Established methods for bone age assessment (BAA), such as the Greulich and Pyle atlas, suffer from variability due to population differences and observer discrepancies. Although automated BAA offers speed and consistency, limited research exists on its performance across different populations using deep learning. This study examines deep learning algorithms on the Turkish population to enhance bone age models by understanding demographic influences. We analyzed reports from Bağcılar Hospital's Health Information Management System between April 2012 and September 2023 using "bone age" as a keyword. Patient images were re-evaluated by an experienced radiologist and anonymized. A total of 2,730 hand radiographs from Bağcılar Hospital (Turkish population), 12,572 from the Radiological Society of North America (RSNA), and 6,185 from the Radiological Hand Pose Estimation (RHPE) public datasets were collected, along with corresponding bone ages and gender information. A random set of 546 radiographs (273 from Bağcılar, 273 from public datasets) was initially randomly split for an internal test set with bone age stratification; the remaining data were used for training and validation. BAAs were generated using a modified InceptionV3 model on 500 × 500-pixel images, selecting the model with the lowest mean absolute error (MAE) on the validation set. Three models were trained and tested based on dataset origin: Bağcılar (Turkish), public (RSNA-RHPE), and a Combined model. Internal test set predictions of the Combined model estimated bone age within less than 6, 12, 18, and 24 months at rates of 44%, 73%, 87%, and 94%, respectively. The MAE was 9.2 months in the overall internal test set, 7 months on the public test set, and 11.5 months on the Bağcılar internal test data. The Bağcılar-only model had an MAE of 12.7 months on the Bağcılar internal test data. Despite less training data, there was no significant difference between the combined and Bağcılar models on the Bağcılar dataset (<i>P</i> > 0.05). The public model showed an MAE of 16.5 months on the Bağcılar dataset, significantly worse than the other models (<i>P</i> < 0.05). We developed an automatic BAA model including the Turkish population, one of the few such studies using deep learning. Despite challenges from population differences and data heterogeneity, these models can be effectively used in various clinical settings. Model accuracy can improve over time with cumulative data, and publicly available datasets may further refine them. Our approach enables more accurate and efficient BAAs, supporting healthcare professionals where traditional methods are time-consuming and variable. The developed automated BAA model for the Turkish population offers a reliable and efficient alternative to traditional methods. By utilizing deep learning with diverse datasets from Bağcılar Hospital and publicly available sources, the model minimizes assessment time and reduces variability. This advancement enhances clinical decision-making, supports standardized BAA practices, and improves patient care in various healthcare settings.

The advent of large-scale vision foundation models, pre-trained on diverse natural images, has marked a paradigm shift in computer vision. However, how the frontier vision foundation models' efficacies transfer to specialized domains remains such as medical imaging remains an open question. This report investigates whether DINOv3, a state-of-the-art self-supervised vision transformer (ViT) that features strong capability in dense prediction tasks, can directly serve as a powerful, unified encoder for medical vision tasks without domain-specific pre-training. To answer this, we benchmark DINOv3 across common medical vision tasks, including 2D/3D classification and segmentation on a wide range of medical imaging modalities. We systematically analyze its scalability by varying model sizes and input image resolutions. Our findings reveal that DINOv3 shows impressive performance and establishes a formidable new baseline. Remarkably, it can even outperform medical-specific foundation models like BiomedCLIP and CT-Net on several tasks, despite being trained solely on natural images. However, we identify clear limitations: The model's features degrade in scenarios requiring deep domain specialization, such as in Whole-Slide Pathological Images (WSIs), Electron Microscopy (EM), and Positron Emission Tomography (PET). Furthermore, we observe that DINOv3 does not consistently obey scaling law in the medical domain; performance does not reliably increase with larger models or finer feature resolutions, showing diverse scaling behaviors across tasks. Ultimately, our work establishes DINOv3 as a strong baseline, whose powerful visual features can serve as a robust prior for multiple complex medical tasks. This opens promising future directions, such as leveraging its features to enforce multiview consistency in 3D reconstruction.

Assessing the severity of rheumatoid arthritis (RA) using the Total Sharp/Van Der Heijde Score (TSS) is crucial, but manual scoring is often time-consuming and subjective. This study introduces an Automated Radiographic Sharp Scoring (ARTSS) framework that leverages deep learning to analyze full-hand X-ray images, aiming to reduce inter- and intra-observer variability. The research uniquely accommodates patients with joint disappearance and variable-length image sequences. We developed ARTSS using data from 970 patients, structured into four stages: I) Image pre-processing and re-orientation using ResNet50, II) Hand segmentation using UNet.3, III) Joint identification using YOLOv7, and IV) TSS prediction using models such as VGG16, VGG19, ResNet50, DenseNet201, EfficientNetB0, and Vision Transformer (ViT). We evaluated model performance with Intersection over Union (IoU), Mean Average Precision (MAP), mean absolute error (MAE), Root Mean Squared Error (RMSE), and Huber loss. The average TSS from two radiologists was used as the ground truth. Model training employed 3-fold cross-validation, with each fold consisting of 452 training and 227 validation samples, and external testing included 291 unseen subjects. Our joint identification model achieved 99% accuracy. The best-performing model, ViT, achieved a notably low Huber loss of 0.87 for TSS prediction. Our results demonstrate the potential of deep learning to automate RA scoring, which can significantly enhance clinical practice. Our approach addresses the challenge of joint disappearance and variable joint numbers, offers timesaving benefits, reduces inter- and intra-reader variability, improves radiologist accuracy, and aids rheumatologists in making more informed decisions.

Quantification of brain atrophy currently requires visual rating scales which are time consuming and automated brain image analysis is warranted. We validated our automated deep learning (DL) tool measuring the Global Cerebral Atrophy (GCA) score against trained human raters, and associations with age and cognitive impairment, in representative older (>65 years) patients. CT-brain scans were obtained from patients in acute medicine (ORCHARD-EPR), acute stroke (OCS studies) and a legacy sample. Scans were divided in a 60/20/20 ratio for training, optimisation and testing. CT-images were assessed by two trained raters (rater-1=864 scans, rater-2=20 scans). Agreement between DL tool-predicted GCA scores (range 0-39) and the visual ratings was evaluated using mean absolute error (MAE) and Cohen's weighted kappa. Among 864 scans (ORCHARD-EPR=578, OCS=200, legacy scans=86), MAE between the DL tool and rater-1 GCA scores was 3.2 overall, 3.1 for ORCHARD-EPR, 3.3 for OCS and 2.6 for the legacy scans and half had DL-predicted GCA error between -2 and 2. Inter-rater agreement was Kappa=0.45 between the DL-tool and rater-1, and 0.41 between the tool and rater- 2 whereas it was lower at 0.28 for rater-1 and rater-2. There was no difference in GCA scores from the DL-tool and the two raters (one-way ANOVA, p=0.35) or in mean GCA scores between the DL-tool and rater-1 (paired t-test, t=-0.43, p=0.66), the tool and rater-2 (t=1.35, p=0.18) or between rater-1 and rater-2 (t=0.99, p=0.32). DL-tool GCA scores correlated with age and cognitive scores (both p<0.001). Our DL CT-brain analysis tool measured GCA score accurately and without user input in real-world scans acquired from older patients. Our tool will enable extraction of standardised quantitative measures of atrophy at scale for use in health data research and will act as proof-of-concept towards a point-of-care clinically approved tool.

AI-assisted radiological interpretation is based on predominantly narrow, single-task models. This approach is impractical for covering the vast spectrum of imaging modalities, diseases, and radiological findings. Foundation models (FMs) hold the promise of broad generalization across modalities and in low-data settings. However, this potential has remained largely unrealized in radiology. We introduce Curia, a foundation model trained on the entire cross-sectional imaging output of a major hospital over several years, which to our knowledge is the largest such corpus of real-world data-encompassing 150,000 exams (130 TB). On a newly curated 19-task external validation benchmark, Curia accurately identifies organs, detects conditions like brain hemorrhages and myocardial infarctions, and predicts outcomes in tumor staging. Curia meets or surpasses the performance of radiologists and recent foundation models, and exhibits clinically significant emergent properties in cross-modality, and low-data regimes. To accelerate progress, we release our base model's weights at https://huggingface.co/raidium/curia.

Applying pre-trained medical deep learning segmentation models to out-of-domain images often yields predictions of insufficient quality. In this study, we propose using a robust generalizing descriptor, along with augmentation, to enable domain-generalized pre-training and test-time adaptation, thereby achieving high-quality segmentation in unseen domains. In this study, five different publicly available datasets, including 3D CT and MRI images, are used to evaluate segmentation performance in out-of-domain scenarios. The settings include abdominal, spine, and cardiac imaging. Domain-generalized pre-training on source data is used to obtain the best initial performance in the target domain. We introduce a combination of the generalizing SSC descriptor and GIN intensity augmentation for optimal generalization. Segmentation results are subsequently optimized at test time, where we propose adapting the pre-trained models for every unseen scan using a consistency scheme with the augmentation-descriptor combination. The proposed generalized pre-training and subsequent test-time adaptation improve model performance significantly in CT to MRI cross-domain prediction for abdominal (+46.2 and +28.2 Dice), spine (+72.9), and cardiac (+14.2 and +55.7 Dice) scenarios (<i>p</i> < 0.001). Our method enables the optimal, independent use of source and target data, successfully bridging domain gaps with a compact and efficient methodology.

Background: Precise breast ultrasound (BUS) segmentation supports reliable measurement, quantitative analysis, and downstream classification, yet remains difficult for small or low-contrast lesions with fuzzy margins and speckle noise. Text prompts can add clinical context, but directly applying weakly localized text-image cues (e.g., CAM/CLIP-derived signals) tends to produce coarse, blob-like responses that smear boundaries unless additional mechanisms recover fine edges. Methods: We propose XBusNet, a novel dual-prompt, dual-branch multimodal model that combines image features with clinically grounded text. A global pathway based on a CLIP Vision Transformer encodes whole-image semantics conditioned on lesion size and location, while a local U-Net pathway emphasizes precise boundaries and is modulated by prompts that describe shape, margin, and Breast Imaging Reporting and Data System (BI-RADS) terms. Prompts are assembled automatically from structured metadata, requiring no manual clicks. We evaluate on the Breast Lesions USG (BLU) dataset using five-fold cross-validation. Primary metrics are Dice and Intersection over Union (IoU); we also conduct size-stratified analyses and ablations to assess the roles of the global and local paths and the text-driven modulation. Results: XBusNet achieves state-of-the-art performance on BLU, with mean Dice of 0.8765 and IoU of 0.8149, outperforming six strong baselines. Small lesions show the largest gains, with fewer missed regions and fewer spurious activations. Ablation studies show complementary contributions of global context, local boundary modeling, and prompt-based modulation. Conclusions: A dual-prompt, dual-branch multimodal design that merges global semantics with local precision yields accurate BUS segmentation masks and improves robustness for small, low-contrast lesions.

Point-of-care ultrasonography has become a valuable tool for assessing diaphragmatic function in critically ill patients receiving invasive mechanical ventilation. However, conventional diaphragm ultrasound assessment remains highly operator-dependent and subjective. Previous research introduced automatic measurement of diaphragmatic excursion and velocity using 2D speckle-tracking technology. This study aimed to develop an artificial intelligence-multimodal learning framework to improve the prediction of weaning failure and guide individualized weaning strategies. This prospective study enrolled critically ill patients older than 18 years who received mechanical ventilation for more than 48 hours and were eligible for a spontaneous breathing trial in 2 intensive care units in Guangzhou, China. Before the spontaneous breathing trial, diaphragm ultrasound videos were collected using a standardized protocol, and automatic measurements of excursion and velocity were obtained. A total of 88 patients were included, with 50 successfully weaned and 38 experiencing weaning failure. Each patient record included 27 clinical and 6 diaphragmatic indicators, selected based on previous literature and phenotyping studies. Clinical variables were preprocessed using OneHotEncoder, normalization, and scaling. Ultrasound videos were interpolated to a uniform resolution of 224×224×96. Artificial intelligence-multimodal learning based on clinical characteristics, laboratory parameters, and diaphragm ultrasonic videos was established. Four experiments were conducted in an ablation setting to evaluate model performance using different combinations of input data: (1) diaphragmatic excursion only, (2) clinical and diaphragmatic indicators, (3) ultrasound videos only, and (4) all modalities combined (multimodal). Metrics for evaluation included classification accuracy, area under the receiver operating characteristic curve (AUC), average precision in the precision-recall curve, and calibration curve. Variable importance was assessed using SHAP (Shapley Additive Explanation) to interpret feature contributions and understand model predictions. The multimodal co-learning model outperformed all single-modal approaches. The accuracy improved when predicted through diaphragm ultrasound video data using Video Vision Transformer (accuracy=0.8095, AUC=0.852), clinical or ultrasound indicators (accuracy=0.7381, AUC=0.746), and the multimodal co-learning (accuracy=0.8331, AUC=0.894). The proposed co-learning model achieved the highest score (average precision=0.91) among the 4 experiments. Furthermore, calibration curve analysis demonstrated that the proposed colearning model was well calibrated, as the curve was closest to the perfectly calibrated line. Combining ultrasound and clinical data for colearning improved the accuracy of the weaning outcome prediction. Multimodal learning based on automatic measurement of point-of-care ultrasonography and automated collection of objective clinical indicators greatly enhanced the practical operability and user-friendliness of the system. The proposed model offered promising potential for widespread clinical application in intensive care settings.

Brain tumors are serious health problems that require early diagnosis due to their high mortality rates. Diagnosing tumors by examining Magnetic Resonance Imaging (MRI) images is a process that requires expertise and is prone to error. Therefore, the need for automated diagnosis systems is increasing day by day. In this context, a robust and explainable Deep Learning (DL) model for the classification of brain tumors is proposed. In this study, a publicly available Figshare dataset containing 3,064 T1-weighted contrast-enhanced brain MRI images of three tumor types was used. First, the classification performance of nine well-known CNN architectures was evaluated to determine the most effective backbone. Among these, EfficientNetV2 demonstrated the best performance and was selected as the backbone for further development. Subsequently, an attention-based MLP-Mixer architecture was integrated into EfficientNetV2 to enhance its classification capability. The performance of the final model was comprehensively compared with basic CNNs and the methods in the literature. Additionally, Grad-CAM visualization was used to interpret and validate the decision-making process of the proposed model. The proposed model's performance was evaluated using the five-fold cross-validation method. The proposed model demonstrated superior performance with 99.50% accuracy, 99.47% precision, 99.52% recall and 99.49% F1 score. The results obtained show that the model outperforms the studies in the literature. Moreover, Grad-CAM visualizations demonstrate that the model effectively focuses on relevant regions of MRI images, thus improving interpretability and clinical reliability. A robust deep learning model for clinical decision support systems has been obtained by combining EfficientNetV2 and attention-based MLP-Mixer, providing high accuracy and interpretability in brain tumor classification.

Vision foundation models like DINOv2 demonstrate remarkable potential in medical imaging despite their origin in natural image domains. However, their design inherently works best for uni-modal image analysis, limiting their effectiveness for multi-modal imaging tasks that are common in many medical fields, such as neurology and oncology. While supervised models perform well in this setting, they fail to leverage unlabeled datasets and struggle with missing modalities, a frequent challenge in clinical settings. To bridge these gaps, we introduce MM-DINOv2, a novel and efficient framework that adapts the pre-trained vision foundation model DINOv2 for multi-modal medical imaging. Our approach incorporates multi-modal patch embeddings, enabling vision foundation models to effectively process multi-modal imaging data. To address missing modalities, we employ full-modality masking, which encourages the model to learn robust cross-modality relationships. Furthermore, we leverage semi-supervised learning to harness large unlabeled datasets, enhancing both the accuracy and reliability of medical predictions. Applied to glioma subtype classification from multi-sequence brain MRI, our method achieves a Matthews Correlation Coefficient (MCC) of 0.6 on an external test set, surpassing state-of-the-art supervised approaches by +11.1%. Our work establishes a scalable and robust solution for multi-modal medical imaging tasks, leveraging powerful vision foundation models pre-trained on natural images while addressing real-world clinical challenges such as missing data and limited annotations.