Medical images like MR scans often show domain shifts across imaging sites due to scanner and protocol differences, which degrade machine learning performance in tasks such as disease classification. Domain harmonization is thus a critical research focus. Recent approaches encode brain images $\boldsymbol{x}$ into a low-dimensional latent space $\boldsymbol{z}$, then disentangle it into $\boldsymbol{z_u}$ (domain-invariant) and $\boldsymbol{z_d}$ (domain-specific), achieving strong results. However, these methods often lack interpretability$-$an essential requirement in medical applications$-$leaving practical issues unresolved. We propose Pseudo-Linear-Style Encoder Adversarial Domain Adaptation (PL-SE-ADA), a general framework for domain harmonization and interpretable representation learning that preserves disease-relevant information in brain MR images. PL-SE-ADA includes two encoders $f_E$ and $f_{SE}$ to extract $\boldsymbol{z_u}$ and $\boldsymbol{z_d}$, a decoder to reconstruct the image $f_D$, and a domain predictor $g_D$. Beyond adversarial training between the encoder and domain predictor, the model learns to reconstruct the input image $\boldsymbol{x}$ by summing reconstructions from $\boldsymbol{z_u}$ and $\boldsymbol{z_d}$, ensuring both harmonization and informativeness. Compared to prior methods, PL-SE-ADA achieves equal or better performance in image reconstruction, disease classification, and domain recognition. It also enables visualization of both domain-independent brain features and domain-specific components, offering high interpretability across the entire framework.

Oral and maxillofacial radiology plays a vital role in dental healthcare, but radiographic image interpretation is limited by a shortage of trained professionals. While AI approaches have shown promise, existing dental AI systems are restricted by their single-modality focus, task-specific design, and reliance on costly labeled data, hindering their generalization across diverse clinical scenarios. To address these challenges, we introduce DentVFM, the first family of vision foundation models (VFMs) designed for dentistry. DentVFM generates task-agnostic visual representations for a wide range of dental applications and uses self-supervised learning on DentVista, a large curated dental imaging dataset with approximately 1.6 million multi-modal radiographic images from various medical centers. DentVFM includes 2D and 3D variants based on the Vision Transformer (ViT) architecture. To address gaps in dental intelligence assessment and benchmarks, we introduce DentBench, a comprehensive benchmark covering eight dental subspecialties, more diseases, imaging modalities, and a wide geographical distribution. DentVFM shows impressive generalist intelligence, demonstrating robust generalization to diverse dental tasks, such as disease diagnosis, treatment analysis, biomarker identification, and anatomical landmark detection and segmentation. Experimental results indicate DentVFM significantly outperforms supervised, self-supervised, and weakly supervised baselines, offering superior generalization, label efficiency, and scalability. Additionally, DentVFM enables cross-modality diagnostics, providing more reliable results than experienced dentists in situations where conventional imaging is unavailable. DentVFM sets a new paradigm for dental AI, offering a scalable, adaptable, and label-efficient model to improve intelligent dental healthcare and address critical gaps in global oral healthcare.

Unsupervised deep generative models are emerging as a promising alternative to supervised methods for detecting and segmenting anomalies in brain imaging. Unlike fully supervised approaches, which require large voxel-level annotated datasets and are limited to well-characterised pathologies, these models can be trained exclusively on healthy data and identify anomalies as deviations from learned normative brain structures. This PRISMA-guided scoping review synthesises recent work on unsupervised deep generative models for anomaly detection in neuroimaging, including autoencoders, variational autoencoders, generative adversarial networks, and denoising diffusion models. A total of 49 studies published between 2018 - 2025 were identified, covering applications to brain MRI and, less frequently, CT across diverse pathologies such as tumours, stroke, multiple sclerosis, and small vessel disease. Reported performance metrics are compared alongside architectural design choices. Across the included studies, generative models achieved encouraging performance for large focal lesions and demonstrated progress in addressing more subtle abnormalities. A key strength of generative models is their ability to produce interpretable pseudo-healthy (also referred to as counterfactual) reconstructions, which is particularly valuable when annotated data are scarce, as in rare or heterogeneous diseases. Looking ahead, these models offer a compelling direction for anomaly detection, enabling semi-supervised learning, supporting the discovery of novel imaging biomarkers, and facilitating within- and cross-disease deviation mapping in unified end-to-end frameworks. To realise clinical impact, future work should prioritise anatomy-aware modelling, development of foundation models, task-appropriate evaluation metrics, and rigorous clinical validation.

Mammography, the current standard for breast cancer screening, has reduced sensitivity in women with dense breast tissue, contributing to missed or delayed diagnoses. Thermalytix, an AI-based thermal imaging modality, captures functional vascular and metabolic cues that may complement mammographic structural data. This study investigates whether a breast density-informed multi-modal AI framework can improve cancer detection by dynamically selecting the appropriate imaging modality based on breast tissue composition. A total of 324 women underwent both mammography and thermal imaging. Mammography images were analyzed using a multi-view deep learning model, while Thermalytix assessed thermal images through vascular and thermal radiomics. The proposed framework utilized Mammography AI for fatty breasts and Thermalytix AI for dense breasts, optimizing predictions based on tissue type. This multi-modal AI framework achieved a sensitivity of 94.55% (95% CI: 88.54-100) and specificity of 79.93% (95% CI: 75.14-84.71), outperforming standalone mammography AI (sensitivity 81.82%, specificity 86.25%) and Thermalytix AI (sensitivity 92.73%, specificity 75.46%). Importantly, the sensitivity of Mammography dropped significantly in dense breasts (67.86%) versus fatty breasts (96.30%), whereas Thermalytix AI maintained high and consistent sensitivity in both (92.59% and 92.86%, respectively). This demonstrates that a density-informed multi-modal AI framework can overcome key limitations of unimodal screening and deliver high performance across diverse breast compositions. The proposed framework is interpretable, low-cost, and easily deployable, offering a practical path to improving breast cancer screening outcomes in both high-resource and resource-limited settings.

The seminal vesicle region plays a crucial role in male reproductive health, and its accurate evaluation is essential for diagnosing infertility and carcinoma. Magnetic resonance imaging (MRI) is the primary modality for assessment; however, manual evaluation is time-consuming and subject to interobserver variability, necessitating automated approaches. This study presents a modified ResNet-based deep learning model specifically developed for automated localization of the seminal vesicle region in prostate MRI scans. Unlike segmentation-based methods, the focus is on robust region-level localization as a precursor to detailed analysis. To the best of our knowledge, this is the first study to address seminal vesicle localization directly from MRI using a deep learning model. Performance was evaluated using classification accuracy, inference time, and true positive (TP) coverage, with a sliding window approach to detect high-confidence regions. The modified ResNet-34 achieved the highest TP coverage (0.885) and classification accuracy (0.979), demonstrating improved localization with minimal computational overhead. Heatmap visualizations confirmed the model's focus on relevant anatomical structures. The proposed approach provides a practical solution for reducing manual effort and interobserver variability, offering a reliable foundation for subsequent segmentation and abnormality detection. Future work may explore the integration of multi-view imaging or 3D CNN architectures to further improve performance.

Accurate prediction of prognosis and risk stratification in patients with laryngeal cancer can inform appropriate treatment decision-making. This study aims to develop a multi-channel deep learning radiomics model based on contrast-enhanced computed tomography (CECT) for predicting postoperative overall survival (OS) in patients. A total of 272 patients with laryngeal cancer were retrospectively recruited from two hospitals between January 2016 and July 2021. Specifically, 156 patients were enrolled from Center 1 as the training cohort, and 116 patients from Center 2 as the external test cohort. Two imaging signatures, reflecting phenotypes of the radiomics and multi-channel deep learning features, were constructed using pretreatment venous-phase CECT images. Feature selection involved reproducibility evaluation, Spearman correlation coefficient, and least absolute shrinkage and selection operator (LASSO) regression. Ten machine learning algorithms were employed to construct the signatures. An integrated Deep Learning Radiomics Nomogram (DLRN) was subsequently developed to predict OS. Predictive performance was evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). Kaplan-Meier survival curves derived from the DLRN were used for patient risk stratification, and subgroup analyses were conducted to validate the model's robustness. The DLRN demonstrated satisfactory prognostic performance for laryngeal cancer. In the external test cohort, the 1-, 2-, and 3-year AUC values were 0.74, 0.75, and 0.80, respectively, and achieved a C-index of 0.73, outperforming individual models. Calibration curves and DCA indicated excellent calibration and the highest clinical net benefit. Subgroup analyses confirmed consistent DLRN performance across clinical stages, age groups, and surgical modalities. The proposed multi-channel deep learning radiomics model showed promising performance for predicting OS in laryngeal cancer patients. This approach may support individualized risk stratification and assist clinical decision-making in patients with laryngeal carcinoma.

Deep learning (DL) models based on ultrasound (US) images can enhance the ability of radiologists to diagnose ovarian tumors. This retrospective study included 916 women with ovarian tumors in southeast China who underwent surgery with clear pathology and preoperative US examination. The data set was divided into a training (80%) and a validation (20%) set. The test set consisted of 81 women with ovarian tumors from southwest and northeast China. DL models based on three backbone architectures, ResNet-50 (residual CNN), VGG16 (plain CNN), and Vision Transformer (ViT), were trained to classify benign, borderline, and malignant ovarian tumors. The diagnostic efficiency of primary US doctors combined with the DL model was compared with the ADNEX model and a US expert. Additionally, we compared the diagnostic performance of primary US doctors before and after being assisted by the integrated framework combining visual DL models and large language models. (1) The accuracy of the ResNet50-based DL model for benign, malignant, and borderline ovarian tumors was 91.8%, 84.61%, and 82.60% for the test sets, respectively. (2) After visual and linguistic DL assistance, the accuracy of primary US doctors all improved in the test set (doctor A: 76.62% to 90.90%, doctor B: 76.62% to 90.90%, doctor C: 79.22% to 94.54%, doctor D: 76.62% to 95.95%, doctor E: 76.60% to 95.95%, respectively). (3) The diagnostic consistency of primary US doctors for validation and test sets also increased (doctor A: 0.671 to 0.912, doctor B: 0.762 to 0.916, doctor C: 0.412 to 0.629, doctor D: 0.588 to 0.701, doctor E: 0.528 to 0.710, respectively). A DL system combining an image-based model (vision model) and a language model was developed to assist radiologists in classifying ovarian tumors in US images and enhance diagnostic efficacy. The established model can assist primary US doctors in preoperative diagnosis and improve the early detection and timely treatment of ovarian tumors. An ultrasound-based deep learning (DL) model was developed for ovarian tumors using multi-center patients. An image-based DL model was combined with a large language model to establish a diagnostic framework for ovarian tumor classification. Our DL model can improve the diagnosis of primary US doctors to the level of experts and might assist in surgical decision-making.

Manual CT protocol selection persists as a time-intensive and error-prone bottleneck in radiology workflows, impeding the realization of fully automated scanning pipelines. To overcome this limitation, we developed a Large Language Model Retrieval-Augmented Generation (LLM-RAG) framework for personalized CT protocol recommendation. This system constructs a protocol knowledge base from historical examination records to deliver institutionally tailored, precision recommendations aligned with clinical preferences. Our system demonstrated compelling performance (min: 88.60% precision, 89.34% recall, 88.08% F1, 96.09% accuracy), with key findings revealing: (1) task-specific parity between Qwen and DeepSeek models at equivalent scales (max Δ = 1.41% at 32B); (2) positive scaling laws where larger models boost accuracy (e.g., DeepSeek 7B → 32B: + 1.55%); and (3) linear GPU memory-cost scaling (7B:25 GB → 32B:95 GB), defining clinical deployment constraints. Error analysis of 225 discordant cases identified three primary patterns: over-recommendation (52.44%), unsuitable recommendation (27.56%), and clinically equivalent choices (20%). Critically, the framework achieves clinically viable accuracy without model retraining requirements-a pivotal advantage enabling significant utility in streamlining scanning operations and accelerating imaging workflow automation.

Artificial intelligence (AI) tools for radiology are commonly unmonitored once deployed. The lack of real-time case-by-case assessments of AI prediction confidence requires users to independently distinguish between trustworthy and unreliable AI predictions, which increases cognitive burden, reduces productivity, and potentially leads to misdiagnoses. To address these challenges, we introduce Ensembled Monitoring Model (EMM), a framework inspired by clinical consensus practices using multiple expert reviews. Designed specifically for black-box commercial AI products, EMM operates independently without requiring access to internal AI components or intermediate outputs, while still providing robust confidence measurements. Using intracranial hemorrhage detection as our test case on a large, diverse dataset of 2919 studies, we demonstrate that EMM can successfully categorize confidence in the AI-generated prediction, suggest appropriate actions, and help physicians recognize low confidence scenarios, ultimately reducing cognitive burden. Importantly, we provide key technical considerations and best practices for successfully translating EMM into clinical settings.

The automatic diagnosis model of medical image based on deep learning can improve the diagnosis efficiency and reduce the diagnosis cost. At present, there is a lack of research on special artificial intelligence models for medical image analysis of fundus disease characteristics. Considering that fundus diseases have both local and global features, this paper proposes a novel deep learning model Local-Global Scale Fusion Network (LGSF-Net). The novelty lies in a dual-stream fusion design that processes global context (Transformer) and local details (CNN) in parallel with residual fusion. On the public fundus dataset, LGSF-Net delivers 96% accuracy with only 18.7K parameters and 0.93 GFLOPs, outperforming existing state-of-the-art universal methods like ResNet50 and ViT. LGSF-Net is more suitable for clinical diagnosis because of its accuracy and lightweight design. The ablation study shows that the concept of LGSF-Net multi-scale fusion understanding has been correctly realized. This work effectively promotes the development of smart medicine and provides a new solution for the design of new deep learning models.