Project description:IntroductionRadiographic bone age (BA) assessment is widely used to evaluate children's growth disorders and predict their future height. Moreover, children are more sensitive and vulnerable to X-ray radiation exposure than adults. The purpose of this study is to develop a new, safer, radiation-free BA assessment method for children by using three-dimensional ultrasound (3D-US) and artificial intelligence (AI), and to test the diagnostic accuracy and reliability of this method.Methods and analysisThis is a prospective, observational study. All participants will be recruited through Paediatric Growth and Development Clinic. All participants will receive left hand 3D-US and X-ray examination at the Shanghai Sixth People's Hospital on the same day, all images will be recorded. These image related data will be collected and randomly divided into training set (80% of all) and test set (20% of all). The training set will be used to establish a cascade network of 3D-US skeletal image segmentation and BA prediction model to achieve end-to-end prediction of image to BA. The test set will be used to evaluate the accuracy of AI BA model of 3D-US. We have developed a new ultrasonic scanning device, which can be proposed to automatic 3D-US scanning of hands. AI algorithms, such as convolutional neural network, will be used to identify and segment the skeletal structures in the hand 3D-US images. We will achieve automatic segmentation of hand skeletal 3D-US images, establish BA prediction model of 3D-US, and test the accuracy of the prediction model.Ethics and disseminationThe Ethics Committee of Shanghai Sixth People's Hospital approved this study. The approval number is 2022-019. A written informed consent will be obtained from their parent or guardian of each participant. Final results will be published in peer-reviewed journals and presented at national and international conferences.Trial registration numberChiCTR2200057236.
Project description:ObjectiveBone age (BA) is a crucial indicator for revealing the growth and development of children. This study tested the performance of a fully automated artificial intelligence (AI) system for BA assessment of Chinese children with abnormal growth and development.Materials and methodsA fully automated AI system based on the Greulich and Pyle (GP) method was developed for Chinese children by using 8,000 BA radiographs from five medical centers nationwide in China. Then, a total of 745 cases (360 boys and 385 girls) with abnormal growth and development from another tertiary medical center of north China were consecutively collected between January and October 2018 to test the system. The reference standard was defined as the result interpreted by two experienced reviewers (a radiologist with 10 years and an endocrinologist with 15 years of experience in BA reading) through consensus using the GP atlas. BA accuracy within 1 year, root mean square error (RMSE), mean absolute difference (MAD), and 95% limits of agreement according to the Bland-Altman plot were statistically calculated.ResultsFor Chinese pediatric patients with abnormal growth and development, the accuracy of this new automated AI system within 1 year was 84.60% as compared to the reference standard, with the highest percentage of 89.45% in the 12- to 18-year group. The RMSE, MAD, and 95% limits of agreement of the AI system were 0.76 years, 0.58 years, and -1.547 to 1.428, respectively, according to the Bland-Altman plot. The largest difference between the AI and experts' BA result was noted for patients of short stature with bone deformities, severe osteomalacia, or different rates of maturation of the carpals and phalanges.ConclusionsThe developed automated AI system could achieve comparable BA results to experienced reviewers for Chinese children with abnormal growth and development.
Project description:BackgroundThe autonomous artificial intelligence (AI) system for bone age rating (BoneXpert) was designed to be used in clinical radiology practice as an AI-replace tool, replacing the radiologist completely.ObjectiveThe aim of this study was to investigate how the tool is used in clinical practice. Are radiologists more inclined to use BoneXpert to assist rather than replace themselves, and how much time is saved?Materials and methodsWe sent a survey consisting of eight multiple-choice questions to 282 radiologists in departments in Europe already using the software.ResultsThe 97 (34%) respondents came from 18 countries. Their answers revealed that before installing the automated method, 83 (86%) of the respondents took more than 2 min per bone age rating; this fell to 20 (21%) respondents after installation. Only 17/97 (18%) respondents used BoneXpert to completely replace the radiologist; the rest used it to assist radiologists to varying degrees. For instance, 39/97 (40%) never overruled the automated reading, while 9/97 (9%) overruled more than 5% of the automated ratings. The majority 58/97 (60%) of respondents checked the radiographs themselves to exclude features of underlying disease.ConclusionBoneXpert significantly reduces reporting times for bone age determination. However, radiographic analysis involves more than just determining bone age. It also involves identification of abnormalities, and for this reason, radiologists cannot be completely replaced. AI systems originally developed to replace the radiologist might be more suitable as AI assist tools, particularly if they have not been validated to work autonomously, including the ability to omit ratings when the image is outside the range of validity.
Project description:BackgroundIn 2020, our center established a Tanner-Whitehouse 3 (TW3) artificial intelligence (AI) system using a convolutional neural network (CNN), which was built upon 9059 radiographs. However, the system, upon which our study is based, lacked a gold standard for comparison and had not undergone thorough evaluation in different working environments.MethodsTo further verify the applicability of the AI system in clinical bone age assessment (BAA) and to enhance the accuracy and homogeneity of BAA, a prospective multi-center validation was conducted. This study utilized 744 left-hand radiographs of patients, ranging from 1 to 20 years of age, with 378 boys and 366 girls. These radiographs were obtained from nine different children's hospitals between August and December 2020. The BAAs were performed using the TW3 AI system and were also reviewed by experienced reviewers. Bone age accuracy within 1 year, root mean square error (RMSE), and mean absolute error (MAE) were statistically calculated to evaluate the accuracy. Kappa test and Bland-Altman (B-A) plot were conducted to measure the diagnostic consistency.ResultsThe system exhibited a high level of performance, producing results that closely aligned with those of the reviewers. It achieved a RMSE of 0.52 years and an accuracy of 94.55% for the radius, ulna, and short bones series. When assessing the carpal series of bones, the system achieved a RMSE of 0.85 years and an accuracy of 80.38%. Overall, the system displayed satisfactory accuracy and RMSE, particularly in patients over 7 years old. The system excelled in evaluating the carpal bone age of patients aged 1-6. Both the Kappa test and B-A plot demonstrated substantial consistency between the system and the reviewers, although the model encountered challenges in consistently distinguishing specific bones, such as the capitate. Furthermore, the system's performance proved acceptable across different genders and age groups, as well as radiography instruments.ConclusionsIn this multi-center validation, the system showcased its potential to enhance the efficiency and consistency of healthy delivery, ultimately resulting in improved patient outcomes and reduced healthcare costs.
Project description:Adenoid hypertrophy may lead to pediatric obstructive sleep apnea and mouth breathing. The routine screening of adenoid hypertrophy in dental practice is helpful for preventing relevant craniofacial and systemic consequences. The purpose of this study was to develop an automated assessment tool for adenoid hypertrophy based on artificial intelligence. A clinical dataset containing 581 lateral cephalograms was used to train the convolutional neural network (CNN). According to Fujioka's method for adenoid hypertrophy assessment, the regions of interest were defined with four keypoint landmarks. The adenoid ratio based on the four landmarks was used for adenoid hypertrophy assessment. Another dataset consisting of 160 patients' lateral cephalograms were used for evaluating the performance of the network. Diagnostic performance was evaluated with statistical analysis. The developed system exhibited high sensitivity (0.906, 95% confidence interval [CI]: 0.750-0.980), specificity (0.938, 95% CI: 0.881-0.973) and accuracy (0.919, 95% CI: 0.877-0.961) for adenoid hypertrophy assessment. The area under the receiver operating characteristic curve was 0.987 (95% CI: 0.974-1.000). These results indicated the proposed assessment system is able to assess AH accurately. The CNN-incorporated system showed high accuracy and stability in the detection of adenoid hypertrophy from children' lateral cephalograms, implying the feasibility of automated adenoid hypertrophy screening utilizing a deep neural network model.
Project description:BackgroundThe accuracy and consistency of bone age assessments (BAA) using standard methods can vary with physicians' level of experience.MethodsTo assess the impact of information from an artificial intelligence (AI) deep learning convolutional neural network (CNN) model on BAA, specialists with different levels of experience (junior, mid-level, and senior) assessed radiographs from 316 children aged 4-18 years that had been randomly divided into two equal sets-group A and group B. Bone age (BA) was assessed independently by each specialist without additional information (group A) and with information from the model (group B). With the mean assessment of four experts as the reference standard, mean absolute error (MAE), and intraclass correlation coefficient (ICC) were calculated to evaluate accuracy and consistency. Individual assessments of 13 bones (radius, ulna, and short bones) were also compared between group A and group B with the rank-sum test.ResultsThe accuracies of senior, mid-level, and junior physicians were significantly better (all P < 0.001) with AI assistance (MAEs 0.325, 0.344, and 0.370, respectively) than without AI assistance (MAEs 0.403, 0.469, and 0.755, respectively). Moreover, for senior, mid-level, and junior physicians, consistency was significantly higher (all P < 0.001) with AI assistance (ICCs 0.996, 0.996, and 0.992, respectively) than without AI assistance (ICCs 0.987, 0.989, and 0.941, respectively). For all levels of experience, accuracy with AI assistance was significantly better than accuracy without AI assistance for assessments of the first and fifth proximal phalanges.ConclusionsInformation from an AI model improves both the accuracy and the consistency of bone age assessments for physicians of all levels of experience. The first and fifth proximal phalanges are difficult to assess, and they should be paid more attention.
Project description:Artificial intelligence (AI) based on convolutional neural networks (CNNs) has a great potential to enhance medical workflow and improve health care quality. Of particular interest is practical implementation of such AI-based software as a cloud-based tool aimed for telemedicine, the practice of providing medical care from a distance using electronic interfaces. Methods: In this study, we used a dataset of labeled 35,900 optical coherence tomography (OCT) images obtained from age-related macular degeneration (AMD) patients and used them to train three types of CNNs to perform AMD diagnosis. Results: Here, we present an AI- and cloud-based telemedicine interaction tool for diagnosis and proposed treatment of AMD. Through deep learning process based on the analysis of preprocessed optical coherence tomography (OCT) imaging data, our AI-based system achieved the same image discrimination rate as that of retinal specialists in our hospital. The AI platform's detection accuracy was generally higher than 90% and was significantly superior (p < 0.001) to that of medical students (69.4% and 68.9%) and equal (p = 0.99) to that of retinal specialists (92.73% and 91.90%). Furthermore, it provided appropriate treatment recommendations comparable to those of retinal specialists. Conclusions: We therefore developed a website for realistic cloud computing based on this AI platform, available at https://www.ym.edu.tw/~AI-OCT/. Patients can upload their OCT images to the website to verify whether they have AMD and require treatment. Using an AI-based cloud service represents a real solution for medical imaging diagnostics and telemedicine.
Project description:Purpose of reviewPredicting treatment response and optimizing treatment regimen in patients with neovascular age-related macular degeneration (nAMD) remains challenging. Artificial intelligence-based tools have the potential to increase confidence in clinical development of new therapeutics, facilitate individual prognostic predictions, and ultimately inform treatment decisions in clinical practice.Recent findingsTo date, most advances in applying artificial intelligence to nAMD have focused on facilitating image analysis, particularly for automated segmentation, extraction, and quantification of imaging-based features from optical coherence tomography (OCT) images. No studies in our literature search evaluated whether artificial intelligence could predict the treatment regimen required for an optimal visual response for an individual patient. Challenges identified for developing artificial intelligence-based models for nAMD include the limited number of large datasets with high-quality OCT data, limiting the patient populations included in model development; lack of counterfactual data to inform how individual patients may have fared with an alternative treatment strategy; and absence of OCT data standards, impairing the development of models usable across devices.SummaryArtificial intelligence has the potential to enable powerful prognostic tools for a complex nAMD treatment landscape; however, additional work remains before these tools are applicable to informing treatment decisions for nAMD in clinical practice.
Project description:BackgroundSurgical site infection (SSI) is one of the most common types of health care-associated infections. It increases mortality, prolongs hospital length of stay, and raises health care costs. Many institutions developed risk assessment models for SSI to help surgeons preoperatively identify high-risk patients and guide clinical intervention. However, most of these models had low accuracies.ObjectiveWe aimed to provide a solution in the form of an Artificial intelligence-based Multimodal Risk Assessment Model for Surgical site infection (AMRAMS) for inpatients undergoing operations, using routinely collected clinical data. We internally and externally validated the discriminations of the models, which combined various machine learning and natural language processing techniques, and compared them with the National Nosocomial Infections Surveillance (NNIS) risk index.MethodsWe retrieved inpatient records between January 1, 2014, and June 30, 2019, from the electronic medical record (EMR) system of Rui Jin Hospital, Luwan Branch, Shanghai, China. We used data from before July 1, 2018, as the development set for internal validation and the remaining data as the test set for external validation. We included patient demographics, preoperative lab results, and free-text preoperative notes as our features. We used word-embedding techniques to encode text information, and we trained the LASSO (least absolute shrinkage and selection operator) model, random forest model, gradient boosting decision tree (GBDT) model, convolutional neural network (CNN) model, and self-attention network model using the combined data. Surgeons manually scored the NNIS risk index values.ResultsFor internal bootstrapping validation, CNN yielded the highest mean area under the receiver operating characteristic curve (AUROC) of 0.889 (95% CI 0.886-0.892), and the paired-sample t test revealed statistically significant advantages as compared with other models (P<.001). The self-attention network yielded the second-highest mean AUROC of 0.882 (95% CI 0.878-0.886), but the AUROC was only numerically higher than the AUROC of the third-best model, GBDT with text embeddings (mean AUROC 0.881, 95% CI 0.878-0.884, P=.47). The AUROCs of LASSO, random forest, and GBDT models using text embeddings were statistically higher than the AUROCs of models not using text embeddings (P<.001). For external validation, the self-attention network yielded the highest AUROC of 0.879. CNN was the second-best model (AUROC 0.878), and GBDT with text embeddings was the third-best model (AUROC 0.872). The NNIS risk index scored by surgeons had an AUROC of 0.651.ConclusionsOur AMRAMS based on EMR data and deep learning methods-CNN and self-attention network-had significant advantages in terms of accuracy compared with other conventional machine learning methods and the NNIS risk index. Moreover, the semantic embeddings of preoperative notes improved the model performance further. Our models could replace the NNIS risk index to provide personalized guidance for the preoperative intervention of SSIs. Through this case, we offered an easy-to-implement solution for building multimodal RAMs for other similar scenarios.