluated consolidation boundaries with LunNet assistance. The intervention significantly improved their measurement accuracy, achieving a Dice coefficient of 0.9221. Comparative analysis revealed that initial diagnostic limitations - including consolidation oversight, misclassification errors, and boundary delineation inaccuracies observed in unaided assessments - were effectively resolved through AI assistance. This enhancement enabled junior ultrasound physicians to attain diagnostic proficiency approaching that of senior practitioners. These findings align with recent research by Phung Tran Huy Nhat et al. (2023), where non-expert clinicians improved their diagnostic accuracy from 68.9–82.9% when supported by the RAILUS (Real-time AI-assisted LUS) system for interpreting retrospective ultrasound clips [ 52 ]. Both studies substantiate that AI assistance can significantly enhance diagnostic capabilities for less-experienced practitioners. However, AI is not suitable for independently performing medical image analysis tasks. In clinical practice, AI should operate under physician supervision [ 53 ], with its primary role being to assist ultrasound physicians in rapidly and accurately identifying lesions. By applying deep learning models to the field of LUS, this innovation not only exemplifies interdisciplinary convergence but also demonstrates substantive clinical applicability in real-world diagnostic workflows. The developed LunNet demonstrates significant advantages in addressing the low diagnostic accuracy of junior physicians in pediatric pneumonia, thereby enhancing clinical workflow efficiency. Based on prior research, we propose that LunNet is particularly suitable for junior ultrasound physicians or clinicians lacking ultrasound experience. Under AI assistance, it effectively reduces their learning curve for LUS and improves diagnostic accuracy in clinical applications. Several limitations emerged during the advancement of this study. First, the accuracy and stability of LunNet depend on deep learning-based lesion features. The model exhibited reduced performance in identifying lesions with indistinct boundaries or those obscured by rib shadows. Second, as a single-center study, variations in ultrasound image acquisition—due to differences in devices and operational protocols—introduced potential biases. To address this, we validated LunNet’s performance on Test Set B, revealing a slight decline in generalizability (dice coefficient: 0.7773). Future efforts will focus on multicenter large-scale external validation to enhance universal applicability. We plan to establish detailed image acquisition and procedural guidelines and continuously enrich the training dataset with multi-institutional data. Additionally, the limited dataset for B-lines and pleural effusion impacted model performance. While pleural effusion, appearing as anechoic or hypoechoic dark areas on ultrasound, can be readily identified even by less experienced physicians, LunNet’s segmentation accuracy for these features (dice: 0.8140) fell below expectations. Future work will prioritize expanding the dataset to improve LunNet’s precision for B-lines and pleural effusion, thereby optimizing its diagnostic utility. In summary, the LunNet system developed in this study enhances the clinical utility of LUS in pediatric lobar pneumonia management through its accurate segmentation of pulmonary consolidations, B-lines, and pleural effusions in ultrasonographic images. This advancement enables more precise diagnostic applications, facilitates longitudinal follow-up assessments, and provides valuable guidance for clinical decision-making in pediatric respiratory care. Abbreviations LUS lung ultrasound AI artificial intelligence Declarations Ethics approval and consent to participate This study was performed in accordance with the Declaration of Helsinki. This single-center retrospective study was approved by the Institutional Review Board of The Sixth Affiliated Hospital of Harbin Medical University (IRB No. LC2024-091). According to the requirements of the Ethics Committee of the Sixth Affiliated Hospital of Harbin Medical University: All child participants provided written informed consent to participate. For participants under 8 years of age, written informed consent was obtained from their legal guardians. For participants aged 8 to 18 years, written informed consent was jointly signed by both the participants and their legal guardians. Consent for publication Not applicable. Availability of data and materials The data supporting this study’s findings are available from the corresponding author, Binbin Guo and Yingtao Zhang, upon reasonable request. Competing interests The authors declare no competing interests. The AI algorithms developed in this study are for research purposes only and have no commercial affiliations. Funding None. Authors' contributions TS: Conceptualization, Investigation, Methodology, Software, Validation, Visualization, Writing-original draft, ST: Conceptualization, Formal analysis, Methodology, Software, Visualization, Writing-review & editing, YL: Formal analysis, Investigation, Visualization, Writing-review & editing, LZ: Formal analysis, Investigation, Supervision, Writing-review & editing, YW: Methodology, Supervision, Visualization, Writing-review & editing, YZ: Conceptualization, Methodology, Project administration, Software, Supervision, Writing-review & editing, BG: Conceptualization, Data curation, Methodology, Project administration, Supervision, Validation, Writing-review & editing. Acknowledgements The authors sincerely thank the ultrasound physicians from the Sixth Affiliated Hospital of Harbin Medical University, Xiaoya Chen, Qiang Qiao, Yao Zhang, and Xinhong Yu, their expertise in ultrasound image acquisition and meticulous execution of the research protocol significantly facilitated the completion of this work. And we appreciate all the doctors in the Department of Pediatrics and Imaging of the Sixth Affiliated Hospital of Harbin Medical University for providing clinical information support. References Walker CLF, Rudan I, Liu L, Nair H, Theodoratou E, Bhutta ZA, et al. Global burden of childhood pneumonia and diarrhoea. The Lancet 2013;381:1405-1416 Smith DC, Kuckel DP, Recidoro AM. Community-Acquired Pneumonia in Children: Rapid Evidence Review. American Family Physician 2021;104:618-625 Marangu D, Zar HJ. Childhood pneumonia in low-and-middle-income countries: An update. Paediatric Respiratory Reviews 2019;32:3-9 Alexopoulou E, Prountzos S, Raissaki M, Mazioti A, Caro-Dominguez P, Hirsch FW, et al. Imaging of Acute Complications of Community-Acquired Pneumonia in the Paediatric Population—From Chest Radiography to MRI. Children 2024;11:122 Picano E, Scali MC, Ciampi Q, Lichtenstein D. Lung Ultrasound for the Cardiologist. JACC: Cardiovascular Imaging 2018;11:1692-1705 Akkus Z, Cai J, Boonrod A, Zeinoddini A, Weston AD, Philbrick KA, et al. A Survey of Deep-Learning Applications in Ultrasound: Artificial Intelligence–Powered Ultrasound for Improving Clinical Workflow. Journal of the American College of Radiology 2019;16:1318-1328 Xiao Y, Liang Q, Zhou L, He X, Lv L, Chen J, et al. Construction of a new automatic grading system for jaw bone mineral density level based on deep learning using cone beam computed tomography. Scientific Reports 2022;12:12841 Valanarasu JMJ, Sindagi VA, Hacihaliloglu I, Patel VM. KiU-Net: Overcomplete Convolutional Architectures for Biomedical Image and Volumetric Segmentation. IEEE Transactions on Medical Imaging 2022;41:965-976 Zunair H, Ben Hamza A. Sharp U-Net: Depthwise convolutional network for biomedical image segmentation. Computers in Biology and Medicine 2021;136:104699 Zhuang Q, Gan S, Zhang L. Human-computer interaction based health diagnostics using ResNet34 for tongue image classification. Computer Methods and Programs in Biomedicine 2022;226:107096 Gu Z, Cheng J, Fu H, Zhou K, Hao H, Zhao Y, et al. CE-Net: Context Encoder Network for 2D Medical Image Segmentation. IEEE Transactions on Medical Imaging 2019;38:2281-2292 Anuar N, Md Sultan AB. Validate Conference Paper using Dice Coefficient. Computer and Information Science 2010;3:16-28 Troeger C, Blacker B, Khalil IA, Rao PC, Cao J, Zimsen SRM, et al. Estimates of the global, regional, and national morbidity, mortality, and aetiologies of lower respiratory infections in 195 countries, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. The Lancet Infectious Diseases 2018;18:1191-1210 Smith DC, Kuckel DP, Recidoro AM. Community-Acquired Pneumonia in Children: Rapid Evidence Review. Am Fam Physician 2021;104:618-625 Hunton R. Updated concepts in the diagnosis and management of community-acquired pneumonia. Jaapa 2019;32:18-23 Cunha BA. The atypical pneumonias: clinical diagnosis and importance. Clinical Microbiology and Infection 2006;12:12-24 Miyashita N. Atypical pneumonia: Pathophysiology, diagnosis, and treatment. Respiratory Investigation 2022;60:56-67 Principi N, Esposito A, Giannitto C, Esposito S. Lung ultrasonography to diagnose community-acquired pneumonia in children. BMC Pulmonary Medicine 2017;17:212 Sperandeo M, Rotondo A, Guglielmi G, Catalano D, Feragalli B, Trovato GM. Transthoracic ultrasound in the assessment of pleural and pulmonary diseases: use and limitations. La radiologia medica 2014;119:729-740 Turner JP, Dankoff J. Thoracic Ultrasound. Emergency Medicine Clinics of North America 2012;30:451-473 Ho M-C, Ker C-R, Hsu J-H, Wu J-R, Dai Z-K, Chen IC. Usefulness of Lung Ultrasound in the Diagnosis of Community-acquired Pneumonia in Children. Pediatrics & Neonatology 2015;56:40-45 Fitzgerald DB, Blakey JD, Joshi P, Kuok YJ, Lee YCG, Thomas R. Ultrasound Clues in Lobar Pneumonia. Chest 2022;161:e59-e62 Light RW. Pleural Effusion. New England Journal of Medicine 2002;346:1971-1977 Liu J, Chen S-W, Liu F, Li Q-P, Kong X-Y, Feng Z-C. The Diagnosis of Neonatal Pulmonary Atelectasis Using Lung Ultrasonography. Chest 2015;147:1013-1019 Chidini G, Raimondi F. Lung ultrasound for the sick child: less harm and more information than a radiograph. European Journal of Pediatrics 2023;183:1079-1089 Beshara M, Bittner EA, Goffi A, Berra L, Chang MG. Nuts and bolts of lung ultrasound: utility, scanning techniques, protocols, and findings in common pathologies. Critical Care 2024;28:328 Pereda MA, Chavez MA, Hooper-Miele CC, Gilman RH, Steinhoff MC, Ellington LE, et al. Lung Ultrasound for the Diagnosis of Pneumonia in Children: A Meta-analysis. Pediatrics 2015;135:714-722 Llamas-Álvarez AM, Tenza-Lozano EM, Latour-Pérez J. Accuracy of Lung Ultrasonography in the Diagnosis of Pneumonia in Adults. Chest 2017;151:374-382 Heuvelings CC, Bélard S, Familusi MA, Spijker R, Grobusch MP, Zar HJ. Chest ultrasound for the diagnosis of paediatric pulmonary diseases: a systematic review and meta-analysis of diagnostic test accuracy. British Medical Bulletin 2019;129:35-51 Trinavarat P, Riccabona M. Potential of ultrasound in the pediatric chest. European Journal of Radiology 2014;83:1507-1518 Khan U, Afrakhteh S, Mento F, Fatima N, De Rosa L, Custode LL, et al. Benchmark methodological approach for the application of artificial intelligence to lung ultrasound data from COVID-19 patients: From frame to prognostic-level. Ultrasonics 2023;132:106994 Cattarossi L. Lung ultrasound: its role in neonatology and pediatrics. Early Human Development 2013;89:S17-S19 Chen X, Wang X, Zhang K, Fung K-M, Thai TC, Moore K, et al. Recent advances and clinical applications of deep learning in medical image analysis. Medical Image Analysis 2022;79:102444 Shen Y-T, Chen L, Yue W-W, Xu H-X. Artificial intelligence in ultrasound. European Journal of Radiology 2021;139:109717 Hammerschmidt DE. About the cover illustration. Journal of Laboratory and Clinical Medicine 2004;143:327 Daugherty EA. Lobar Pneumonia. Medical Clinics of North America 1947;31:1432-1441 Zinserling VA, Swistunov VV, Botvinkin AD, Stepanenko LA, Makarova AE. Lobar (croupous) pneumonia: old and new data. Infection 2021;50:235-242 Reynolds JH, McDonald G, Alton H, Gordon SB. Pneumonia in the immunocompetent patient. The British Journal of Radiology 2010;83:998-1009 Basse P, Grégory J, Lavenne R, Foucrier A. Right pan-lobar pneumonia due to Streptococcus pneumoniae. Intensive Care Medicine 2022;48:1647-1647 Malla D, Rathi V, Gomber S, Upreti L. Can lung ultrasound differentiate between bacterial and viral pneumonia in children? Journal of Clinical Ultrasound 2020;49:91-100 Botana Rial M, Pérez Pallarés J, Cases Viedma E, López González FJ, Porcel JM, Rodríguez M, et al. Diagnosis and Treatment of Pleural Effusion. Recommendations of the Spanish Society of Pulmonology and Thoracic Surgery. Update 2022. Archivos de Bronconeumología 2023;59:27-35 Fatima N, Khan U, Han X, Zannin E, Rigotti C, Cattaneo F, et al. Deep learning approaches for automated classification of neonatal lung ultrasound with assessment of human-to-AI interrater agreement. Computers in Biology and Medicine 2024;183:109315 Lichtenstein D. Lung ultrasound in the critically ill. Current Opinion in Critical Care 2014;20:315-322 Evertsson M, Graneli C, Vernersson A, Wiaczek O, Hagelsteen K, Erlöv T, et al. Design of a Pediatric Rectal Ultrasound Probe Intended for Ultra-High Frequency Ultrasound Diagnostics. Diagnostics 2023;13:1667 Chen J, Li J, He C, Li W, Li Q. Automated Pleural Line Detection Based on Radon Transform Using Ultrasound. Ultrasonic Imaging 2020;43:19-28 Muñoz Moreno JF, Rubio Prieto E, Magro Martín MÁ. Lung Ultrasound in ARDS: B-lines Pattern and Shred Sign. Archivos de Bronconeumología 2024;60：180 Beaudoin S, Gonzalez AV. Evaluation of the patient with pleural effusion. Canadian Medical Association Journal 2018;190:E291-E295 Mayo PH, Copetti R, Feller-Kopman D, Mathis G, Maury E, Mongodi S, et al. Thoracic ultrasonography: a narrative review. Intensive Care Medicine 2019;45:1200-1211 Brogi E, Gargani L, Bignami E, Barbariol F, Marra A, Forfori F, et al. Thoracic ultrasound for pleural effusion in the intensive care unit: a narrative review from diagnosis to treatment. Critical Care 2017;21:325 Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, et al. A survey on deep learning in medical image analysis. Medical Image Analysis 2017;42:60-88 Lee J-G, Jun S, Cho Y-W, Lee H, Kim GB, Seo JB, et al. Deep Learning in Medical Imaging: General Overview. Korean Journal of Radiology 2017;18:570 Nhat PTH, Van Hao N, Tho PV, Kerdegari H, Pisani L, Thu LNM, et al. Clinical benefit of AI-assisted lung ultrasound in a resource-limited intensive care unit. Critical Care 2023;27:257 Peng S, Liu Y, Lv W, Liu L, Zhou Q, Yang H, et al. Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre diagnostic study. The Lancet Digital Health 2021;3:e250-e259 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 08 Jun, 2025 Reviewers agreed at journal 23 Apr, 2025 Reviewers agreed at journal 23 Apr, 2025 Reviewers invited by journal 22 Apr, 2025 Editor assigned by journal 22 Apr, 2025 Submission checks completed at journal 21 Apr, 2025 First submitted to journal 21 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6410448","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":447043509,"identity":"8eeaaf51-715f-4ee8-82dc-65e22bac8b83","order_by":0,"name":"Tong Su","email":"","orcid":"","institution":"the Sixth Affiliated Hospital of Harbin Medical University, Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Tong","middleName":"","lastName":"Su","suffix":""},{"id":447043510,"identity":"192d1671-66a1-46f0-a341-937a8a68fd68","order_by":1,"name":"Sipeng Tang","email":"","orcid":"","institution":"Harbin Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Sipeng","middleName":"","lastName":"Tang","suffix":""},{"id":447043514,"identity":"014fe452-eef3-465f-a551-5e76733a8e3e","order_by":2,"name":"Yang Li","email":"","orcid":"","institution":"the Sixth Affiliated Hospital of Harbin Medical University, Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Li","suffix":""},{"id":447043517,"identity":"b695def3-4038-4e92-b0ab-b7acd071446d","order_by":3,"name":"Lei Zhong","email":"","orcid":"","institution":"the Sixth Affiliated Hospital of Harbin Medical University, Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Zhong","suffix":""},{"id":447043519,"identity":"5f9742b2-a03c-4f58-8e2c-ed10f23408b9","order_by":4,"name":"Yajun Wang","email":"","orcid":"","institution":"the Sixth Affiliated Hospital of Harbin Medical University, Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yajun","middleName":"","lastName":"Wang","suffix":""},{"id":447043521,"identity":"4e467d61-6cd3-4411-8a7a-a704c5e6cf08","order_by":5,"name":"Yingtao Zhang","email":"","orcid":"","institution":"Harbin Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Yingtao","middleName":"","lastName":"Zhang","suffix":""},{"id":447043523,"identity":"6f04e4af-60c5-452c-ae0d-6d3928a6e654","order_by":6,"name":"Binbin Guo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2UlEQVRIiWNgGAWjYBACfmbmg4//VNgkMEiA+cyEtUi2tyUb8JxJI0GLQc8ZMwnetsOkaJFIMJCQYDufZz67O02CocI6sYH97AG8WswlEhIMDHhuF8vcObtNguFMemIDT14CXi2WMxIOJCRI3E6cIZG7TYKx7XBigwSPAX6H3UhsOHDA4BxUyz9itJw5zNjYkHAAqqWBCC3AQGZmZjiQnDhD5uxmi4Rj6cZtPDn4tfAz83//zfjPLnGGdO/GGx9qrGX72c/g14IKEoCYjQT1o2AUjIJRMApwAAA0GkeVlgJDwgAAAABJRU5ErkJggg==","orcid":"","institution":"the Sixth Affiliated Hospital of Harbin Medical University, Harbin Medical University","correspondingAuthor":true,"prefix":"","firstName":"Binbin","middleName":"","lastName":"Guo","suffix":""}],"badges":[],"createdAt":"2025-04-09 09:38:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6410448/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6410448/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82120754,"identity":"4a59f310-41f1-451c-a5ef-d5d58fea33fd","added_by":"auto","created_at":"2025-05-07 03:19:44","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":213753,"visible":true,"origin":"","legend":"\u003cp\u003eTechnical methodology\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6410448/v1/f491297dd1e1ffa743058742.jpeg"},{"id":82119833,"identity":"e45f47c8-21fe-4a0f-96a5-5dff3fc751f1","added_by":"auto","created_at":"2025-05-07 03:11:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":194327,"visible":true,"origin":"","legend":"\u003cp\u003eSegmentation results of LunNet for different lesions.\u003cstrong\u003e \u003c/strong\u003eEach result is composed of 3 rows. First row shows the original ultrasound images, and ground truth lesion mask is delineated by two senior ultrasound physicians as validation criteria in second row. Automatic description by LunNet is provided in third row.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6410448/v1/509773b5a50b439c50b229db.png"},{"id":82119836,"identity":"747c2e93-3d7d-430b-a397-0fb65ae99708","added_by":"auto","created_at":"2025-05-07 03:11:45","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":175010,"visible":true,"origin":"","legend":"\u003cp\u003eSegmentation results of ultrasound physicians for lung consolidation.\u003cstrong\u003e \u003c/strong\u003eEach result is composed of 5 rows. First row shows the original ultrasound images, and ground truth lesion mask is delineated by two senior ultrasound physicians as validation criteria in second row. Ground truth lesion mask is delineated by senior ultrasound physicians in third row. Ground truth lesion mask is delineated by junior ultrasound physicians in fourth row. Ground truth lesion mask is delineated by junior ultrasound physicians assisted by LunNet in fifth row.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6410448/v1/7674278c336302ddb051adb4.png"},{"id":82119840,"identity":"1567dbc1-b527-45d4-a8c2-83057b0d51c0","added_by":"auto","created_at":"2025-05-07 03:11:45","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":140770,"visible":true,"origin":"","legend":"\u003cp\u003eSegmentation results of LunNet on lung consolidation images of varying quality.\u003cstrong\u003e \u003c/strong\u003eEach result is composed of 3 rows. First row shows the original ultrasound images of varying quality, and ground truth lesion mask is delineated by two senior ultrasound physicians as validation criteria in second row. Automatic description by LunNet is provided in third row.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6410448/v1/2cc813ed91a7f08844cd6bfe.png"},{"id":82123798,"identity":"6b908036-ded0-4727-a43e-18e3d6cdfa74","added_by":"auto","created_at":"2025-05-07 03:35:45","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1550229,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6410448/v1/1dfd82c4-8205-4604-8de9-4416d145bf1e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Application of Deep Learning-Based Artificial Intelligence Model in Lung Ultrasound for Pediatric Lobar Pneumonia","fulltext":[{"header":"Background","content":"\u003cp\u003ePediatric pneumonia remains a major global threat to child health, characterized by high morbidity and mortality rates [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], particularly in resource-limited regions [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Current primary imaging modalities for pneumonia diagnosis include chest X-ray and CT. However, both techniques carry potential radiation risks, making them less suitable for pediatric populations [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In contrast, ultrasound offers a radiation-free alternative with advantages in safety and sensitivity for diagnosing pediatric pneumonia, Its non-invasive nature permits serial examinations, making it particularly suitable for longitudinal disease monitoring and therapeutic guidance in children [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. While sonographic diagnosis of pediatric pneumonia currently focuses on identifying consolidation patterns, B-lines, and potential pleural effusion complications, the clinical implementation of LUS remains limited. This disparity primarily stems from its relative novelty as an imaging modality, resulting in insufficient operator expertise and standardized diagnostic criteria. AI offers a promising solution to bridge this clinical gap by assisting image interpretation, enhancing diagnostic accuracy, and accelerating the learning curve for practitioners [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. This study therefore aims to develop an enhanced U-Net-based AI model using ultrasound imaging features of lobar pneumonia in children, with the ultimate goals of improving diagnostic efficacy, facilitating clinical adoption of pulmonary ultrasound, and optimizing its therapeutic monitoring capabilities.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatient Selection\u003c/h2\u003e \u003cp\u003eThe inclusion criteria were: (1) pediatric patients who underwent LUS at the The Sixth Affiliated Hospital of Harbin Medical University between May 2024 and December 2024; (2) clinically diagnosed with pneumonia; and (3) radiologically confirmed as having lobar pneumonia. Patients with lesions not reaching the subpleural region were excluded. General clinical data, including age, sex, and imaging findings, were recorded for all enrolled subjects.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eImage Selection and Quality Control\u003c/h3\u003e\n\u003cp\u003eClinician expertise levels were self-assessed using two predefined categories: (1) junior ultrasound physicians, defined as those with less than six months of LUS experience, and (2) senior ultrasound physicians, defined as those with over five years of LUS experience.\u003c/p\u003e \u003cp\u003eA total of 1,383 ultrasound images from 419 pediatric patients meeting the inclusion criteria were retrospectively collected. For LunNet development and evaluation, ultrasound images were acquired by two senior ultrasound physicians using a Philips EPIQ7 ultrasound system equipped with a high-frequency linear probe (eL18-4). These images were divided into non-overlapping training and validation sets at a ratio of 4:1. Test Set A, used to evaluate physician performance, was also acquired by the same senior physicians using identical equipment. To assess model generalizability, Test Set B was collected by different operators using diverse ultrasound systems and probes. The composition of the dataset is detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComposition of the dataset\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDateset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of Patients\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTotal Images\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eConsolidation Images\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eB-line Images\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePleural Effusion Images\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTraining Set\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e236\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e867\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e534\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e165\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eValidation Set\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e219\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e134\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTest Set A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTest Set B\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e239\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e239\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAll datasets, including the training set, were annotated by two senior ultrasound physicians based on medical knowledge and clinical experience. Lesion boundaries were delineated using LabelMe [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], a robust deep learning image annotation tool, and validated against CT scans to ensure consistency.\u003c/p\u003e\n\u003ch3\u003eModel Architecture\u003c/h3\u003e\n\u003cp\u003eLunNet adopts a modified U-Net [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] encoder-decoder architecture, which incorporates three innovative modules: a BottleNeck for multi-scale feature fusion, a ResNet34-based encoder, and a transposed convolution-driven decoder.\u003c/p\u003e \u003cp\u003eThe BottleNeck integrates a dual-path feature enhancement mechanism:(1) Dense Atrous Convolution (DAC): This module progressively increases convolutional depth to capture multi-scale receptive field features. (2) Residual Multi-kernel Pooling (RMP): A residual multi-scale pooling mechanism enables parallel attention to lesion features of varying sizes. The synergistic interaction of DAC and RMP significantly enhances the sensitivity and specificity of lesion localization.\u003c/p\u003e \u003cp\u003eThe encoder inherits the ResNet34 network architecture [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], utilizing ResBlock modules with shortcut connections to facilitate deep feature extraction. This design not only effectively mitigates the vanishing gradient problem but also allows the network to deepen while approaching optimal performance, thereby significantly accelerating model convergence.\u003c/p\u003e \u003cp\u003eThe decoder employs transposed convolution for feature map upsampling. Compared to traditional bilinear interpolation, this approach leverages learnable adaptive mappings to better restore fine lesion details. By integrating low-level semantic features from the encoder with high-level semantic information from the decoder via skip connections, the model enhances the characterization of subtle lesions while preserving spatial localization accuracy. Ultimately, a progressive upsampling process generates high-resolution mask maps, demonstrating superior detail reconstruction performance in ultrasound image analysis.\u003c/p\u003e\n\u003ch3\u003eModel Training\u003c/h3\u003e\n\u003cp\u003eThe training set was input into the model for development and training, enabling the model to learn annotated features from the dataset. By integrating the DAC and RMP modules, the model assigns higher weights to lesion regions and their corresponding feature maps [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], ultimately yielding the AI model LunNet capable of segmenting lung consolidation, B-lines, and pleural effusion. Binary Cross-Entropy Loss (BCELoss) was employed as the loss function, and the model was trained end-to-end. The technical workflow is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eModel Validation\u003c/h3\u003e\n\u003cp\u003eIn our experiment, the dice coefficient was employed as the primary evaluation metric to quantify the similarity between segmentation outputs and reference standards [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The validation set was first processed through LunNet for lesion segmentation on ultrasound images, followed by quantitative validation using the Dice coefficient.\u003c/p\u003e \u003cp\u003eTo compare LunNet's performance with clinical practitioners, Test Set A was independently analyzed by both a junior ultrasound physician and a senior ultrasound physician for pathological area delineation. When their Dice coefficients for pulmonary consolidation segmentation underperformed compared to LunNet's benchmarks, the ultrasound physicians subsequently utilized LunNet's outputs as reference guidance to reassess lesion boundaries. This iterative process, validated through repeated Dice coefficient measurements, was designed to investigate LunNet's potential clinical utility in enhancing diagnostic workflows. The experimental paradigm specifically examined how AI-driven segmentation could complement and refine human interpretation in real-world ultrasound practice.\u003c/p\u003e \u003cp\u003eAdditionally, the storage quality of ultrasound images may vary due to differences in examination techniques among ultrasound physicians and the use of diverse ultrasound devices. Such variability could introduce errors in LunNet\u0026rsquo;s analysis of lesion images, thereby compromising the generalizability of results. To validate this hypothesis and prepare for model optimization, Test Set B\u0026mdash;collected under heterogeneous imaging conditions (different operators and equipment)\u0026mdash;was processed by LunNet using the same experimental protocol to evaluate its performance.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003ePatient Characteristics\u003c/h2\u003e \u003cp\u003eFrom May 2024 to December 2024, a total of 450 pediatric patients with lobar pneumonia underwent LUS examinations at our institution. Among these, 31 patients were excluded due to lesions not reaching the subpleural region. Consequently, 419 patients were enrolled in this study. The detailed characteristics are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGeneral clinical data of 419 pediatric patients\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining Group\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;236)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation Group\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;58)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTest Group A\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;25)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTest Group B\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;100)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;419)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMean Age\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.3\u0026plusmn;3.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.2\u0026plusmn;2.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.7\u0026plusmn;3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.7\u0026plusmn;2.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7.1\u0026plusmn;3.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e114(27.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27(6.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11(2.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e47(11.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e199(47.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e122(29.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31(7.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14(3.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e53(12.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e220(52.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eImaging Results\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLobar Pneumonia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e236\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e419\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eLesion Segmentation Performance of the Model\u003c/h3\u003e\n\u003cp\u003eAfter sufficient training on 534 ultrasound images of lung consolidation, 168 images of B-lines, and 165 images of pleural effusion from 236 pediatric patients (training set), and validation using 134 images of lung consolidation, 43 images of B-lines, and 42 images of pleural effusion from 58 patients (validation set), LunNet achieved mean dice coefficients of 0.8401 (95% CI: 0.8191\u0026ndash;0.8610), 0.8274 (95% CI: 0.7874\u0026ndash;0.8673), and 0.8140 (95% CI: 0.7808\u0026ndash;0.8472) for lung consolidation, B-lines, and pleural effusion segmentation, respectively. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, LunNet successfully localized lesion regions with high concordance between algorithmic outputs and ground truth annotations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePerformance of Consolidation Segmentation by Ultrasound Physicians and AI-Assisted Physicians\u003c/h2\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, the segmentation results of lung consolidation were evaluated using the dice coefficient for ultrasound physicians of different expertise levels and those assisted by LunNet, based on 58 ultrasound images from 25 pediatric patients (Test Set A). The senior ultrasound physicians achieved a mean dice coefficient of 0.9441 (95% CI: 0.9352\u0026ndash;0.9530), while junior physicians attained 0.6946 (95% CI: 0.6312\u0026ndash;0.7581). LunNet\u0026rsquo;s prediction accuracy for lung consolidation fell between these two groups, outperforming junior physicians but remaining below senior physicians. Notably, when junior physicians were assisted by LunNet, their mean dice coefficient significantly improved to 0.9221 (95% CI: 0.8191\u0026ndash;0.8610), (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eGeneralization Performance of the Model\u003c/h2\u003e \u003cp\u003eTo validate the external applicability and enhance the credibility of the model\u0026rsquo;s generalization capability, Test Set B comprising 239 ultrasound images of lung consolidation from 100 pediatric patients which was collected by different operators using diverse imaging devices. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the segmentation results for Test Set B were evaluated using dice coefficient. LunNet achieved a mean dice coefficient of 0.7773 (95% CI: 0.9108\u0026ndash;0.9335), indicating a slight decline in generalization performance compared to internal validation sets but maintaining a high level of accuracy.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003ePneumonia remains the leading single infectious cause of mortality among children worldwide [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Current diagnostic approaches for pneumonia primarily rely on physical examinations, imaging studies, and laboratory tests [\u003cspan additionalcitationids=\"CR15 CR16\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], among which imaging plays a pivotal role. Chest X-ray and CT are widely used modalities for pneumonia diagnosis. While chest X-ray is rapid and convenient, and CT offers superior resolution and sensitivity, both involve radiation exposure risks, Magnetic Resonance Imaging (MRI), although free of ionizing radiation, is time-consuming and costly [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Ultrasound, in contrast, is radiation-free and repeatable but remains underutilized in clinical practice [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn practice, LUS has demonstrated sufficient diagnostic accuracy for pulmonary diseases through the analysis of various artifacts [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Specifically, LUS aids in pneumonia diagnosis by detecting subpleural lesions, primarily lung consolidation, in pediatric patients. Experienced ultrasound physicians can monitor disease progression, assess ventilation status, and evaluate therapeutic outcomes by observing dynamic changes in consolidation and air bronchogram signs [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Furthermore, LUS sensitively identifies pneumonia complications such as pleural effusion and atelectasis [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Its portability also facilitates follow-up assessments in pneumonia management [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Accumulated clinical experience in pediatric populations underscores the significance of LUS across different stages of pediatric lobar pneumonia [\u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], establishing it as a reliable diagnostic tool for pediatricians [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWith the increasing application of LUS in pediatric respiratory diseases, the realization of its clinical value faces two critical challenges: the subjective interpretation of ultrasound images, which heavily relies on operator experience, and the efficiency limitations of traditional manual analysis. In this context, AI offers a transformative approach by translating diagnostic expertise into deep learning models for objective segmentation of pathological features. This study not only pioneers an innovative pathway to enhance the standardization and efficiency of LUS diagnosis for pediatric lobar pneumonia but, more importantly, enables precise assessment of lesion extent through objective segmentation. Such advancements provide an objective foundation for evaluating disease progression across stages, guiding therapeutic strategies, and facilitating post-discharge follow-up, thereby demonstrating significant clinical relevance. Previous studies have confirmed that AI-based medical image analysis matches or surpasses human expert performance [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], with successful implementations in ultrasound-based anatomical structure analysis, including breast and thyroid imaging [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Based on an in-depth analysis of existing lesion classification research, this study explores AI applications in LUS from a novel perspective, with a primary focus on developing a novel deep learning model for objective pathological segmentation. Our investigation specifically targets pediatric patients with lobar pneumonia, as the consolidation morphology during the consolidation phase typically aligns with lobar or segmental pulmonary patterns [\u003cspan additionalcitationids=\"CR36\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], which predominantly occur subpleurally and exhibit excellent visualization in ultrasound imaging [\u003cspan additionalcitationids=\"CR39\" citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. The developed model enables precise segmentation of subpleural abnormalities, including tissue-like hepatization of lung consolidation, B-lines, and associated complications such as pleural effusion [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. While conventional classification approaches merely serve diagnostic purposes for pneumonia identification, the real-time and repeatable nature of ultrasound highlights its critical clinical value in evaluating pneumonia severity through quantitative measurement of consolidation size. Furthermore, serial comparisons of ultrasonographic measurements during disease progression and treatment provide vital evidence for therapeutic efficacy assessment and clinical decision-making. This underscores the essential requirement for high-precision segmentation in our model to support dynamic monitoring and evidence-based clinical management.\u003c/p\u003e \u003cp\u003eBoth this study and the automated LUS classification research by Fatima et al. (2024) aim to optimize pediatric respiratory disease diagnosis through AI-enhanced ultrasound analysis, employing deep learning models for medical image processing. However, our investigation presents distinct methodological and clinical advancements. While Fatima et al. primarily focused on multi-category scoring of dynamic LUS video sequences, leveraging temporal information to capture pathological motion patterns [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e], our work addresses the critical challenge of static image segmentation in pediatric lobar pneumonia cases. This technical distinction enables superior spatial resolution in lesion boundary delineation, effectively addressing the limitations of conventional video-based scoring systems in detecting subtle parenchymal abnormalities. The proposed framework provides clinicians with quantitatively precise measurements of consolidation extent through pixel-wise characterization of pulmonary infiltrates, a significant improvement over the regional scoring paradigm. Furthermore, we have empirically validated the clinical utility of this AI system in enhancing diagnostic accuracy among junior clinicians, demonstrating substantially improved operational practicality compared to previous video-dependent approaches in real-world clinical workflows.\u003c/p\u003e \u003cp\u003eThe accurate identification of pulmonary consolidation plays a pivotal role in the management of pediatric lobar pneumonia. Precise sonographic assessment of consolidation extent by practitioners is crucial for providing clinicians with diagnostically relevant information and therapeutic guidance. In this study, dedicated efforts were directed toward training LunNet for pulmonary consolidation segmentation, thereby capitalizing on ultrasonography's inherent advantages including excellent reproducibility and serial examination comparability. This technical enhancement facilitates objective longitudinal monitoring of pathological changes. To our knowledge, this is the first study to employ AI for segmenting consolidation in pediatric ultrasound images. Our results demonstrate that LunNet achieves precise localization of subpleural lesions, effectively distinguishing rib shadows and other anatomical structures (e.g., heart, liver, and blood vessels) without misregistration. Traditionally, convex probes with larger near- and far-field views, optimized for adults, were used for LUS. In contrast, LunNet was trained on images acquired using a high-frequency linear probe with enhanced near-field resolution, which improves differentiation between the pleural line and adjacent structures (e.g., gastric wall) in pediatric patients. Given children\u0026rsquo;s thinner chest walls, this approach yields clearer images, minimizing potential misdiagnosis [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e] while enhancing diagnostic accuracy and efficiency for pediatric pneumonia.\u003c/p\u003e \u003cp\u003eLunNet not only localizes lesions accurately but also excels in segmentation. Its outputs reliably identify the pleural line and shred sign at lesion margins. The pleural line, a key diagnostic marker, typically requires subjective evaluation by experienced clinicians, posing challenges for novices [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. The shred sign, characterized by irregular hyperechoic interfaces between normal and consolidated lung tissue [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e], serves as a critical boundary marker for measuring consolidation extent. Based on these two aspects, we implemented a series of innovative experimental designs by specifically annotating pleural lines and shred signs in the training set of lung consolidation images. The results demonstrated that nearly all LunNet outputs achieved precise identification of pleural lines and shred signs, with the segmentation performance reaching a Dice coefficient of 0.8401. This procedural innovation significantly enhanced the model's segmentation capability, which not only assists ultrasound physicians in more accurately delineating pathological regions in real clinical scenarios but also provides a valuable methodological framework for subsequent studies. Specifically, our technical approach establishes a reference paradigm for improving computer-aided diagnosis systems in pulmonary ultrasound imaging, while offering clinically interpretable features that align with ultrasound physicians' diagnostic reasoning processes.\u003c/p\u003e \u003cp\u003eLunNet demonstrated high accuracy in identifying B-lines and pleural effusion. B-lines manifest as hyperechoic vertical artifacts arising from the pleural line and extending to the bottom of the screen without attenuation in ultrasound imaging. Their clinical significance lies in the positive correlation between their quantity, density, and tendency for coalescence with the severity of pulmonary involvement. Precise segmentation of B-lines enables quantitative assessment of regional lung pathology. Validation studies revealed LunNet achieved a Dice coefficient of 0.8274 for B-line segmentation. Pleural effusion, a critical pneumonia complication resulting from inflammatory pleural involvement, requires prompt diagnosis due to its association with dyspnea and chest tightness in pediatric patients [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. Ultrasound, as the primary imaging modality for detecting pleural effusion [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e], is indispensable from diagnosis to clinical management [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Sonographically characterized by anechoic or hypoechoic spaces, accurate delineation of effusion volume and distribution significantly impacts clinical decision-making. LunNet attained a Dice coefficient of 0.8140 in pleural effusion segmentation, demonstrating sufficient reliability for assisting ultrasound physicians in localization and quantification. Notably, LunNet exhibited promising diagnostic potential even with limited training data for these features. The integration of three diagnostic models (lung consolidation, B-lines, pleural effusion) achieved anticipated performance in segmenting fundamental pathological features of pneumonia.\u003c/p\u003e \u003cp\u003ePrevious studies have demonstrated that certain results obtained through deep learning algorithms for medical image analysis surpass those achieved by professional clinicians [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. To validate this, we compared the segmentation performance between ultrasound physicians and LunNet using Test Set A. The senior ultrasound physician achieved a dice coefficient of 0.9441, while the junior ultrasound physician obtained 0.6946. While LunNet underperformed compared to senior ultrasound physicians, it demonstrated superior diagnostic capabilities over junior practitioners in pulmonary consolidation assessment. To investigate LunNet's potential for enhancing junior ultrasound physicians' performance, a validation study was conducted where these practitioners re-evaluated consolidation boundaries with LunNet assistance. The intervention significantly improved their measurement accuracy, achieving a Dice coefficient of 0.9221. Comparative analysis revealed that initial diagnostic limitations - including consolidation oversight, misclassification errors, and boundary delineation inaccuracies observed in unaided assessments - were effectively resolved through AI assistance. This enhancement enabled junior ultrasound physicians to attain diagnostic proficiency approaching that of senior practitioners. These findings align with recent research by Phung Tran Huy Nhat et al. (2023), where non-expert clinicians improved their diagnostic accuracy from 68.9\u0026ndash;82.9% when supported by the RAILUS (Real-time AI-assisted LUS) system for interpreting retrospective ultrasound clips [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. Both studies substantiate that AI assistance can significantly enhance diagnostic capabilities for less-experienced practitioners.\u003c/p\u003e \u003cp\u003eHowever, AI is not suitable for independently performing medical image analysis tasks. In clinical practice, AI should operate under physician supervision [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e], with its primary role being to assist ultrasound physicians in rapidly and accurately identifying lesions. By applying deep learning models to the field of LUS, this innovation not only exemplifies interdisciplinary convergence but also demonstrates substantive clinical applicability in real-world diagnostic workflows. The developed LunNet demonstrates significant advantages in addressing the low diagnostic accuracy of junior physicians in pediatric pneumonia, thereby enhancing clinical workflow efficiency. Based on prior research, we propose that LunNet is particularly suitable for junior ultrasound physicians or clinicians lacking ultrasound experience. Under AI assistance, it effectively reduces their learning curve for LUS and improves diagnostic accuracy in clinical applications.\u003c/p\u003e \u003cp\u003eSeveral limitations emerged during the advancement of this study. First, the accuracy and stability of LunNet depend on deep learning-based lesion features. The model exhibited reduced performance in identifying lesions with indistinct boundaries or those obscured by rib shadows. Second, as a single-center study, variations in ultrasound image acquisition\u0026mdash;due to differences in devices and operational protocols\u0026mdash;introduced potential biases. To address this, we validated LunNet\u0026rsquo;s performance on Test Set B, revealing a slight decline in generalizability (dice coefficient: 0.7773). Future efforts will focus on multicenter large-scale external validation to enhance universal applicability. We plan to establish detailed image acquisition and procedural guidelines and continuously enrich the training dataset with multi-institutional data. Additionally, the limited dataset for B-lines and pleural effusion impacted model performance. While pleural effusion, appearing as anechoic or hypoechoic dark areas on ultrasound, can be readily identified even by less experienced physicians, LunNet\u0026rsquo;s segmentation accuracy for these features (dice: 0.8140) fell below expectations. Future work will prioritize expanding the dataset to improve LunNet\u0026rsquo;s precision for B-lines and pleural effusion, thereby optimizing its diagnostic utility.\u003c/p\u003e \u003cp\u003eIn summary, the LunNet system developed in this study enhances the clinical utility of LUS in pediatric lobar pneumonia management through its accurate segmentation of pulmonary consolidations, B-lines, and pleural effusions in ultrasonographic images. This advancement enables more precise diagnostic applications, facilitates longitudinal follow-up assessments, and provides valuable guidance for clinical decision-making in pediatric respiratory care.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eLUS \u0026nbsp; lung ultrasound\u003c/p\u003e\n\u003cp\u003eAI \u0026nbsp; \u0026nbsp;artificial intelligence\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was performed in accordance with the Declaration of Helsinki. This single-center retrospective study was approved by the Institutional Review Board\u0026nbsp;of The Sixth Affiliated Hospital of Harbin Medical University\u0026nbsp;(IRB No. LC2024-091). According to the requirements of the Ethics Committee of the Sixth Affiliated Hospital of Harbin Medical University: All child participants provided written informed consent to participate. For participants under 8 years of age, written informed consent was obtained from their legal guardians. For participants aged 8 to 18 years, written informed consent was jointly signed by both the participants and their legal guardians.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data supporting this study\u0026rsquo;s findings are available from the corresponding author, Binbin Guo and Yingtao Zhang, upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests. The AI algorithms developed in this study are for research purposes only and have no commercial affiliations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTS: Conceptualization, Investigation, Methodology, Software, Validation, Visualization, Writing-original draft, ST: Conceptualization, Formal analysis, Methodology, Software, Visualization, Writing-review \u0026amp; editing, YL: Formal analysis, Investigation, Visualization, Writing-review \u0026amp; editing, LZ: Formal analysis, Investigation, Supervision, Writing-review \u0026amp; editing, YW: Methodology, Supervision, Visualization, Writing-review \u0026amp; editing, YZ: Conceptualization, Methodology, Project administration, Software, Supervision, Writing-review \u0026amp; editing, BG: Conceptualization, Data curation, Methodology, Project administration, Supervision, Validation, Writing-review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors sincerely thank the ultrasound physicians from the Sixth Affiliated Hospital of Harbin Medical University, Xiaoya Chen, Qiang Qiao, Yao Zhang, and Xinhong Yu, their expertise in ultrasound image acquisition and meticulous execution of the research protocol significantly facilitated the completion of this work. And we appreciate all the doctors in the Department of Pediatrics and Imaging of the Sixth Affiliated Hospital of Harbin Medical University for providing clinical information support.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWalker CLF, Rudan I, Liu L, Nair H, Theodoratou E, Bhutta ZA, et al. Global burden of childhood pneumonia and diarrhoea. The Lancet 2013;381:1405-1416\u003c/li\u003e\n\u003cli\u003eSmith DC, Kuckel DP, Recidoro AM. Community-Acquired Pneumonia in Children: Rapid Evidence Review. American Family Physician 2021;104:618-625\u003c/li\u003e\n\u003cli\u003eMarangu D, Zar HJ. Childhood pneumonia in low-and-middle-income countries: An update. Paediatric Respiratory Reviews 2019;32:3-9\u003c/li\u003e\n\u003cli\u003eAlexopoulou E, Prountzos S, Raissaki M, Mazioti A, Caro-Dominguez P, Hirsch FW, et al. Imaging of Acute Complications of Community-Acquired Pneumonia in the Paediatric Population\u0026mdash;From Chest Radiography to MRI. Children 2024;11:122\u003c/li\u003e\n\u003cli\u003ePicano E, Scali MC, Ciampi Q, Lichtenstein D. Lung Ultrasound for the Cardiologist. JACC: Cardiovascular Imaging 2018;11:1692-1705\u003c/li\u003e\n\u003cli\u003eAkkus Z, Cai J, Boonrod A, Zeinoddini A, Weston AD, Philbrick KA, et al. A Survey of Deep-Learning Applications in Ultrasound: Artificial Intelligence\u0026ndash;Powered Ultrasound for Improving Clinical Workflow. Journal of the American College of Radiology 2019;16:1318-1328\u003c/li\u003e\n\u003cli\u003eXiao Y, Liang Q, Zhou L, He X, Lv L, Chen J, et al. Construction of a new automatic grading system for jaw bone mineral density level based on deep learning using cone beam computed tomography. Scientific Reports 2022;12:12841\u003c/li\u003e\n\u003cli\u003eValanarasu JMJ, Sindagi VA, Hacihaliloglu I, Patel VM. KiU-Net: Overcomplete Convolutional Architectures for Biomedical Image and Volumetric Segmentation. IEEE Transactions on Medical Imaging 2022;41:965-976\u003c/li\u003e\n\u003cli\u003eZunair H, Ben Hamza A. Sharp U-Net: Depthwise convolutional network for biomedical image segmentation. Computers in Biology and Medicine 2021;136:104699\u003c/li\u003e\n\u003cli\u003eZhuang Q, Gan S, Zhang L. Human-computer interaction based health diagnostics using ResNet34 for tongue image classification. Computer Methods and Programs in Biomedicine 2022;226:107096\u003c/li\u003e\n\u003cli\u003eGu Z, Cheng J, Fu H, Zhou K, Hao H, Zhao Y, et al. CE-Net: Context Encoder Network for 2D Medical Image Segmentation. IEEE Transactions on Medical Imaging 2019;38:2281-2292\u003c/li\u003e\n\u003cli\u003eAnuar N, Md Sultan AB. Validate Conference Paper using Dice Coefficient. Computer and Information Science 2010;3:16-28\u003c/li\u003e\n\u003cli\u003eTroeger C, Blacker B, Khalil IA, Rao PC, Cao J, Zimsen SRM, et al. Estimates of the global, regional, and national morbidity, mortality, and aetiologies of lower respiratory infections in 195 countries, 1990\u0026ndash;2016: a systematic analysis for the Global Burden of Disease Study 2016. The Lancet Infectious Diseases 2018;18:1191-1210\u003c/li\u003e\n\u003cli\u003eSmith DC, Kuckel DP, Recidoro AM. Community-Acquired Pneumonia in Children: Rapid Evidence Review. Am Fam Physician 2021;104:618-625\u003c/li\u003e\n\u003cli\u003eHunton R. Updated concepts in the diagnosis and management of community-acquired pneumonia. Jaapa 2019;32:18-23\u003c/li\u003e\n\u003cli\u003eCunha BA. The atypical pneumonias: clinical diagnosis and importance. Clinical Microbiology and Infection 2006;12:12-24\u003c/li\u003e\n\u003cli\u003eMiyashita N. Atypical pneumonia: Pathophysiology, diagnosis, and treatment. Respiratory Investigation 2022;60:56-67\u003c/li\u003e\n\u003cli\u003ePrincipi N, Esposito A, Giannitto C, Esposito S. Lung ultrasonography to diagnose community-acquired pneumonia in children. BMC Pulmonary Medicine 2017;17:212\u003c/li\u003e\n\u003cli\u003eSperandeo M, Rotondo A, Guglielmi G, Catalano D, Feragalli B, Trovato GM. Transthoracic ultrasound in the assessment of pleural and pulmonary diseases: use and limitations. La radiologia medica 2014;119:729-740\u003c/li\u003e\n\u003cli\u003eTurner JP, Dankoff J. Thoracic Ultrasound. Emergency Medicine Clinics of North America 2012;30:451-473\u003c/li\u003e\n\u003cli\u003eHo M-C, Ker C-R, Hsu J-H, Wu J-R, Dai Z-K, Chen IC. Usefulness of Lung Ultrasound in the Diagnosis of Community-acquired Pneumonia in Children. Pediatrics \u0026amp; Neonatology 2015;56:40-45\u003c/li\u003e\n\u003cli\u003eFitzgerald DB, Blakey JD, Joshi P, Kuok YJ, Lee YCG, Thomas R. Ultrasound Clues in Lobar Pneumonia. Chest 2022;161:e59-e62\u003c/li\u003e\n\u003cli\u003eLight RW. Pleural Effusion. New England Journal of Medicine 2002;346:1971-1977\u003c/li\u003e\n\u003cli\u003eLiu J, Chen S-W, Liu F, Li Q-P, Kong X-Y, Feng Z-C. The Diagnosis of Neonatal Pulmonary Atelectasis Using Lung Ultrasonography. Chest 2015;147:1013-1019\u003c/li\u003e\n\u003cli\u003eChidini G, Raimondi F. Lung ultrasound for the sick child: less harm and more information than a radiograph. European Journal of Pediatrics 2023;183:1079-1089\u003c/li\u003e\n\u003cli\u003eBeshara M, Bittner EA, Goffi A, Berra L, Chang MG. Nuts and bolts of lung ultrasound: utility, scanning techniques, protocols, and findings in common pathologies. Critical Care 2024;28:328\u003c/li\u003e\n\u003cli\u003ePereda MA, Chavez MA, Hooper-Miele CC, Gilman RH, Steinhoff MC, Ellington LE, et al. Lung Ultrasound for the Diagnosis of Pneumonia in Children: A Meta-analysis. Pediatrics 2015;135:714-722\u003c/li\u003e\n\u003cli\u003eLlamas-\u0026Aacute;lvarez AM, Tenza-Lozano EM, Latour-P\u0026eacute;rez J. Accuracy of Lung Ultrasonography in the Diagnosis of Pneumonia in Adults. Chest 2017;151:374-382\u003c/li\u003e\n\u003cli\u003eHeuvelings CC, B\u0026eacute;lard S, Familusi MA, Spijker R, Grobusch MP, Zar HJ. Chest ultrasound for the diagnosis of paediatric pulmonary diseases: a systematic review and meta-analysis of diagnostic test accuracy. British Medical Bulletin 2019;129:35-51\u003c/li\u003e\n\u003cli\u003eTrinavarat P, Riccabona M. Potential of ultrasound in the pediatric chest. European Journal of Radiology 2014;83:1507-1518\u003c/li\u003e\n\u003cli\u003eKhan U, Afrakhteh S, Mento F, Fatima N, De Rosa L, Custode LL, et al. Benchmark methodological approach for the application of artificial intelligence to lung ultrasound data from COVID-19 patients: From frame to prognostic-level. Ultrasonics 2023;132:106994\u003c/li\u003e\n\u003cli\u003eCattarossi L. Lung ultrasound: its role in neonatology and pediatrics. Early Human Development 2013;89:S17-S19\u003c/li\u003e\n\u003cli\u003eChen X, Wang X, Zhang K, Fung K-M, Thai TC, Moore K, et al. Recent advances and clinical applications of deep learning in medical image analysis. Medical Image Analysis 2022;79:102444\u003c/li\u003e\n\u003cli\u003eShen Y-T, Chen L, Yue W-W, Xu H-X. Artificial intelligence in ultrasound. European Journal of Radiology 2021;139:109717\u003c/li\u003e\n\u003cli\u003eHammerschmidt DE. About the cover illustration. Journal of Laboratory and Clinical Medicine 2004;143:327\u003c/li\u003e\n\u003cli\u003eDaugherty EA. Lobar Pneumonia. Medical Clinics of North America 1947;31:1432-1441\u003c/li\u003e\n\u003cli\u003eZinserling VA, Swistunov VV, Botvinkin AD, Stepanenko LA, Makarova AE. Lobar (croupous) pneumonia: old and new data. Infection 2021;50:235-242\u003c/li\u003e\n\u003cli\u003eReynolds JH, McDonald G, Alton H, Gordon SB. Pneumonia in the immunocompetent patient. The British Journal of Radiology 2010;83:998-1009\u003c/li\u003e\n\u003cli\u003eBasse P, Gr\u0026eacute;gory J, Lavenne R, Foucrier A. Right pan-lobar pneumonia due to Streptococcus pneumoniae. Intensive Care Medicine 2022;48:1647-1647\u003c/li\u003e\n\u003cli\u003eMalla D, Rathi V, Gomber S, Upreti L. Can lung ultrasound differentiate between bacterial and viral pneumonia in children? Journal of Clinical Ultrasound 2020;49:91-100\u003c/li\u003e\n\u003cli\u003eBotana Rial M, P\u0026eacute;rez Pallar\u0026eacute;s J, Cases Viedma E, L\u0026oacute;pez Gonz\u0026aacute;lez FJ, Porcel JM, Rodr\u0026iacute;guez M, et al. Diagnosis and Treatment of Pleural Effusion. Recommendations of the Spanish Society of Pulmonology and Thoracic Surgery. Update 2022. Archivos de Bronconeumolog\u0026iacute;a 2023;59:27-35\u003c/li\u003e\n\u003cli\u003eFatima N, Khan U, Han X, Zannin E, Rigotti C, Cattaneo F, et al. Deep learning approaches for automated classification of neonatal lung ultrasound with assessment of human-to-AI interrater agreement. Computers in Biology and Medicine 2024;183:109315\u003c/li\u003e\n\u003cli\u003eLichtenstein D. Lung ultrasound in the critically ill. Current Opinion in Critical Care 2014;20:315-322\u003c/li\u003e\n\u003cli\u003eEvertsson M, Graneli C, Vernersson A, Wiaczek O, Hagelsteen K, Erl\u0026ouml;v T, et al. Design of a Pediatric Rectal Ultrasound Probe Intended for Ultra-High Frequency Ultrasound Diagnostics. Diagnostics 2023;13:1667\u003c/li\u003e\n\u003cli\u003eChen J, Li J, He C, Li W, Li Q. Automated Pleural Line Detection Based on Radon Transform Using Ultrasound. Ultrasonic Imaging 2020;43:19-28\u003c/li\u003e\n\u003cli\u003eMu\u0026ntilde;oz Moreno JF, Rubio Prieto E, Magro Mart\u0026iacute;n M\u0026Aacute;. Lung Ultrasound in ARDS: B-lines Pattern and Shred Sign. Archivos de Bronconeumolog\u0026iacute;a 2024;60：180\u003c/li\u003e\n\u003cli\u003eBeaudoin S, Gonzalez AV. Evaluation of the patient with pleural effusion. Canadian Medical Association Journal 2018;190:E291-E295\u003c/li\u003e\n\u003cli\u003eMayo PH, Copetti R, Feller-Kopman D, Mathis G, Maury E, Mongodi S, et al. Thoracic ultrasonography: a narrative review. Intensive Care Medicine 2019;45:1200-1211\u003c/li\u003e\n\u003cli\u003eBrogi E, Gargani L, Bignami E, Barbariol F, Marra A, Forfori F, et al. Thoracic ultrasound for pleural effusion in the intensive care unit: a narrative review from diagnosis to treatment. Critical Care 2017;21:325\u003c/li\u003e\n\u003cli\u003eLitjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, et al. A survey on deep learning in medical image analysis. Medical Image Analysis 2017;42:60-88\u003c/li\u003e\n\u003cli\u003eLee J-G, Jun S, Cho Y-W, Lee H, Kim GB, Seo JB, et al. Deep Learning in Medical Imaging: General Overview. Korean Journal of Radiology 2017;18:570\u003c/li\u003e\n\u003cli\u003eNhat PTH, Van Hao N, Tho PV, Kerdegari H, Pisani L, Thu LNM, et al. Clinical benefit of AI-assisted lung ultrasound in a resource-limited intensive care unit. Critical Care 2023;27:257\u003c/li\u003e\n\u003cli\u003ePeng S, Liu Y, Lv W, Liu L, Zhou Q, Yang H, et al. Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre diagnostic study. The Lancet Digital Health 2021;3:e250-e259\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-pulmonary-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pulm","sideBox":"Learn more about [BMC Pulmonary Medicine](http://bmcpulmmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pulm/default.aspx","title":"BMC Pulmonary Medicine","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Lung ultrasound, Artificial intelligence, Pediatric pneumonia, Deep learning, Lobar pneumonia","lastPublishedDoi":"10.21203/rs.3.rs-6410448/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6410448/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePneumonia remains a major contributor to global childhood morbidity and mortality, posing significant public health challenges. Lung ultrasound (LUS) serves as a critical tool for phased assessment of pneumonia progression and guidance of clinical management. This study developed a deep learning artificial intelligence (AI) model (LunNet) to automatically identify and precisely segment lesion characteristics in LUS images, aiming to assist ultrasound physicians in accurate lesion measurement for longitudinal disease monitoring and treatment guidance.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe retrospectively analyzed 419 pediatric patients diagnosed with lobar pneumonia (male : female, 199:220; mean age 7.1\u0026thinsp;\u0026plusmn;\u0026thinsp;3.0 years) who underwent LUS examinations between May 2024 and December 2024. A total of 1,383 images from this cohort were used for LunNet (modified U-Net) development and validation. The model's lesion segmentation performance was evaluated using the dice coefficient and compared with the performance of ultrasound physicians.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eLunNet demonstrated robust performance in automatically identifying and segmenting lung consolidation, B-lines, and pleural effusion, achieving mean dice coefficients of 0.8401 (95% CI: 0.8191\u0026ndash;0.8610), 0.8274 (95% CI: 0.7874\u0026ndash;0.8673), and 0.8140 (95% CI: 0.7808\u0026ndash;0.8472), respectively. The segmentation performance for lung consolidation exhibited marked disparity between junior and senior ultrasound physicians, with mean dice coefficients of 0.6946 (95% CI: 0.6312\u0026ndash;0.7581) and 0.9441 (95% CI: 0.9352\u0026ndash;0.9530), respectively. Notably, when assisted by LunNet, junior ultrasound physicians exhibited substantial improvement in lung consolidation segmentation, attaining a mean dice coefficient of 0.9221 (95% CI: 0.8191\u0026ndash;0.8610), (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). In the generalizability validation experiment, LunNet maintained competent performance for lung consolidation segmentation with a Dice coefficient of 0.7773 (95% CI: 0.9108\u0026ndash;0.9335).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe LunNet AI model demonstrates excellent segmentation capabilities for pediatric lobar pneumonia lesions in ultrasound imaging. It effectively assists ultrasound physicians in precise quantification of pathological findings and significantly enhances diagnostic efficiency for novice practitioners. These results underscore LunNet's potential clinical value in supporting diagnosis, longitudinal monitoring, and therapeutic decision-making for lobar pneumonia.\u003c/p\u003e","manuscriptTitle":"Application of Deep Learning-Based Artificial Intelligence Model in Lung Ultrasound for Pediatric Lobar Pneumonia","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-07 03:11:40","doi":"10.21203/rs.3.rs-6410448/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2025-06-08T16:37:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"75585218233837609704693148297978817508","date":"2025-04-23T06:33:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"120404124463081768260700223499091494793","date":"2025-04-23T06:20:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-23T01:50:40+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-22T11:31:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-21T05:57:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Pulmonary Medicine","date":"2025-04-21T05:56:22+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-pulmonary-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pulm","sideBox":"Learn more about [BMC Pulmonary Medicine](http://bmcpulmmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pulm/default.aspx","title":"BMC Pulmonary Medicine","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"39cce6bd-df19-465d-bd39-a3fc41266e0e","owner":[],"postedDate":"May 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-05-07T03:11:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-07 03:11:40","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6410448","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6410448","identity":"rs-6410448","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}