AI-assisted ultrasound for early liver trauma: Animal models & clinical validation

AI-assisted ultrasound for early liver trauma: Animal models & clinical validation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article AI-assisted ultrasound for early liver trauma: Animal models & clinical validation Qing Song, Xuelei He, Yanjie Wang, Hanjing Gao, Li Tan, Jun Ma, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4454754/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Jul, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract The study aimed to develop an AI-assisted ultrasound model for early liver trauma identification, using data from Bama miniature pigs and patients in Beijing, China. A deep learning model was created and fine-tuned with animal and clinical data, achieving high accuracy metrics. In internal tests, the model outperformed both Junior and Senior sonographers. External tests showed the model's effectiveness, with a Dice Similarity Coefficient of 0.74, True Positive Rate of 0.80, Positive Predictive Value of 0.74, and 95% Hausdorff distance of 14.84. The model's performance was comparable to Junior sonographers and slightly lower than Senior sonographers. This AI model shows promise for liver injury detection, offering a valuable tool with diagnostic capabilities similar to those of less experienced human operators. Biological sciences/Computational biology and bioinformatics Biological sciences/Computational biology and bioinformatics/Computational models Health sciences/Diseases Liver Contusions Lacerations Deep Learning Artificial Intelligence Animal Models Predictive Value of Tests Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction The liver is one of the most frequently injured organs. Liver trauma is responsible for about 15% of all abdominal injuries( 1 , 2 ) and approximately 5% of all emergency admissions ( 3 ). Liver trauma is often combined with further complications, such as secondary bleeding, bile duct injury, bile leakage, and peritonitis syndrome, leading to high mortality that can exceed 75%( 4 , 5 ). The prognosis of patients could be notably improved by quick evaluation of the injury and minimally invasive treatment ( 6 , 7 ). Thus, timely assessment of the liver is the priority for trauma patients, especially under emergency conditions. Ultrasound (US) is a commonly used method to assess liver trauma. It is an affordable real-time imaging approach, allowing dynamic, convenient, and continuous examination( 2 , 8 ). However, compared to other imaging techniques, the quality of US imaging is significantly affected by the physiological characteristics of patients and is more dependent on the operation methods and human factors ( 9 ). Thus, artificial intelligence (AI)-empowered ultrasonography has the potential to accelerate further the use of medical ultrasound in various clinical settings by analyzing the patient's clinical information, gene information, and multimodal imaging information, simplifying the operation steps and avoiding subjective differences, saving physician resources, shortening the reporting time, and improving the diagnostic efficiency( 10 – 13 ). At present, AI studies are mainly focused on investigating breast nodules and liver tumors; AI in the thyroid, obstetrics, gynecology, pelvic floor, and vascular diseases have rarely been reported. In the context of liver injury, AI can potentially find the differences of subtle texture features in images that cannot be recognized by the naked eye( 14 ), which may be helpful for the early diagnosis of liver parenchymal contusion and laceration missed by doctors in US images. Aside from liver tumors, successful usage of the AI was reported for non-alcoholic fatty liver disease( 15 ), cholangitis( 16 ), and focal liver lesions( 17 ). Therefore, we established and validated a new AI-assisted US model for identifying liver trauma (parenchymal contusion and laceration). We hypothesized the AI-assisted US model could achieved comparable diagnostic efficiency with sonographers, and could assist early identifying of liver trauma. Results Establishment of the CNDLM A total of 1603 US images (1373 on the training set, 230 in the validation set; 116 were selected as the test set) were obtained from pigs and 126 from patients admitted to 3 centers (30 from the first center were selected as training set for fine-tuning of CNDLM, 18 were selected as a clinical internal validation set, and 78 were selected as clinical external validation set). There were 3 animal CV models and 3 clinical CV models after training and fine-tuning, with acceptable accuracy in animal test set ( Table S1 and Fig. 1 ). Then the CNDLM was established. Performance of CNDLM The CNDLM achieved significantly better performance (DSC = 0.91 [0.87–0.95]; TPR = 0.91 [0.86–0.95]; PPV = 0.95 [0.93–0.97]; 95_HD = 2.55 [1.85–3.56]) compared to Junior and Senior sonographers (all P < 0.001) (Table 1 , Fig. 2 ) in animal test set. The CNDLM also demonstrated satisfactory performance (Dice = 0.88 [0.84–0.93]; TPR = 0.87 [0.81–0.92]; PPV = 0.92 [0.86–0.96]; 95_HD = 11.78 [4.08–22.04]) in the clinical internal test set (Table 1 , Fig. 3 ), showing superior diagnostic performance compared to Junior (P = 0.001, 0.012, respectively) and Senior sonographers (both P = 0.004). In the external test set, the CNDLM achieved the DSC of 0.74 (0.66–0.81), TPR of 0.80 (0.73–0.87), PPV of 0.74(0.65–0.82), and HD-95 of 14.84 (8.34–23.11). The performance was significantly better than one of the junior sonographers, but significantly inferior to another junior sonographers (Table 1 , Fig. 4 ). Table 1 The detailed quantitative metrics of the ensembled model and different level sonographers DSC TPR PPV HD-95 P* Clinical internal test set Junior 1 0.70(0.63–0.77) 0.79(0.70–0.86) 0.67(0.58–0.75) 21.19(10.90-34.21) 0.0011 Junior 2 0.72(0.66–0.79) 0.90(0.85–0.94) 0.64(0.55–0.72) 13.70(8.90–19.80) 0.0123 Seniority 1 0.79(0.74–0.83) 0.94(0.90–0.97) 0.70(0.63–0.76) 15.92(8.12–26.94) 0.0040 Seniority 2 0.71(0.63–0.77) 0.92(0.87–0.96) 0.60(0.52–0.68) 15.55(11.17–19.98) 0.0040 CNDLM 0.88(0.84–0.93) 0.87(0.81–0.92) 0.92(0.86–0.96) 11.78(4.08–22.04) - Clinical external test set Junior 1 0.75(0.69–0.80) 0.75(0.69–0.80) 0.79(0.72–0.85) 13.66(9.17–19.13) 0.0025 Junior 2 0.72(0.64–0.78) 0.84(0.78–0.90) 0.68(0.60–0.76) 18.85(12.86–26.37) 0.0025 Seniority 1 0.79(0.75–0.83) 0.86(0.81–0.90) 0.78(0.73–0.82) 11.77(7.99–16.19) 0.0025 Seniority 2 0.76(0.72–0.80) 0.94(0.91–0.97) 0.66(0.61–0.71) 10.01(7.35–13.18) 0.0025 CNDLM 0.74(0.66–0.81) 0.80(0.73–0.87) 0.74(0.65–0.82) 14.84(8.34–23.11) - Animal test set Junior 1 0.33(0.28–0.38) 0.60(0.52–0.67) 0.25(0.20–0.29) 48.21(41.16–55.71) < 0.001 Junior 2 0.31(0.26–0.37) 0.54(0.46–0.62 0.24(0.20–0.29) 67.89(57.06–78.03) < 0.001 Seniority 1 0.64(0.57–0.69) 0.71(0.65–0.78) 0.60(0.54–0.66) 22.65(16.71–29.47) < 0.001 Seniority 2 0.58(0.52–0.64) 0.65(0.59–0.71) 0.57(0.51–0.63) 38.21(28.52–47.74) < 0.001 CNDLM 0.91(0.87–0.95) 0.91(0.86–0.95) 0.95(0.93–0.97) 2.55(1.85–3.56) - *: Comparing the Dice Similarity Coefficient (DSC) with convolutional neural network-based liver injury diagnostic model (CNDLM). TPR: true positive rate; PPV: true positive rate; HD-95: 95% Hausdorff distance; CNDLM: convolutional neural network-based liver injury diagnostic model Discussion In this study, a deep learning model was proposed for liver injury detection, which could achieve satisfactory diagnostic performance in assessing clinical US images comparable to that of the trained sonographers. To the best of our knowledge, this is the first study that proposed the clinical DL model for liver trauma evaluation that demonstrated good accuracy. Applying AI methods in the liver US could potentially increase the quality of imaging while limiting the influence of the human factor to achieve an early, rapid, and accurate diagnosis of liver trauma. The Bama miniature pig’s liver resembles the human liver in terms of anatomy, morphology, and texture echo ( 18 ). Thus, these animals were selected as the experimental animal in this study. The automatic segmentation results of the model were compared with the standard range manually outlined by the trained sonographers, and DSC evaluated the output performance of the model. It is generally believed that the Dice coefficient > 0.7 indicates that the model has better performance ( 19 , 20 ). In this study, the DSCs were all above 0.70, which confirmed that the automatically segmented liver injury area and the real injury area overlapped. The average positive predictive value (PPV) reached 0.884, and the average true positive rate (TPR) was about 0.780, which is comparable with the previous reports by Sato et al. ( 21 ) or Saillard et al. ( 22 ) for the DL models in the diagnosis of hepatocellular carcinoma. The obtained results confirmed that the model not only accurately located the wound location but also outlined the wound contour more reliably. Despite the promising accuracy, peripheral false positives or internal false negatives were reported when the lesions were over-segmented or under-segmented. This may be at least partly explained by the following two points: firstly, the grayscale of the ultrasound image of liver trauma changes with time. Within a week of the formation of liver trauma, the trauma area often shows a heterogeneous echo area or a slightly hyperechoic area, and sometimes the grayscale contrast with the normal liver parenchyma is not significant. With time, the echo in the lesion area gradually decreases, and the grayscale contrast with normal tissue also gradually increases. Most lesions have echogenicity similar to the surrounding liver parenchyma in about two months ( 23 ). Secondly, in the case of slight injury, the grayscale contrast with normal tissue is not significant, and the more severe the injury, the greater the grayscale difference from normal tissue. Therefore, the sensitivity and specificity of the proposed model should be further evaluated based on the above data. One of the strengths of this study is the large number of animal US image data collected for training, which were used to guide the model and maintain the existing low-level features through pre-trained weights. Additionally, the proposed AI model was fine-tuned using ultrasound image data of patients to correct the difference between human and animals. This approach could potentially increase the diagnostic accuracy to the level of human sonographers, as was previously reported( 24 ). To test the potential of the model, this study compared the diagnostic accuracy of physicians with different seniority and AI. The results showed that the model outperformed some junior physicians in external test set, and the ensemble model had better diagnostic performance than physicians of all levels in the internal test set, which is consistent with previous studies ( 24 , 25 ). Thus, in small hospitals with comparable conditions, as well as teaching hospitals that undertake the teaching work of young residents, the use of AI-assisted diagnosis could potentially help to improve the quality of ultrasound diagnosis. Finally, the concept of “golden time” is emphasized in trauma first aid, with time becoming the most important factor affecting survival( 26 ). Therefore, the methods that shorten the assessment time before the surgery are most important. In emergency situations, US can effectively guide clinical classification and treatment, but its interpretation is significantly affected by the physiological characteristics of patients( 27 ). Moreover, ultrasound examination has certain subjective differences, and training professionals require a lot of long-term practice and learning. In contrast, combining US with AI can simplify operation steps, avoid subjective differences, save physician resources, shorten examination time, and improve diagnostic efficiency. This study demonstrated that the application of AI to diagnosing liver trauma could potentially maximize the portability and speed of real-time US interpretation, which was consistent with previous studies ( 28 – 30 ). This study has several limitations. Firstly, the number of clinical US images was limited due to the low incidence of liver trauma, resulting in an imbalance of image sources. For this issue, this study supplemented clinical data to fine-tune the animal model; through data amplification and batch normalization, overfitting was reduced as much as possible, and robustness was enhanced. Secondly, clinical data in this study were collected retrospectively, with could have led to a potential bias. The transfer learning method was adopted due to the lack of clinical US data, which cannot meet the needs of building deep learning models. Although we used methods such as increasing the data set, enhancing the data, reducing the learning rate, and using regularization parameters for data preprocessing, there was still a certain overfitting problem in the model training, which needs to be increased in future research. Finally, clinical data of patients (including weight and blood pressure) were not taken into account at this stage. Future studies should focus on validating the model in the clinical setting. In conclusion, this study proposed a deep learning model based on ResNet50 structure with ImageNet pre-trained weights for liver injury detection from 2D ultrasound images, which achieved comparable diagnostic performance with Junior sonographers. Methods The study was undertaken in two stages. The first stage included the animal experimental US images as a training set and verification set. The second stage included the clinical US images as a test set. The animal study strictly complied with the National Institute of Animal Health regulation in experimental design, animal feeding, and sample acquisition. This whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01). Establishment of the experimental model The animal experimental stage took place between January 2021 and April 2021. Eighteen Bama miniature pigs weighing 30–35 kg underwent intramuscular anesthesia (0.3 ml/kg, mix midazolam and serrazine hydrochloride injection in the ratio of 1:1) and endotracheal intubation. The animals were fixed, the venous channel was established through the marginal ear vein, and the liver parenchyma contusion and laceration were modeled with a self-made percussion device ( Supplementary Figure S1 ). After modeling, the contrast-enhanced ultrasound (Ultrasound Diagnostic Instrument with 5 ~ 15 MHz Convex probe, 3 ~ 5MHz frequency) was performed to obtain multiple gray-scale (GS) ultrasound images of the trauma area from different angles. The liver trauma region was defined as region without enhancement in US images, evaluated by a junior sonographer (with less than 3 years of experience). Clinical data collection Clinical US images were collected from patients with liver trauma treated in the first medical center of PLA General Hospital, the seventh medical center of PLA General Hospital, and Beijing Chaoyang emergency center between January 2015 and April 2021. The inclusion criteria were: 1) trauma history within one week; 2) liver trauma confirmed by surgery or contrast-enhanced ultrasound; 3) complete clinical and imaging information. The exclusion criteria were: 1) poor image quality; 2) chronic liver lesions (such as liver cirrhosis); 3) non-traumatic liver rupture (such as rupture of liver tumor). The golden standard for liver trauma region was defined as a region of no enhancement in contrast-enhanced US imaging, drawn by the same sonographer. The true positive was defined as the overlap between the liver trauma region defined and the liver trauma region in the gold standard. Thirty clinical US images from the data collected from the first medical center of PLA General Hospital and the seventh medical center of PLA General Hospital were randomly selected as fine tuning set, and the rest images from the two centers were selected as clinical internal test set. The clinical US images collected from Beijing Chaoyang emergency center were selected as clinical external test. This whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01) and informed consent was obtained from all subjects. Establishment and evaluation of the deep learning (DL) model US images obtained from pigs were randomly divided into a training set and validation set with a ratio of 6:1, and a test set was randomly selected. Three animal cross-validation (CV) models were established and evaluated. In addition, 3 clinical CV models and a convolutional neural network-based liver injury diagnostic model (CNDLM) were established by fine-tuning clinical US images from the first center. The performance of 3 CV models was compared using animal training, validation, and test set, and the performance of CNDLM was compared with data obtained by other sonographers using the animal test set, clinical internal validation set, and clinical external validation set. The workflow of the segmentation DL model is shown in Fig. 1 . The segmentation mask was extracted through the Visual Object Tagging Tool (VOTT, Microsoft, USA) and OpenCV package in python. Then, the images and segmentation mask were resized (224*224 pixels). The model was developed with the RefineNet ( Supplementary Figure S2 ). A Resnet101 network pre-trained on the ImageNet dataset was employed in RefineNet ( Supplementary Figure S3 ). The final model was evaluated by internal test cohort and external test cohorts with human US images. In the training process, the parameters of the model were iteratively updated through backpropagation, whose input and output was the image and segmentation mask, respectively. The dice loss of the output and segmentation mask was used as the loss function. The data enhancement strategy was employed to reduce the impact of overfitting, including random flipping and rotation of the input image. A detailed description of the model establishment and evaluation are shown in the Supplementary Methods section. To compare the diagnostic efficiency of the final model, the animal and clinical images in validation were evaluated by two junior sonographers (with less than 3 years of experiences) and two senior sonographers (with more than 10 years of experiences). Statistical analysis R software (version 3.3.1; R 21 Foundation for Statistical Computing, Vienna, Austria) was used for statistical analysis. The Dice Similarity Coefficient (DSC), True Positive Rate (TPR), Positive Predictive Value (PPV), and 95% Hausdorff distance (HD-95) were used to evaluate the performance of CV, CNDLM, and sonographers. Models were compared for DSC in physician identification using the Kruskal-Wallis test. All statistical tests were two-sided, and the statistical significance level was set at two-sided P < 0.05. Declarations Acknowledgments Not applicable Competing interests The author(s) declare no competing interests. Ethics approval for animal The animalstudy strictly complied with the National Institute of Animal Health regulation in experimental design, animal feeding, and sample acquisition. All methods were carried out in compliance with the ARRIVE guidelines. Ethics approval for human This whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01) and informed consent was obtained from all subjects. All methods were performed in accordance with the relevant guidelines and regulations. Availability of data and materials All data generated or analysed during this study are included in this article and its supplementary information file Funding This study was supported by the China Postdoctoral Scientific Foundation (2018M643876). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Authors' contributions LYK and SQ conceived and supervised the study; LYK, TL and WK designed experiments; WYJ, MJ, KLL and HP performed experiments; WYJ and HXL analysed data; SQ and HXL wrote the manuscript; GHJ and HXL made manuscript revisions. All authors reviewed the results and approved the final version of the manuscript. References Buci, S., et al. The rate of success of the conservative management of liver trauma in a developing country. World J Emerg Surg. 12,24 (2017). Tarchouli, M., et al. Liver trauma: What current management? Hepatobiliary Pancreat Dis Int. 17,39–44 (2018). Saviano, A., et al. Liver Trauma: Management in the Emergency Setting and Medico-Legal Implications. Diagnostics (Basel) . 12, (2022). Pillai, A.S., Kumar, G. & Pillai, A.K. Hepatic Trauma Interventions. Semin Intervent Radiol. 38,96–104 (2021). Sánchez-Bueno, F., et al. [Changes in the diagnosis and therapeutic management of hepatic trauma. A retrospective study comparing 2 series of cases in different (1997 – 1984 vs. 2001–2008)]. Cir Esp. 89,439–447 (2011). Kanani, A., Sandve, K.O. & Søreide, K. Management of severe liver injuries: push, pack, pringle - and plug! Scand J Trauma Resusc Emerg Med. 29,93 (2021). Naylor, J.F., et al. Emergency department imaging of pediatric trauma patients during combat operations in Iraq and Afghanistan. Pediatr Radiol. 48,620–625 (2018). Pillai, A.S., Srinivas, S., Kumar, G. & Pillai, A.K. Where Does Interventional Radiology Fit in with Trauma Management Algorithm? Semin Intervent Radiol. 38,3–8 (2021). Lada, N.E., et al. Liver trauma: hepatic vascular injury on computed tomography as a predictor of patient outcome. Eur Radiol. 31,3375–3382 (2021). Gong, J., et al. A deep residual learning network for predicting lung adenocarcinoma manifesting as ground-glass nodule on CT images. Eur Radiol. 30,1847–1855 (2020). Veerankutty, F.H., et al. Artificial Intelligence in hepatology, liver surgery and transplantation: Emerging applications and frontiers of research. World J Hepatol. 13,1977–1990 (2021). Cai, S.L., et al. Using a deep learning system in endoscopy for screening of early esophageal squamous cell carcinoma (with video). Gastrointest Endosc. 90,745–753.e742 (2019). Li, C., et al. Development and validation of an endoscopic images-based deep learning model for detection with nasopharyngeal malignancies. Cancer Commun (Lond). 38,59 (2018). Ahn, S.H., et al. Comparative clinical evaluation of atlas and deep-learning-based auto-segmentation of organ structures in liver cancer. Radiat Oncol. 14,213 (2019). Biswas, M., et al. Symtosis: A liver ultrasound tissue characterization and risk stratification in optimized deep learning paradigm. Comput Methods Programs Biomed. 155,165–177 (2018). Ringe, K.I., et al. Fully automated detection of primary sclerosing cholangitis (PSC)-compatible bile duct changes based on 3D magnetic resonance cholangiopancreatography using machine learning. Eur Radiol. 31,2482–2489 (2021). Schmauch, B., et al. Diagnosis of focal liver lesions from ultrasound using deep learning. Diagn Interv Imaging. 100,227–233 (2019). Yekuo, L., et al. Contrast-enhanced ultrasound for blunt hepatic trauma: an animal experiment. Am J Emerg Med. 28,828–833 (2010). Li, L., et al. Automatic multi-plaque tracking and segmentation in ultrasonic videos. Med Image Anal. 74,102201 (2021). Tajbakhsh, N., et al. Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation. Med Image Anal. 63,101693 (2020). Sato, M., et al. Machine-learning Approach for the Development of a Novel Predictive Model for the Diagnosis of Hepatocellular Carcinoma. Sci Rep. 9,7704 (2019). Saillard, C., et al. Predicting Survival After Hepatocellular Carcinoma Resection Using Deep Learning on Histological Slides. Hepatology. 72,2000–2013 (2020). Padalino, P., et al. Healing of Blunt Liver Injury After Non-Operative Management: Role of Ultrasonography Follow-Up. Eur J Trauma Emerg Surg. 35,364–370 (2009). Zhao, C., et al. Feasibility of computer-assisted diagnosis for breast ultrasound: the results of the diagnostic performance of S-detect from a single center in China. Cancer Manag Res. 11,921–930 (2019). Li, J., et al. The value of S-Detect for the differential diagnosis of breast masses on ultrasound: a systematic review and pooled meta-analysis. Med Ultrason. 22,211–219 (2020). Hussmann, B. & Lendemans, S. Pre-hospital and early in-hospital management of severe injuries: changes and trends. Injury. 45 Suppl 3,S39-42 (2014). Akkus, Z., et al. A Survey of Deep-Learning Applications in Ultrasound: Artificial Intelligence-Powered Ultrasound for Improving Clinical Workflow. J Am Coll Radiol. 16,1318–1328 (2019). Ciompi, F., et al. Automatic classification of pulmonary peri-fissural nodules in computed tomography using an ensemble of 2D views and a convolutional neural network out-of-the-box. Med Image Anal. 26,195–202 (2015). Wang, X., et al. Searching for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus non-deep learning. Sci Rep. 7,15415 (2017). Lassau, N., et al. Five simultaneous artificial intelligence data challenges on ultrasound, CT, and MRI. Diagn Interv Imaging. 100,199–209 (2019). Additional Declarations No competing interests reported. Supplementary Files SupplementalMaterials.docx Cite Share Download PDF Status: Published Journal Publication published 02 Jul, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 29 Jul, 2024 Reviews received at journal 25 Jul, 2024 Reviews received at journal 10 Jul, 2024 Reviewers agreed at journal 09 Jul, 2024 Reviewers agreed at journal 09 Jul, 2024 Reviewers invited by journal 09 Jul, 2024 Editor assigned by journal 09 Jul, 2024 Editor invited by journal 04 Jun, 2024 Submission checks completed at journal 31 May, 2024 First submitted to journal 21 May, 2024 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-4454754","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":312545685,"identity":"049ca247-861a-4123-b1b5-f3996b825943","order_by":0,"name":"Qing Song","email":"","orcid":"","institution":"First Medical Center of General Hospital of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Qing","middleName":"","lastName":"Song","suffix":""},{"id":312545686,"identity":"58750774-1869-4bca-aff7-29dc0b78eff9","order_by":1,"name":"Xuelei He","email":"","orcid":"","institution":"Northwest University","correspondingAuthor":false,"prefix":"","firstName":"Xuelei","middleName":"","lastName":"He","suffix":""},{"id":312545687,"identity":"4687d67f-8f53-4c81-8ee8-6329bf35fcc8","order_by":2,"name":"Yanjie Wang","email":"","orcid":"","institution":"Shandong Province Maternal and Child Health Care Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yanjie","middleName":"","lastName":"Wang","suffix":""},{"id":312545688,"identity":"34ee921d-c839-4967-86cb-321927b6c743","order_by":3,"name":"Hanjing Gao","email":"","orcid":"","institution":"General Hospital of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Hanjing","middleName":"","lastName":"Gao","suffix":""},{"id":312545689,"identity":"c29e9e88-7de9-4b56-b1f7-e940e84a8bc3","order_by":4,"name":"Li Tan","email":"","orcid":"","institution":"Beijing Da Wang Lu Emergency Hospital","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Tan","suffix":""},{"id":312545690,"identity":"3e1a19c2-b9aa-4b85-9b79-afd48691d534","order_by":5,"name":"Jun Ma","email":"","orcid":"","institution":"First Medical Center of General Hospital of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Jun","middleName":"","lastName":"Ma","suffix":""},{"id":312545691,"identity":"7d54d398-c8e5-416e-81ae-2fa134617241","order_by":6,"name":"Linli Kang","email":"","orcid":"","institution":"First Medical Center of General Hospital of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Linli","middleName":"","lastName":"Kang","suffix":""},{"id":312545692,"identity":"2a4b24e3-bc02-4015-8a8a-42fcfc042fe8","order_by":7,"name":"Peng Han","email":"","orcid":"","institution":"First Medical Center of General Hospital of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Han","suffix":""},{"id":312545693,"identity":"e81c76f3-ee99-4b79-ae60-a63e4bd25e48","order_by":8,"name":"Yukun Luo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAArklEQVRIie3PsQoCMQyA4RyFukTraNGHCHQQ4dBXKQh1cfARCvcS9UVuzuGuD+ByIDj3dge5wT1ugv2nDPkIASiVfrAZaNV7qtGYKCQatKZ8CiubWE7ApnypKXopWQTvkG5IwFUejjLCD6Q7rlVU9tyKyCG6kWwiazWVkiXSFYm9mAS2ifgbgs9AmfZoU9fIfjGT4Hr/2u6Mabo8SAjM/WeqomR/PMPCxVKpVPrf3l66MjT8/3wKAAAAAElFTkSuQmCC","orcid":"","institution":"First Medical Center of General Hospital of Chinese PLA","correspondingAuthor":true,"prefix":"","firstName":"Yukun","middleName":"","lastName":"Luo","suffix":""},{"id":312545694,"identity":"f44997ae-fd76-4c90-8d15-670bd88937b7","order_by":9,"name":"Kun Wang","email":"","orcid":"","institution":"Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Kun","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2024-05-21 12:17:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4454754/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4454754/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-91900-5","type":"published","date":"2025-07-02T15:58:04+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":58596328,"identity":"3fe03938-b36b-43a4-9b15-ad5547b8dc70","added_by":"auto","created_at":"2024-06-18 16:46:03","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":666738,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy flow chart.\u003c/strong\u003eThe three steps were: data preparation, model construction, and model evaluation. The data preparation included collecting animal and clinical US images. The model construction was a model training process. Three-fold cross-validation was used to train the animal CV model, which was then transferred into clinical data, finally ensembling 3 CV models. The model evaluation was performed as a comparison with the trained sonographers using visual methods and quantitative metrics.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/fdafe10081da4db59a949f48.png"},{"id":58595338,"identity":"bbc065c5-a393-41d7-a4e1-0909906fc345","added_by":"auto","created_at":"2024-06-18 16:38:03","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1500133,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe segmentation results of ensembled models and different level sonographers in test animal set.\u003c/strong\u003e The red region is a mask, the green region is a model prediction, and the yellow region is an overlap region of mask and model predictions.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/583fcff59c36b716638ed281.png"},{"id":58595337,"identity":"f396749c-c6b9-4a83-b122-a0a0c83f53e8","added_by":"auto","created_at":"2024-06-18 16:38:03","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":879715,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe segmentation results of the ensembled model and different level sonographers in the clinical internal test.\u003c/strong\u003e The red region is a mask, the green region is a model prediction, and the yellow region is an overlap region of mask and model predictions.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/7931c7564e1049c127600ca5.png"},{"id":58595342,"identity":"1bd15333-592d-4ab8-bdae-2251b414664c","added_by":"auto","created_at":"2024-06-18 16:38:04","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1120670,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe segmentation results of the ensembled models and different level sonographers in the clinical external test set.\u003c/strong\u003e The red region is a mask, the green region is a model prediction, and the yellow region is an overlap region of mask and model predictions.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/18b4a730784abcf92c214d49.png"},{"id":86179164,"identity":"a582e47c-c67c-47ec-8a81-442c6b7f397d","added_by":"auto","created_at":"2025-07-07 16:16:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6723647,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/5554dfbb-a7a4-42b0-8477-a24ce7d0f34d.pdf"},{"id":58595340,"identity":"7fe9164a-2196-42ac-9243-9582bf06469c","added_by":"auto","created_at":"2024-06-18 16:38:03","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":411911,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-4454754/v1/e4a8eb1810660115bc85e334.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-assisted ultrasound for early liver trauma: Animal models \u0026 clinical validation","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe liver is one of the most frequently injured organs. Liver trauma is responsible for about 15% of all abdominal injuries(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) and approximately 5% of all emergency admissions (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Liver trauma is often combined with further complications, such as secondary bleeding, bile duct injury, bile leakage, and peritonitis syndrome, leading to high mortality that can exceed 75%(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). The prognosis of patients could be notably improved by quick evaluation of the injury and minimally invasive treatment (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Thus, timely assessment of the liver is the priority for trauma patients, especially under emergency conditions.\u003c/p\u003e \u003cp\u003eUltrasound (US) is a commonly used method to assess liver trauma. It is an affordable real-time imaging approach, allowing dynamic, convenient, and continuous examination(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). However, compared to other imaging techniques, the quality of US imaging is significantly affected by the physiological characteristics of patients and is more dependent on the operation methods and human factors (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Thus, artificial intelligence (AI)-empowered ultrasonography has the potential to accelerate further the use of medical ultrasound in various clinical settings by analyzing the patient's clinical information, gene information, and multimodal imaging information, simplifying the operation steps and avoiding subjective differences, saving physician resources, shortening the reporting time, and improving the diagnostic efficiency(\u003cspan additionalcitationids=\"CR11 CR12\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAt present, AI studies are mainly focused on investigating breast nodules and liver tumors; AI in the thyroid, obstetrics, gynecology, pelvic floor, and vascular diseases have rarely been reported. In the context of liver injury, AI can potentially find the differences of subtle texture features in images that cannot be recognized by the naked eye(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), which may be helpful for the early diagnosis of liver parenchymal contusion and laceration missed by doctors in US images. Aside from liver tumors, successful usage of the AI was reported for non-alcoholic fatty liver disease(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), cholangitis(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), and focal liver lesions(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Therefore, we established and validated a new AI-assisted US model for identifying liver trauma (parenchymal contusion and laceration). We hypothesized the AI-assisted US model could achieved comparable diagnostic efficiency with sonographers, and could assist early identifying of liver trauma.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eEstablishment of the CNDLM\u003c/h2\u003e \u003cp\u003eA total of 1603 US images (1373 on the training set, 230 in the validation set; 116 were selected as the test set) were obtained from pigs and 126 from patients admitted to 3 centers (30 from the first center were selected as training set for fine-tuning of CNDLM, 18 were selected as a clinical internal validation set, and 78 were selected as clinical external validation set). There were 3 animal CV models and 3 clinical CV models after training and fine-tuning, with acceptable accuracy in animal test set (\u003cb\u003eTable \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Then the CNDLM was established.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003ePerformance of CNDLM\u003c/h2\u003e \u003cp\u003eThe CNDLM achieved significantly better performance (DSC\u0026thinsp;=\u0026thinsp;0.91 [0.87\u0026ndash;0.95]; TPR\u0026thinsp;=\u0026thinsp;0.91 [0.86\u0026ndash;0.95]; PPV\u0026thinsp;=\u0026thinsp;0.95 [0.93\u0026ndash;0.97]; 95_HD\u0026thinsp;=\u0026thinsp;2.55 [1.85\u0026ndash;3.56]) compared to Junior and Senior sonographers (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) in animal test set. The CNDLM also demonstrated satisfactory performance (Dice\u0026thinsp;=\u0026thinsp;0.88 [0.84\u0026ndash;0.93]; TPR\u0026thinsp;=\u0026thinsp;0.87 [0.81\u0026ndash;0.92]; PPV\u0026thinsp;=\u0026thinsp;0.92 [0.86\u0026ndash;0.96]; 95_HD\u0026thinsp;=\u0026thinsp;11.78 [4.08\u0026ndash;22.04]) in the clinical internal test set (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), showing superior diagnostic performance compared to Junior (P\u0026thinsp;=\u0026thinsp;0.001, 0.012, respectively) and Senior sonographers (both P\u0026thinsp;=\u0026thinsp;0.004). In the external test set, the CNDLM achieved the DSC of 0.74 (0.66\u0026ndash;0.81), TPR of 0.80 (0.73\u0026ndash;0.87), PPV of 0.74(0.65\u0026ndash;0.82), and HD-95 of 14.84 (8.34\u0026ndash;23.11). The performance was significantly better than one of the junior sonographers, but significantly inferior to another junior sonographers (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\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\u003e\u003cb\u003eThe detailed quantitative metrics of the ensembled model and different level sonographers\u003c/b\u003e\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=\"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\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDSC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTPR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHD-95\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eP*\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical internal test set\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\u003eJunior 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.70(0.63\u0026ndash;0.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.79(0.70\u0026ndash;0.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.67(0.58\u0026ndash;0.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e21.19(10.90-34.21)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJunior 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.72(0.66\u0026ndash;0.79)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.90(0.85\u0026ndash;0.94)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.64(0.55\u0026ndash;0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13.70(8.90\u0026ndash;19.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0123\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.79(0.74\u0026ndash;0.83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.94(0.90\u0026ndash;0.97)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.70(0.63\u0026ndash;0.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e15.92(8.12\u0026ndash;26.94)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0040\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.71(0.63\u0026ndash;0.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.92(0.87\u0026ndash;0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.60(0.52\u0026ndash;0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e15.55(11.17\u0026ndash;19.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0040\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNDLM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88(0.84\u0026ndash;0.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87(0.81\u0026ndash;0.92)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92(0.86\u0026ndash;0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11.78(4.08\u0026ndash;22.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical external test set\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\u003eJunior 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.75(0.69\u0026ndash;0.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.75(0.69\u0026ndash;0.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.79(0.72\u0026ndash;0.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13.66(9.17\u0026ndash;19.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0025\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJunior 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.72(0.64\u0026ndash;0.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84(0.78\u0026ndash;0.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.68(0.60\u0026ndash;0.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18.85(12.86\u0026ndash;26.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0025\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.79(0.75\u0026ndash;0.83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.86(0.81\u0026ndash;0.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.78(0.73\u0026ndash;0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11.77(7.99\u0026ndash;16.19)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0025\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.76(0.72\u0026ndash;0.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.94(0.91\u0026ndash;0.97)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.66(0.61\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10.01(7.35\u0026ndash;13.18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0025\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNDLM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.74(0.66\u0026ndash;0.81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.80(0.73\u0026ndash;0.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.74(0.65\u0026ndash;0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.84(8.34\u0026ndash;23.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnimal test set\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\u003eJunior 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.33(0.28\u0026ndash;0.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.60(0.52\u0026ndash;0.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.25(0.20\u0026ndash;0.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e48.21(41.16\u0026ndash;55.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJunior 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.31(0.26\u0026ndash;0.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.54(0.46\u0026ndash;0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.24(0.20\u0026ndash;0.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e67.89(57.06\u0026ndash;78.03)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.64(0.57\u0026ndash;0.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.71(0.65\u0026ndash;0.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.60(0.54\u0026ndash;0.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22.65(16.71\u0026ndash;29.47)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeniority 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.58(0.52\u0026ndash;0.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.65(0.59\u0026ndash;0.71)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.57(0.51\u0026ndash;0.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e38.21(28.52\u0026ndash;47.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNDLM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.91(0.87\u0026ndash;0.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.91(0.86\u0026ndash;0.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.95(0.93\u0026ndash;0.97)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.55(1.85\u0026ndash;3.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e*: Comparing the Dice Similarity Coefficient (DSC) with convolutional neural network-based liver injury diagnostic model (CNDLM). TPR: true positive rate; PPV: true positive rate; HD-95: 95% Hausdorff distance; CNDLM: convolutional neural network-based liver injury diagnostic model\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, a deep learning model was proposed for liver injury detection, which could achieve satisfactory diagnostic performance in assessing clinical US images comparable to that of the trained sonographers. To the best of our knowledge, this is the first study that proposed the clinical DL model for liver trauma evaluation that demonstrated good accuracy. Applying AI methods in the liver US could potentially increase the quality of imaging while limiting the influence of the human factor to achieve an early, rapid, and accurate diagnosis of liver trauma.\u003c/p\u003e \u003cp\u003eThe Bama miniature pig\u0026rsquo;s liver resembles the human liver in terms of anatomy, morphology, and texture echo (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Thus, these animals were selected as the experimental animal in this study. The automatic segmentation results of the model were compared with the standard range manually outlined by the trained sonographers, and DSC evaluated the output performance of the model. It is generally believed that the Dice coefficient\u0026thinsp;\u0026gt;\u0026thinsp;0.7 indicates that the model has better performance (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). In this study, the DSCs were all above 0.70, which confirmed that the automatically segmented liver injury area and the real injury area overlapped. The average positive predictive value (PPV) reached 0.884, and the average true positive rate (TPR) was about 0.780, which is comparable with the previous reports by Sato \u003cem\u003eet al.\u003c/em\u003e(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e) or Saillard \u003cem\u003eet al.\u003c/em\u003e(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e) for the DL models in the diagnosis of hepatocellular carcinoma. The obtained results confirmed that the model not only accurately located the wound location but also outlined the wound contour more reliably.\u003c/p\u003e \u003cp\u003eDespite the promising accuracy, peripheral false positives or internal false negatives were reported when the lesions were over-segmented or under-segmented. This may be at least partly explained by the following two points: firstly, the grayscale of the ultrasound image of liver trauma changes with time. Within a week of the formation of liver trauma, the trauma area often shows a heterogeneous echo area or a slightly hyperechoic area, and sometimes the grayscale contrast with the normal liver parenchyma is not significant. With time, the echo in the lesion area gradually decreases, and the grayscale contrast with normal tissue also gradually increases. Most lesions have echogenicity similar to the surrounding liver parenchyma in about two months (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). Secondly, in the case of slight injury, the grayscale contrast with normal tissue is not significant, and the more severe the injury, the greater the grayscale difference from normal tissue. Therefore, the sensitivity and specificity of the proposed model should be further evaluated based on the above data.\u003c/p\u003e \u003cp\u003eOne of the strengths of this study is the large number of animal US image data collected for training, which were used to guide the model and maintain the existing low-level features through pre-trained weights. Additionally, the proposed AI model was fine-tuned using ultrasound image data of patients to correct the difference between human and animals. This approach could potentially increase the diagnostic accuracy to the level of human sonographers, as was previously reported(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). To test the potential of the model, this study compared the diagnostic accuracy of physicians with different seniority and AI. The results showed that the model outperformed some junior physicians in external test set, and the ensemble model had better diagnostic performance than physicians of all levels in the internal test set, which is consistent with previous studies (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). Thus, in small hospitals with comparable conditions, as well as teaching hospitals that undertake the teaching work of young residents, the use of AI-assisted diagnosis could potentially help to improve the quality of ultrasound diagnosis.\u003c/p\u003e \u003cp\u003eFinally, the concept of \u0026ldquo;golden time\u0026rdquo; is emphasized in trauma first aid, with time becoming the most important factor affecting survival(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). Therefore, the methods that shorten the assessment time before the surgery are most important. In emergency situations, US can effectively guide clinical classification and treatment, but its interpretation is significantly affected by the physiological characteristics of patients(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Moreover, ultrasound examination has certain subjective differences, and training professionals require a lot of long-term practice and learning. In contrast, combining US with AI can simplify operation steps, avoid subjective differences, save physician resources, shorten examination time, and improve diagnostic efficiency. This study demonstrated that the application of AI to diagnosing liver trauma could potentially maximize the portability and speed of real-time US interpretation, which was consistent with previous studies (\u003cspan additionalcitationids=\"CR29\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis study has several limitations. Firstly, the number of clinical US images was limited due to the low incidence of liver trauma, resulting in an imbalance of image sources. For this issue, this study supplemented clinical data to fine-tune the animal model; through data amplification and batch normalization, overfitting was reduced as much as possible, and robustness was enhanced. Secondly, clinical data in this study were collected retrospectively, with could have led to a potential bias. The transfer learning method was adopted due to the lack of clinical US data, which cannot meet the needs of building deep learning models. Although we used methods such as increasing the data set, enhancing the data, reducing the learning rate, and using regularization parameters for data preprocessing, there was still a certain overfitting problem in the model training, which needs to be increased in future research. Finally, clinical data of patients (including weight and blood pressure) were not taken into account at this stage. Future studies should focus on validating the model in the clinical setting.\u003c/p\u003e \u003cp\u003eIn conclusion, this study proposed a deep learning model based on ResNet50 structure with ImageNet pre-trained weights for liver injury detection from 2D ultrasound images, which achieved comparable diagnostic performance with Junior sonographers.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThe study was undertaken in two stages. The first stage included the animal experimental US images as a training set and verification set. The second stage included the clinical US images as a test set. The animal study strictly complied with the National Institute of Animal Health regulation in experimental design, animal feeding, and sample acquisition. This whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01).\u003c/p\u003e\n\u003ch3\u003eEstablishment of the experimental model\u003c/h3\u003e\n\u003cp\u003eThe animal experimental stage took place between January 2021 and April 2021. Eighteen Bama miniature pigs weighing 30\u0026ndash;35 kg underwent intramuscular anesthesia (0.3 ml/kg, mix midazolam and serrazine hydrochloride injection in the ratio of 1:1) and endotracheal intubation. The animals were fixed, the venous channel was established through the marginal ear vein, and the liver parenchyma contusion and laceration were modeled with a self-made percussion device (\u003cb\u003eSupplementary Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e). After modeling, the contrast-enhanced ultrasound (Ultrasound Diagnostic Instrument with 5\u0026thinsp;~\u0026thinsp;15 MHz Convex probe, 3\u0026thinsp;~\u0026thinsp;5MHz frequency) was performed to obtain multiple gray-scale (GS) ultrasound images of the trauma area from different angles. The liver trauma region was defined as region without enhancement in US images, evaluated by a junior sonographer (with less than 3 years of experience).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eClinical data collection\u003c/h2\u003e \u003cp\u003eClinical US images were collected from patients with liver trauma treated in the first medical center of PLA General Hospital, the seventh medical center of PLA General Hospital, and Beijing Chaoyang emergency center between January 2015 and April 2021. The inclusion criteria were: 1) trauma history within one week; 2) liver trauma confirmed by surgery or contrast-enhanced ultrasound; 3) complete clinical and imaging information. The exclusion criteria were: 1) poor image quality; 2) chronic liver lesions (such as liver cirrhosis); 3) non-traumatic liver rupture (such as rupture of liver tumor). The golden standard for liver trauma region was defined as a region of no enhancement in contrast-enhanced US imaging, drawn by the same sonographer. The true positive was defined as the overlap between the liver trauma region defined and the liver trauma region in the gold standard. Thirty clinical US images from the data collected from the first medical center of PLA General Hospital and the seventh medical center of PLA General Hospital were randomly selected as fine tuning set, and the rest images from the two centers were selected as clinical internal test set. The clinical US images collected from Beijing Chaoyang emergency center were selected as clinical external test. This whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01) and informed consent was obtained from all subjects.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eEstablishment and evaluation of the deep learning (DL) model\u003c/h2\u003e \u003cp\u003eUS images obtained from pigs were randomly divided into a training set and validation set with a ratio of 6:1, and a test set was randomly selected. Three animal cross-validation (CV) models were established and evaluated. In addition, 3 clinical CV models and a convolutional neural network-based liver injury diagnostic model (CNDLM) were established by fine-tuning clinical US images from the first center. The performance of 3 CV models was compared using animal training, validation, and test set, and the performance of CNDLM was compared with data obtained by other sonographers using the animal test set, clinical internal validation set, and clinical external validation set.\u003c/p\u003e \u003cp\u003eThe workflow of the segmentation DL model is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The segmentation mask was extracted through the Visual Object Tagging Tool (VOTT, Microsoft, USA) and OpenCV package in python. Then, the images and segmentation mask were resized (224*224 pixels). The model was developed with the RefineNet (\u003cb\u003eSupplementary Figure S2\u003c/b\u003e). A Resnet101 network pre-trained on the ImageNet dataset was employed in RefineNet (\u003cb\u003eSupplementary Figure S3\u003c/b\u003e). The final model was evaluated by internal test cohort and external test cohorts with human US images. In the training process, the parameters of the model were iteratively updated through backpropagation, whose input and output was the image and segmentation mask, respectively. The dice loss of the output and segmentation mask was used as the loss function. The data enhancement strategy was employed to reduce the impact of overfitting, including random flipping and rotation of the input image. A detailed description of the model establishment and evaluation are shown in the Supplementary Methods section.\u003c/p\u003e \u003cp\u003eTo compare the diagnostic efficiency of the final model, the animal and clinical images in validation were evaluated by two junior sonographers (with less than 3 years of experiences) and two senior sonographers (with more than 10 years of experiences).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eR software (version 3.3.1; R 21 Foundation for Statistical Computing, Vienna, Austria) was used for statistical analysis. The Dice Similarity Coefficient (DSC), True Positive Rate (TPR), Positive Predictive Value (PPV), and 95% Hausdorff distance (HD-95) were used to evaluate the performance of CV, CNDLM, and sonographers. Models were compared for DSC in physician identification using the Kruskal-Wallis test. All statistical tests were two-sided, and the statistical significance level was set at two-sided P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author(s) declare no competing interests.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval for animal\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe animalstudy strictly complied with the National Institute of Animal Health regulation in experimental design, animal feeding, and sample acquisition. All methods were carried out in compliance with the ARRIVE guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval for human\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis whole study was approved by the Ethics Committee of the PLA General Hospital (approval number: S2020-323-01) and informed consent was obtained from all subjects. All methods were performed in accordance with the relevant guidelines and regulations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated or analysed during this study are included in this article and its supplementary information file\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by the China Postdoctoral Scientific Foundation (2018M643876). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLYK and SQ conceived and supervised the study; LYK, TL and WK designed experiments; WYJ, MJ, KLL and HP performed experiments; WYJ and HXL analysed data; SQ and HXL wrote the manuscript; GHJ and HXL made manuscript revisions. All authors reviewed the results and approved the final version of the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBuci, S., \u003cem\u003eet al.\u003c/em\u003e The rate of success of the conservative management of liver trauma in a developing country. World J Emerg Surg. 12,24 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTarchouli, M., \u003cem\u003eet al.\u003c/em\u003e Liver trauma: What current management? Hepatobiliary Pancreat Dis Int. 17,39\u0026ndash;44 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaviano, A., \u003cem\u003eet al.\u003c/em\u003e Liver Trauma: Management in the Emergency Setting and Medico-Legal Implications. \u003cem\u003eDiagnostics (Basel)\u003c/em\u003e. 12, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePillai, A.S., Kumar, G. \u0026amp; Pillai, A.K. Hepatic Trauma Interventions. Semin Intervent Radiol. 38,96\u0026ndash;104 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS\u0026aacute;nchez-Bueno, F., \u003cem\u003eet al.\u003c/em\u003e [Changes in the diagnosis and therapeutic management of hepatic trauma. A retrospective study comparing 2 series of cases in different (1997\u0026thinsp;\u0026ndash;\u0026thinsp;1984 vs. 2001\u0026ndash;2008)]. Cir Esp. 89,439\u0026ndash;447 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanani, A., Sandve, K.O. \u0026amp; S\u0026oslash;reide, K. Management of severe liver injuries: push, pack, pringle - and plug! Scand J Trauma Resusc Emerg Med. 29,93 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaylor, J.F., \u003cem\u003eet al.\u003c/em\u003e Emergency department imaging of pediatric trauma patients during combat operations in Iraq and Afghanistan. Pediatr Radiol. 48,620\u0026ndash;625 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePillai, A.S., Srinivas, S., Kumar, G. \u0026amp; Pillai, A.K. Where Does Interventional Radiology Fit in with Trauma Management Algorithm? Semin Intervent Radiol. 38,3\u0026ndash;8 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLada, N.E., \u003cem\u003eet al.\u003c/em\u003e Liver trauma: hepatic vascular injury on computed tomography as a predictor of patient outcome. Eur Radiol. 31,3375\u0026ndash;3382 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong, J., \u003cem\u003eet al.\u003c/em\u003e A deep residual learning network for predicting lung adenocarcinoma manifesting as ground-glass nodule on CT images. Eur Radiol. 30,1847\u0026ndash;1855 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVeerankutty, F.H., \u003cem\u003eet al.\u003c/em\u003e Artificial Intelligence in hepatology, liver surgery and transplantation: Emerging applications and frontiers of research. World J Hepatol. 13,1977\u0026ndash;1990 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai, S.L., \u003cem\u003eet al.\u003c/em\u003e Using a deep learning system in endoscopy for screening of early esophageal squamous cell carcinoma (with video). Gastrointest Endosc. 90,745\u0026ndash;753.e742 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, C., \u003cem\u003eet al.\u003c/em\u003e Development and validation of an endoscopic images-based deep learning model for detection with nasopharyngeal malignancies. Cancer Commun (Lond). 38,59 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhn, S.H., \u003cem\u003eet al.\u003c/em\u003e Comparative clinical evaluation of atlas and deep-learning-based auto-segmentation of organ structures in liver cancer. Radiat Oncol. 14,213 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBiswas, M., \u003cem\u003eet al.\u003c/em\u003e Symtosis: A liver ultrasound tissue characterization and risk stratification in optimized deep learning paradigm. Comput Methods Programs Biomed. 155,165\u0026ndash;177 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRinge, K.I., \u003cem\u003eet al.\u003c/em\u003e Fully automated detection of primary sclerosing cholangitis (PSC)-compatible bile duct changes based on 3D magnetic resonance cholangiopancreatography using machine learning. Eur Radiol. 31,2482\u0026ndash;2489 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmauch, B., \u003cem\u003eet al.\u003c/em\u003e Diagnosis of focal liver lesions from ultrasound using deep learning. Diagn Interv Imaging. 100,227\u0026ndash;233 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYekuo, L., \u003cem\u003eet al.\u003c/em\u003e Contrast-enhanced ultrasound for blunt hepatic trauma: an animal experiment. Am J Emerg Med. 28,828\u0026ndash;833 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, L., \u003cem\u003eet al.\u003c/em\u003e Automatic multi-plaque tracking and segmentation in ultrasonic videos. Med Image Anal. 74,102201 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTajbakhsh, N., \u003cem\u003eet al.\u003c/em\u003e Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation. Med Image Anal. 63,101693 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSato, M., \u003cem\u003eet al.\u003c/em\u003e Machine-learning Approach for the Development of a Novel Predictive Model for the Diagnosis of Hepatocellular Carcinoma. Sci Rep. 9,7704 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaillard, C., \u003cem\u003eet al.\u003c/em\u003e Predicting Survival After Hepatocellular Carcinoma Resection Using Deep Learning on Histological Slides. Hepatology. 72,2000\u0026ndash;2013 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePadalino, P., \u003cem\u003eet al.\u003c/em\u003e Healing of Blunt Liver Injury After Non-Operative Management: Role of Ultrasonography Follow-Up. Eur J Trauma Emerg Surg. 35,364\u0026ndash;370 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, C., \u003cem\u003eet al.\u003c/em\u003e Feasibility of computer-assisted diagnosis for breast ultrasound: the results of the diagnostic performance of S-detect from a single center in China. Cancer Manag Res. 11,921\u0026ndash;930 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, J., \u003cem\u003eet al.\u003c/em\u003e The value of S-Detect for the differential diagnosis of breast masses on ultrasound: a systematic review and pooled meta-analysis. Med Ultrason. 22,211\u0026ndash;219 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHussmann, B. \u0026amp; Lendemans, S. Pre-hospital and early in-hospital management of severe injuries: changes and trends. Injury. 45 Suppl 3,S39-42 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkkus, Z., \u003cem\u003eet al.\u003c/em\u003e A Survey of Deep-Learning Applications in Ultrasound: Artificial Intelligence-Powered Ultrasound for Improving Clinical Workflow. J Am Coll Radiol. 16,1318\u0026ndash;1328 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiompi, F., \u003cem\u003eet al.\u003c/em\u003e Automatic classification of pulmonary peri-fissural nodules in computed tomography using an ensemble of 2D views and a convolutional neural network out-of-the-box. Med Image Anal. 26,195\u0026ndash;202 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, X., \u003cem\u003eet al.\u003c/em\u003e Searching for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus non-deep learning. Sci Rep. 7,15415 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLassau, N., \u003cem\u003eet al.\u003c/em\u003e Five simultaneous artificial intelligence data challenges on ultrasound, CT, and MRI. Diagn Interv Imaging. 100,199\u0026ndash;209 (2019).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Liver Contusions, Lacerations, Deep Learning, Artificial Intelligence, Animal Models, Predictive Value of Tests","lastPublishedDoi":"10.21203/rs.3.rs-4454754/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4454754/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe study aimed to develop an AI-assisted ultrasound model for early liver trauma identification, using data from Bama miniature pigs and patients in Beijing, China. A deep learning model was created and fine-tuned with animal and clinical data, achieving high accuracy metrics. In internal tests, the model outperformed both Junior and Senior sonographers. External tests showed the model's effectiveness, with a Dice Similarity Coefficient of 0.74, True Positive Rate of 0.80, Positive Predictive Value of 0.74, and 95% Hausdorff distance of 14.84. The model's performance was comparable to Junior sonographers and slightly lower than Senior sonographers. This AI model shows promise for liver injury detection, offering a valuable tool with diagnostic capabilities similar to those of less experienced human operators.\u003c/p\u003e","manuscriptTitle":"AI-assisted ultrasound for early liver trauma: Animal models \u0026amp; clinical validation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-06-18 16:37:59","doi":"10.21203/rs.3.rs-4454754/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-07-29T04:53:11+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-25T22:41:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-10T16:25:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"199710151380496820712198361781949710048","date":"2024-07-09T15:39:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"255428616708882516790635963837336959930","date":"2024-07-09T14:47:24+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-07-09T14:08:28+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-07-09T13:57:31+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-06-04T13:06:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-31T10:54:00+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-05-21T12:16:12+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8c7ccab5-bb7e-46d3-9503-62e44849b7ce","owner":[],"postedDate":"June 18th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":33034532,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":33034533,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"},{"id":33034534,"name":"Health sciences/Diseases"}],"tags":[],"updatedAt":"2025-07-07T16:05:07+00:00","versionOfRecord":{"articleIdentity":"rs-4454754","link":"https://doi.org/10.1038/s41598-025-91900-5","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-07-02 15:58:04","publishedOnDateReadable":"July 2nd, 2025"},"versionCreatedAt":"2024-06-18 16:37:59","video":"","vorDoi":"10.1038/s41598-025-91900-5","vorDoiUrl":"https://doi.org/10.1038/s41598-025-91900-5","workflowStages":[]},"version":"v1","identity":"rs-4454754","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4454754","identity":"rs-4454754","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}