The determination of the mid-sagittal plane (MSP) on three-dimensional (3D) head imaging is key to the assessment of facial asymmetry. The aim of this study was to evaluate the reliability of an automated landmark-based MSP to quantify mandibular asymmetry on head computed tomography (CT) scans. A dataset of 368 CT scans, including orthognathic surgery patients, was automatically annotated with 3D cephalometric landmarks via a previously published deep learning-based method. Five of these landmarks were used to automatically construct an MSP orthogonal to the Frankfurt horizontal plane. The reliability of automatic MSP construction was compared with the reliability of manual MSP construction based on 6 manual localizations by 3 experienced operators on 19 randomly selected CT scans. The mandibular asymmetry of the 368 CT scans with respect to the MSP was calculated and compared with clinical expert judgment. The construction of the MSP was found to be highly reliable, both manually and automatically. The manual reproducibility 95% limit of agreement was less than 1 mm for -y translation and less than 1.1° for -x and -z rotation, and the automatic measurement lied within the confidence interval of the manual method. The automatic MSP construction was shown to be clinically relevant, with the mandibular asymmetry measures being consistent with the expertly assessed levels of asymmetry. The proposed automatic landmark-based MSP construction was found to be as reliable as manual construction and clinically relevant in assessing the mandibular asymmetry of 368 head CT scans. Once implemented in a clinical software, fully automated landmark-based MSP construction could be clinically used to assess mandibular asymmetry on head CT scans.

Deep learning has been utilized to segment teeth from computed tomography (CT) or cone-beam CT (CBCT). However, the performance of deep learning is unknown due to multiple models and diverse evaluation metrics. This systematic review and meta-analysis aims to evaluate the evolution and performance of deep learning in tooth segmentation. We systematically searched PubMed, Web of Science, Scopus, IEEE Xplore, arXiv.org, and ACM for studies investigating deep learning in human tooth segmentation from CT/CBCT. Included studies were assessed using the Quality Assessment of Diagnostic Accuracy Study (QUADAS-2) tool. Data were extracted for meta-analyses by random-effects models. A total of 30 studies were included in the systematic review, and 28 of them were included for meta-analyses. Various deep learning algorithms were categorized according to the backbone network, encompassing single-stage convolutional models, convolutional models with U-Net architecture, Transformer models, convolutional models with attention mechanisms, and combinations of multiple models. Convolutional models with U-Net architecture were the most commonly used deep learning algorithms. The integration of attention mechanism within convolutional models has become a new topic. 29 evaluation metrics were identified, with Dice Similarity Coefficient (DSC) being the most popular. The pooled results were 0.93 [0.93, 0.93] for DSC, 0.86 [0.85, 0.87] for Intersection over Union (IoU), 0.22 [0.19, 0.24] for Average Symmetric Surface Distance (ASSD), 0.92 [0.90, 0.94] for sensitivity, 0.71 [0.26, 1.17] for 95% Hausdorff distance, and 0.96 [0.93, 0.98] for precision. No significant difference was observed in the segmentation of single-rooted or multi-rooted teeth. No obvious correlation between sample size and segmentation performance was observed. Multiple deep learning algorithms have been successfully applied to tooth segmentation from CT/CBCT and their evolution has been well summarized and categorized according to their backbone structures. In future, studies are needed with standardized protocols and open labelled datasets.

Using massive datasets, foundation models are large-scale, pre-trained models that perform a wide range of tasks. These models have shown consistently improved results with the introduction of new methods. It is crucial to analyze how these trends impact the medical field and determine whether these advancements can drive meaningful change. This study investigates the application of recent state-of-the-art foundation models, DINOv2, MAE, VMamba, CoCa, SAM2, and AIMv2, for medical image classification. We explore their effectiveness on datasets including CBIS-DDSM for mammography, ISIC2019 for skin lesions, APTOS2019 for diabetic retinopathy, and CHEXPERT for chest radiographs. By fine-tuning these models and evaluating their configurations, we aim to understand the potential of these advancements in medical image classification. The results indicate that these advanced models significantly enhance classification outcomes, demonstrating robust performance despite limited labeled data. Based on our results, AIMv2, DINOv2, and SAM2 models outperformed others, demonstrating that progress in natural domain training has positively impacted the medical domain and improved classification outcomes. Our code is publicly available at: https://github.com/sajjad-sh33/Medical-Transfer-Learning.

In contrast to vision transformers, which model long-range dependencies through global self-attention, large kernel convolutions provide a more efficient and scalable alternative, particularly in high-resolution 3D volumetric settings. However, naively increasing kernel size often leads to optimization instability and degradation in performance. Motivated by the spatial bias observed in effective receptive fields (ERFs), we hypothesize that different kernel elements converge at variable rates during training. To support this, we derive a theoretical connection between element-wise gradients and first-order optimization, showing that structurally re-parameterized convolution blocks inherently induce spatially varying learning rates. Building on this insight, we introduce Rep3D, a 3D convolutional framework that incorporates a learnable spatial prior into large kernel training. A lightweight two-stage modulation network generates a receptive-biased scaling mask, adaptively re-weighting kernel updates and enabling local-to-global convergence behavior. Rep3D adopts a plain encoder design with large depthwise convolutions, avoiding the architectural complexity of multi-branch compositions. We evaluate Rep3D on five challenging 3D segmentation benchmarks and demonstrate consistent improvements over state-of-the-art baselines, including transformer-based and fixed-prior re-parameterization methods. By unifying spatial inductive bias with optimization-aware learning, Rep3D offers an interpretable, and scalable solution for 3D medical image analysis. The source code is publicly available at https://github.com/leeh43/Rep3D.

Limited Angle Computed Tomography (LACT) often faces significant challenges due to missing angular information. Unlike previous methods that operate in the image domain, we propose a new method that focuses on sinogram inpainting. We leverage MR-SDEs, a variant of diffusion models that characterize the diffusion process with mean-reverting stochastic differential equations, to fill in missing angular data at the projection level. Furthermore, by combining distillation with constraining the output of the model using the pseudo-inverse of the inpainting matrix, the diffusion process is accelerated and done in a step, enabling efficient and accurate sinogram completion. A subsequent post-processing module back-projects the inpainted sinogram into the image domain and further refines the reconstruction, effectively suppressing artifacts while preserving critical structural details. Quantitative experimental results demonstrate that the proposed method achieves state-of-the-art performance in both perceptual and fidelity quality, offering a promising solution for LACT reconstruction in scientific and clinical applications.

Advanced autoregressive models have reshaped multimodal AI. However, their transformative potential in medical imaging remains largely untapped due to the absence of a unified visual tokenizer -- one capable of capturing fine-grained visual structures for faithful image reconstruction and realistic image synthesis, as well as rich semantics for accurate diagnosis and image interpretation. To this end, we present MedITok, the first unified tokenizer tailored for medical images, encoding both low-level structural details and high-level clinical semantics within a unified latent space. To balance these competing objectives, we introduce a novel two-stage training framework: a visual representation alignment stage that cold-starts the tokenizer reconstruction learning with a visual semantic constraint, followed by a textual semantic representation alignment stage that infuses detailed clinical semantics into the latent space. Trained on the meticulously collected large-scale dataset with over 30 million medical images and 2 million image-caption pairs, MedITok achieves state-of-the-art performance on more than 30 datasets across 9 imaging modalities and 4 different tasks. By providing a unified token space for autoregressive modeling, MedITok supports a wide range of tasks in clinical diagnostics and generative healthcare applications. Model and code will be made publicly available at: https://github.com/Masaaki-75/meditok.

Most publicly available medical segmentation datasets are only partially labeled, with annotations provided for a subset of anatomical structures. When multiple datasets are combined for training, this incomplete annotation poses challenges, as it limits the model's ability to learn shared anatomical representations among datasets. Furthermore, vision-only frameworks often fail to capture complex anatomical relationships and task-specific distinctions, leading to reduced segmentation accuracy and poor generalizability to unseen datasets. In this study, we proposed a novel CLIP-DINO Prompt-Driven Segmentation Network (CDPDNet), which combined a self-supervised vision transformer with CLIP-based text embedding and introduced task-specific text prompts to tackle these challenges. Specifically, the framework was constructed upon a convolutional neural network (CNN) and incorporated DINOv2 to extract both fine-grained and global visual features, which were then fused using a multi-head cross-attention module to overcome the limited long-range modeling capability of CNNs. In addition, CLIP-derived text embeddings were projected into the visual space to help model complex relationships among organs and tumors. To further address the partial label challenge and enhance inter-task discriminative capability, a Text-based Task Prompt Generation (TTPG) module that generated task-specific prompts was designed to guide the segmentation. Extensive experiments on multiple medical imaging datasets demonstrated that CDPDNet consistently outperformed existing state-of-the-art segmentation methods. Code and pretrained model are available at: https://github.com/wujiong-hub/CDPDNet.git.

To evaluate feasibility of large language models (LLMs) to convert radiologist-generated report summaries into personalized report templates, and assess its impact on scan reporting time and quality. In this retrospective study, 100 CT scans from oncology patients were randomly divided into two equal sets. Two radiologists generated conventional reports for one set and summary reports for the other, and vice versa. Three LLMs - GPT-4, Google Gemini, and Claude Opus - generated complete reports from the summaries using institution-specific generic templates. Two expert radiologists qualitatively evaluated the radiologist summaries and LLM-generated reports using the ACR RADPEER scoring system, using conventional radiologist reports as reference. Reporting time for conventional versus summary-based reports was compared, and LLM-generated reports were analyzed for errors. Quantitative similarity and linguistic metrics were computed to assess report alignment across models with the original radiologist-generated report summaries. Statistical analyses were performed using Python 3.0 to identify significant differences in reporting times, error rates and quantitative metrics. The average reporting time was significantly shorter for summary method (6.76 min) compared to conventional method (8.95 min) (p < 0.005). Among the 100 radiologist summaries, 10 received RADPEER scores worse than 1, with three deemed to have clinically significant discrepancies. Only one LLM-generated report received a worse RADPEER score than its corresponding summary. Error frequencies among LLM-generated reports showed no significant differences across models, with template-related errors being most common (χ² = 1.146, p = 0.564). Quantitative analysis indicated significant differences in similarity and linguistic metrics among the three LLMs (p < 0.05), reflecting unique generation patterns. Summary-based scan reporting along with use of LLMs to generate complete personalized report templates can shorten reporting time while maintaining the report quality. However, there remains a need for human oversight to address errors in the generated reports. Summary-based reporting of radiological studies along with the use of large language models to generate tailored reports using generic templates has the potential to make the workflow more efficient by shortening the reporting time while maintaining the quality of reporting.

Recent advances in reinforcement learning with verifiable, rule-based rewards have greatly enhanced the reasoning capabilities and out-of-distribution generalization of VLMs/LLMs, obviating the need for manually crafted reasoning chains. Despite these promising developments in the general domain, their translation to medical imaging remains limited. Current medical reinforcement fine-tuning (RFT) methods predominantly focus on close-ended VQA, thereby restricting the model's ability to engage in world knowledge retrieval and flexible task adaptation. More critically, these methods fall short of addressing the critical clinical demand for open-ended, reasoning-intensive decision-making. To bridge this gap, we introduce \textbf{MedCCO}, the first multimodal reinforcement learning framework tailored for medical VQA that unifies close-ended and open-ended data within a curriculum-driven RFT paradigm. Specifically, MedCCO is initially fine-tuned on a diverse set of close-ended medical VQA tasks to establish domain-grounded reasoning capabilities, and is then progressively adapted to open-ended tasks to foster deeper knowledge enhancement and clinical interpretability. We validate MedCCO across eight challenging medical VQA benchmarks, spanning both close-ended and open-ended settings. Experimental results show that MedCCO consistently enhances performance and generalization, achieving a 11.4\% accuracy gain across three in-domain tasks, and a 5.7\% improvement on five out-of-domain benchmarks. These findings highlight the promise of curriculum-guided RL in advancing robust, clinically-relevant reasoning in medical multimodal language models.

Most publicly available medical segmentation datasets are only partially labeled, with annotations provided for a subset of anatomical structures. When multiple datasets are combined for training, this incomplete annotation poses challenges, as it limits the model's ability to learn shared anatomical representations among datasets. Furthermore, vision-only frameworks often fail to capture complex anatomical relationships and task-specific distinctions, leading to reduced segmentation accuracy and poor generalizability to unseen datasets. In this study, we proposed a novel CLIP-DINO Prompt-Driven Segmentation Network (CDPDNet), which combined a self-supervised vision transformer with CLIP-based text embedding and introduced task-specific text prompts to tackle these challenges. Specifically, the framework was constructed upon a convolutional neural network (CNN) and incorporated DINOv2 to extract both fine-grained and global visual features, which were then fused using a multi-head cross-attention module to overcome the limited long-range modeling capability of CNNs. In addition, CLIP-derived text embeddings were projected into the visual space to help model complex relationships among organs and tumors. To further address the partial label challenge and enhance inter-task discriminative capability, a Text-based Task Prompt Generation (TTPG) module that generated task-specific prompts was designed to guide the segmentation. Extensive experiments on multiple medical imaging datasets demonstrated that CDPDNet consistently outperformed existing state-of-the-art segmentation methods. Code and pretrained model are available at: https://github.com/wujiong-hub/CDPDNet.git.