INTRODUCTION: In this study, we aimed to compare the performance of a convolutional neural network (CNN)-based deep learning model that was trained on a dataset of normal and abnormal paediatric elbow radiographs with that of paediatric emergency department (ED) physicians on a binomial classification task. METHODS: A total of 1,314 paediatric elbow lateral radiographs (patient mean age 8.2 years) were retrospectively retrieved and classified based on annotation as normal or abnormal (with pathology). They were then randomly partitioned to a development set (993 images); first and second tuning (validation) sets (109 and 100 images, respectively); and a test set (112 images). An artificial intelligence (AI) model was trained on the development set using the EfficientNet B1 network architecture. Its performance on the test set was compared to that of five physicians (inter-rater agreement: fair). Performance of the AI model and the physician group was tested using McNemar test. RESULTS: The accuracy of the AI model on the test set was 80.4% (95% confidence interval [CI] 71.8%-87.3%), and the area under the receiver operating characteristic curve (AUROC) was 0.872 (95% CI 0.831-0.947). The performance of the AI model vs. the physician group on the test set was: sensitivity 79.0% (95% CI: 68.4%-89.5%) vs. 64.9% (95% CI: 52.5%-77.3%; P = 0.088); and specificity 81.8% (95% CI: 71.6%-92.0%) vs. 87.3% (95% CI: 78.5%-96.1%; P = 0.439). CONCLUSION The AI model showed good AUROC values and higher sensitivity, with the P-value at nominal significance when compared to the clinician group.
Humans ; Deep Learning ; Child ; Retrospective Studies ; Male ; Female ; Radiography/methods* ; ROC Curve ; Elbow/diagnostic imaging* ; Neural Networks, Computer ; Child, Preschool ; Elbow Joint/diagnostic imaging* ; Emergency Service, Hospital ; Adolescent ; Infant ; Artificial Intelligence

OBJECTIVE: To investigate the effect of manual acupuncture manipulations (MAMs) on subcutaneous muscle tissue, by developing quantitative models of "lifting and thrusting" and "twisting and rotating", based on machine learning techniques. METHODS: A depth camera was used to capture the acupuncture operator's hand movements during "lifting and thrusting" and "twisting and rotating" of needle. Simultaneously, the ultrasound imaging was employed to record the muscle tissue responses of the participants. Amplitude and angular features were extracted from the movement data of operators, and muscle fascicle slope features were derived from the data of ultrasound images. The dynamic time warping barycenter averaging algorithm was adopted to align the dual-source data. Various machine learning techniques were applied to build quantitative models, and the performance of each model was compared. The most optimal model was further analyzed for its interpretability. RESULTS: Among the quantitative models built for the two types of MAMs, the random forest model demonstrated the best performance. For the quantitative model of the "lifting and thrusting" technique, the coefficient of determination (R²) was 0.825. For the "twisting and rotating" technique, R² reached 0.872. CONCLUSION Machine learning can be used to effectively develop the models and quantify the effects of MAMs on subcutaneous muscle tissue. It provides a new perspective to understand the mechanism of acupuncture therapy and lays a foundation for optimizing acupuncture technology and designing personalized treatment regimen in the future.
Humans ; Acupuncture Therapy/methods* ; Machine Learning ; Male ; Adult ; Female ; Subcutaneous Tissue/diagnostic imaging* ; Young Adult

INTRODUCTION: Interpretation and analysis of magnetic resonance imaging (MRI) scans in clinical settings comprise time-consuming visual ratings and complex neuroimage processing that require trained professionals. To combat these challenges, artificial intelligence (AI) techniques can aid clinicians in interpreting brain MRI for accurate diagnosis of neurodegenerative diseases but they require extensive validation. Thus, the aim of this study was to validate the use of AI-based AQUA (Neurophet Inc., Seoul, Republic of Korea) segmentation software in a Southeast Asian community-based cohort with normal cognition, mild cognitive impairment (MCI) and dementia. METHOD: Study participants belonged to the community-based Biomarker and Cognition Study in Singapore. Participants aged between 30 and 95 years, having cognitive concerns, with no diagnosis of major psychiatric, neurological or systemic disorders who were recruited consecutively between April 2022 and July 2023 were included. Participants underwent neuropsychological assessments and structural MRI, and were classified as cognitively normal, with MCI or with dementia. MRI pre-processing using automated pipelines, along with human-based visual ratings, were compared against AI-based automated AQUA output. Default mode network grey matter (GM) volumes were compared between cognitively normal, MCI and dementia groups. RESULTS: A total of 90 participants (mean age at visit was 63.32±10.96 years) were included in the study (30 cognitively normal, 40 MCI and 20 dementia). Non-parametric Spearman correlation analysis indicated that AQUA-based and human-based visual ratings were correlated with total (ρ=0.66; P<0.0001), periventricular (ρ=0.50; P<0.0001) and deep (ρ=0.57; P<0.0001) white matter hyperintensities (WMH). Additionally, volumetric WMH obtained from AQUA and automated pipelines was also strongly correlated (ρ=0.84; P<0.0001) and these correlations remained after controlling for age at visit, sex and diagnosis. Linear regression analyses illustrated significantly different AQUA-derived default mode network GM volumes between cognitively normal, MCI and dementia groups. Dementia participants had significant atrophy in the posterior cingulate cortex compared to cognitively normal participants (P=0.021; 95% confidence interval [CI] -1.25 to -0.08) and in the hippocampus compared to cognitively normal (P=0.0049; 95% CI -1.05 to -0.16) and MCI participants (P=0.0036; 95% CI -1.02 to -0.17). CONCLUSION Our findings demonstrate high concordance between human-based visual ratings and AQUA-based ratings of WMH. Additionally, the AQUA GM segmentation pipeline showed good differentiation in key regions between cognitively normal, MCI and dementia participants. Based on these findings, the automated AQUA software could aid clinicians in examining MRI scans of patients with cognitive impairment.
Humans ; Cognitive Dysfunction/pathology* ; Magnetic Resonance Imaging/methods* ; Male ; Middle Aged ; Female ; Aged ; Artificial Intelligence ; Software ; Dementia/diagnostic imaging* ; Aged, 80 and over ; Adult ; Singapore ; Neuropsychological Tests ; Brain/pathology* ; Cohort Studies ; Gray Matter/pathology* ; Southeast Asian People

BACKGROUND: Iron deposition plays a crucial role in the pathophysiology of Parkinson's disease (PD), yet the distribution pattern of iron deposition in the subcortical nuclei has been inconsistent across previous studies. We aimed to assess the difference patterns of iron deposition detected by quantitative iron-sensitive magnetic resonance imaging (MRI) between patients with PD and patients with atypical parkinsonian syndromes (APSs), and between patients with PD and healthy controls (HCs). METHODS: A systematic literature search was conducted on PubMed, Embase, and Web of Science databases to identify studies investigating the iron content in PD patients using the iron-sensitive MRI techniques (R2 * and quantitative susceptibility mapping [QSM]), up until May 1, 2023. The quality assessment of case-control and cohort studies was performed using the Newcastle-Ottawa Scale, whereas diagnostic studies were assessed using the Quality Assessment of Diagnostic Accuracy Studies-2. Standardized mean differences and summary estimates of sensitivity, specificity, and area under the curve (AUC) were calculated for iron content, using a random effects model. We also conducted the subgroup-analysis based on the MRI sequence and meta-regression. RESULTS: Seventy-seven studies with 3192 PD, 209 multiple system atrophy (MSA), 174 progressive supranuclear palsy (PSP), and 2447 HCs were included. Elevated iron content in substantia nigra (SN) pars reticulata ( P <0.001) and compacta ( P <0.001), SN ( P <0.001), red nucleus (RN, P <0.001), globus pallidus ( P <0.001), putamen (PUT, P = 0.021), and thalamus ( P = 0.029) were found in PD patients compared with HCs. PD patients showed lower iron content in PUT ( P <0.001), RN ( P = 0.003), SN ( P = 0.017), and caudate nucleus ( P = 0.017) than MSA patients, and lower iron content in RN ( P = 0.001), PUT ( P <0.001), globus pallidus ( P = 0.004), SN ( P = 0.015), and caudate nucleus ( P = 0.001) than PSP patients. The highest diagnostic accuracy distinguishing PD from HCs was observed in SN (AUC: 0.85), and that distinguishing PD from MSA was found in PUT (AUC: 0.90). In addition, the best diagnostic performance was achieved in the RN for distinguishing PD from PSP (AUC: 0.86). CONCLUSIONS: Quantitative iron-sensitive MRI could quantitatively detect the iron content of subcortical nuclei in PD and APSs, while it may be insufficient to accurately diagnose PD. Future studies are needed to explore the role of multimodal MRI in the diagnosis of PD. REGISTRISION PROSPERO (CRD42022344413).
Humans ; Parkinson Disease/diagnostic imaging* ; Magnetic Resonance Imaging/methods* ; Iron/metabolism*

BACKGROUND: Neoadjuvant chemoradiotherapy followed by radical surgery has been a common practice for patients with locally advanced rectal cancer, but the response rate varies among patients. This study aimed to develop a ResNet-Vision Transformer based magnetic resonance imaging (MRI)-endoscopy fusion model to precisely predict treatment response and provide personalized treatment. METHODS: In this multicenter study, 366 eligible patients who had undergone neoadjuvant chemoradiotherapy followed by radical surgery at eight Chinese tertiary hospitals between January 2017 and June 2024 were recruited, with 2928 pretreatment colonic endoscopic images and 366 pelvic MRI images. An MRI-endoscopy fusion model was constructed based on the ResNet backbone and Transformer network using pretreatment MRI and endoscopic images. Treatment response was defined as good response or non-good response based on the tumor regression grade. The Delong test and the Hanley-McNeil test were utilized to compare prediction performance among different models and different subgroups, respectively. The predictive performance of the MRI-endoscopy fusion model was comprehensively validated in the test sets and was further compared to that of the single-modal MRI model and single-modal endoscopy model. RESULTS: The MRI-endoscopy fusion model demonstrated favorable prediction performance. In the internal validation set, the area under the curve (AUC) and accuracy were 0.852 (95% confidence interval [CI]: 0.744-0.940) and 0.737 (95% CI: 0.712-0.844), respectively. Moreover, the AUC and accuracy reached 0.769 (95% CI: 0.678-0.861) and 0.729 (95% CI: 0.628-0.821), respectively, in the external test set. In addition, the MRI-endoscopy fusion model outperformed the single-modal MRI model (AUC: 0.692 [95% CI: 0.609-0.783], accuracy: 0.659 [95% CI: 0.565-0.775]) and the single-modal endoscopy model (AUC: 0.720 [95% CI: 0.617-0.823], accuracy: 0.713 [95% CI: 0.612-0.809]) in the external test set. CONCLUSION The MRI-endoscopy fusion model based on ResNet-Vision Transformer achieved favorable performance in predicting treatment response to neoadjuvant chemoradiotherapy and holds tremendous potential for enabling personalized treatment regimens for locally advanced rectal cancer patients.
Humans ; Rectal Neoplasms/diagnostic imaging* ; Magnetic Resonance Imaging/methods* ; Male ; Female ; Middle Aged ; Neoadjuvant Therapy/methods* ; Aged ; Adult ; Chemoradiotherapy/methods* ; Endoscopy/methods* ; Treatment Outcome

Artificial intelligence (AI), particularly deep learning, has demonstrated remarkable performance in medical imaging across a variety of modalities, including X-ray, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, positron emission tomography (PET), and pathological imaging. However, most existing state-of-the-art AI techniques are task-specific and focus on a limited range of imaging modalities. Compared to these task-specific models, emerging foundation models represent a significant milestone in AI development. These models can learn generalized representations of medical images and apply them to downstream tasks through zero-shot or few-shot fine-tuning. Foundation models have the potential to address the comprehensive and multifactorial challenges encountered in clinical practice. This article reviews the clinical applications of both task-specific and foundation models, highlighting their differences, complementarities, and clinical relevance. We also examine their future research directions and potential challenges. Unlike the replacement relationship seen between deep learning and traditional machine learning, task-specific and foundation models are complementary, despite inherent differences. While foundation models primarily focus on segmentation and classification, task-specific models are integrated into nearly all medical image analyses. However, with further advancements, foundation models could be applied to other clinical scenarios. In conclusion, all indications suggest that task-specific and foundation models, especially the latter, have the potential to drive breakthroughs in medical imaging, from image processing to clinical workflows.
Humans ; Artificial Intelligence ; Deep Learning ; Diagnostic Imaging/methods* ; Magnetic Resonance Imaging ; Tomography, X-Ray Computed ; Positron-Emission Tomography

BACKGROUND: Gait impairment is closely related to quality of life in patients with Parkinson's disease (PD). This study aimed to explore alterations in brain microstructure in PD patients and healthy controls (HCs) and to identify the correlation of gait impairment in the ON and OFF states of patients with PD, respectively. METHODS: We enrolled 24 PD patients and 29 HCs from the Movement Disorders Program at Beijing Friendship Hospital Capital Medical University between 2019 and 2020. We acquired magnetic resonance imaging (MRI) scans and processed the diffusion kurtosis imaging (DKI) images. Preprocessing of diffusion-weighted data was performed with Mrtrix3 software, using a directional distribution function to track participants' main white matter fiber bundles. Demographic and clinical characteristics were recorded. Quantitative gait and clinical scales were used to assess the status of medication ON and OFF in PD patients. RESULTS: The axial kurtosis (AK), mean kurtosis (MK), and radial kurtosis (RK) of five specific white matter fiber tracts, the bilateral corticospinal tract, left superior longitudinal fasciculus, left anterior thalamic radiation, forceps minor, and forceps major were significantly higher in PD patients compared to HCs. Additionally, the MK values were negatively correlated with Timed Up and Go Test (TUG) scores in both the ON and OFF in PD patients. Within the PD group, higher AK, MK, and RK values, whether the patients were ON or OFF, were associated with better gait performance (i.e., higher velocity and stride length). CONCLUSIONS PD exhibits characteristic regional patterns of white matter microstructural degradation. Correlations between objective gait parameters and DKI values suggest that dopamine-responsive gait function depends on preserved white matter microstructure. DKI-based Tract-Based Spatial Statistics (TBSS) analysis may serve as a tool for evaluating PD-related motor impairments (e.g., gait impairment) and could yield potential neuroimaging biomarkers.
Humans ; Parkinson Disease/diagnostic imaging* ; White Matter/physiopathology* ; Male ; Female ; Middle Aged ; Aged ; Gait/physiology* ; Diffusion Magnetic Resonance Imaging/methods* ; Diffusion Tensor Imaging/methods*

Recent advances in large models demonstrate significant prospects for transforming the field of medical imaging. These models, including large language models, large visual models, and multimodal large models, offer unprecedented capabilities in processing and interpreting complex medical data across various imaging modalities. By leveraging self-supervised pretraining on vast unlabeled datasets, cross-modal representation learning, and domain-specific medical knowledge adaptation through fine-tuning, large models can achieve higher diagnostic accuracy and more efficient workflows for key clinical tasks. This review summarizes the concepts, methods, and progress of large models in medical imaging, highlighting their potential in precision medicine. The article first outlines the integration of multimodal data under large model technologies, approaches for training large models with medical datasets, and the need for robust evaluation metrics. It then explores how large models can revolutionize applications in critical tasks such as image segmentation, disease diagnosis, personalized treatment strategies, and real-time interactive systems, thus pushing the boundaries of traditional imaging analysis. Despite their potential, the practical implementation of large models in medical imaging faces notable challenges, including the scarcity of high-quality medical data, the need for optimized perception of imaging phenotypes, safety considerations, and seamless integration with existing clinical workflows and equipment. As research progresses, the development of more efficient, interpretable, and generalizable models will be critical to ensuring their reliable deployment across diverse clinical environments. This review aims to provide insights into the current state of the field and provide directions for future research to facilitate the broader adoption of large models in clinical practice.
Humans ; Diagnostic Imaging/methods* ; Precision Medicine/methods* ; Image Processing, Computer-Assisted/methods*

With the emergence of deep learning techniques based on convolutional neural networks, artificial intelligence (AI) has driven transformative developments in the field of medical image analysis. Recently, large language models (LLMs) such as ChatGPT have also started to achieve distinction in this domain. Increasing research shows the undeniable role of AI in reshaping various aspects of medical image analysis, including processes such as image enhancement, segmentation, detection in image preprocessing, and postprocessing related to medical diagnosis and prognosis in clinical settings. However, despite the significant progress in AI research, studies investigating the recent advances in AI technology in the aforementioned aspects, the changes in research hotspot trajectories, and the performance of studies in addressing key clinical challenges in this field are limited. This article provides an overview of recent advances in AI for medical image analysis and discusses the methodological profiles, advantages, disadvantages, and future trends of AI technologies.
Artificial Intelligence ; Humans ; Image Processing, Computer-Assisted/methods* ; Neural Networks, Computer ; Deep Learning ; Diagnostic Imaging/methods*

Synthetic CT (sCT) generated from CBCT has proven effective in artifact reduction and CT number correction, facilitating precise radiation dose calculation. However, the quality of different regions in sCT images is severely imbalanced, with soft tissue region exhibiting notably inferior quality compared to others. To address this imbalance, we proposed a Multi-Task Attention Network (MuTA-Net) based on VGG-16, specifically focusing the enhancement of image quality in soft tissue region of sCT. First, we introduced a multi-task learning strategy that divides the sCT generation task into three sub-tasks: global image generation, soft tissue region generation and bone region segmentation. This approach ensured the quality of overall sCT image while enhancing the network's focus on feature extraction and generation for soft tissues region. The result of bone region segmentation task guided the fusion of sub-tasks results. Then, we designed an attention module to further optimize feature extraction capabilities of the network. Finally, by employing a results fusion module, the results of three sub-tasks were integrated, generating a high-quality sCT image. Experimental results on head and neck CBCT demonstrated that the sCT images generated by the proposed MuTA-Net exhibited a 12.52% reduction in mean absolute error in soft tissue region, compared to the best performance among the three comparative methods, including ResNet, U-Net, and U-Net++. It can be seen that MuTA-Net is suitable for high-quality sCT image generation and has potential application value in the field of CBCT guided adaptive radiation therapy.
Cone-Beam Computed Tomography/methods* ; Humans ; Image Processing, Computer-Assisted/methods* ; Artifacts ; Algorithms ; Bone and Bones/diagnostic imaging* ; Neural Networks, Computer