Recently, combining the strategy of consistency regularization with uncertainty estimation has shown promising performance on semi-supervised medical image segmentation tasks. However, most existing methods estimate the uncertainty solely based on the outputs of a single neural network, which results in imprecise uncertainty estimations and eventually degrades the segmentation performance. In this paper, we propose a novel Uncertainty Co-estimator (UnCo) framework to deal with this problem. Inspired by the co-training technique, UnCo establishes two different mean-teacher modules (i.e., two pairs of teacher and student models), and estimates three types of uncertainty from the multi-source predictions generated by these models. Through combining these uncertainties, their differences will help to filter out incorrect noise in each estimate, thus allowing the final fused uncertainty maps to be more accurate. These resulting maps are then used to enhance a cross-consistency regularization imposed between the two modules. In addition, UnCo also designs an internal consistency regularization within each module, so that the student models can aggregate diverse feature information from both modules, thus promoting the semi-supervised segmentation performance. Finally, an adversarial constraint is introduced to maintain the model diversity. Experimental results on four medical image datasets indicate that UnCo can achieve new state-of-the-art performance on both 2D and 3D semi-supervised segmentation tasks. The source code will be available at https://github.com/z1010x/UnCo.

We introduce CheXGenBench, a rigorous and multifaceted evaluation framework for synthetic chest radiograph generation that simultaneously assesses fidelity, privacy risks, and clinical utility across state-of-the-art text-to-image generative models. Despite rapid advancements in generative AI for real-world imagery, medical domain evaluations have been hindered by methodological inconsistencies, outdated architectural comparisons, and disconnected assessment criteria that rarely address the practical clinical value of synthetic samples. CheXGenBench overcomes these limitations through standardised data partitioning and a unified evaluation protocol comprising over 20 quantitative metrics that systematically analyse generation quality, potential privacy vulnerabilities, and downstream clinical applicability across 11 leading text-to-image architectures. Our results reveal critical inefficiencies in the existing evaluation protocols, particularly in assessing generative fidelity, leading to inconsistent and uninformative comparisons. Our framework establishes a standardised benchmark for the medical AI community, enabling objective and reproducible comparisons while facilitating seamless integration of both existing and future generative models. Additionally, we release a high-quality, synthetic dataset, SynthCheX-75K, comprising 75K radiographs generated by the top-performing model (Sana 0.6B) in our benchmark to support further research in this critical domain. Through CheXGenBench, we establish a new state-of-the-art and release our framework, models, and SynthCheX-75K dataset at https://raman1121.github.io/CheXGenBench/

Minimally invasive surgery involves entering the body through small incisions or natural orifices, using a medical endoscope for observation and clinical procedures. However, traditional endoscopic images often suffer from low texture and uneven illumination, which can negatively impact surgical and diagnostic outcomes. To address these challenges, many researchers have applied deep learning methods to enhance the processing of endoscopic images. This paper proposes a monocular medical endoscopic image depth estimation method based on a window-adaptive asymmetric dual-branch Siamese network. In this network, one branch focuses on processing global image information, while the other branch concentrates on local details. An improved lightweight Squeeze-and-Excitation (SE) module is added to the final layer of each branch, dynamically adjusting the inter-channel weights through self-attention. The outputs from both branches are then integrated using a lightweight cross-attention feature fusion module, enabling cross-branch feature interaction and enhancing the overall feature representation capability of the network. Extensive ablation and comparative experiments were conducted on medical datasets (EAD2019, Hamlyn, M2caiSeg, UCL) and a non-medical dataset (NYUDepthV2), with both qualitative and quantitative results-measured in terms of RMSE, AbsRel, FLOPs and running time-demonstrating the superiority of the proposed model. Additionally, comparisons with CT images show good organ boundary matching capability, highlighting the potential of our method for clinical applications. The key code of this paper is available at: https://github.com/superchongcnn/AttenAdapt_DE .

Deep neural networks excel in medical imaging but remain prone to biases, leading to fairness gaps across demographic groups. We provide the first systematic exploration of Human-AI alignment and fairness in this domain. Our results show that incorporating human insights consistently reduces fairness gaps and enhances out-of-domain generalization, though excessive alignment can introduce performance trade-offs, emphasizing the need for calibrated strategies. These findings highlight Human-AI alignment as a promising approach for developing fair, robust, and generalizable medical AI systems, striking a balance between expert guidance and automated efficiency. Our code is available at https://github.com/Roypic/Aligner.

Positron Emission Tomography (PET) is an important clinical imaging tool but inevitably introduces radiation exposure to patients and healthcare providers. Reducing the tracer injection dose and eliminating the CT acquisition for attenuation correction can reduce the overall radiation dose, but often results in PET with high noise and bias. Thus, it is desirable to develop 3D methods to translate the non-attenuation-corrected low-dose PET (NAC-LDPET) into attenuation-corrected standard-dose PET (AC-SDPET). Recently, diffusion models have emerged as a new state-of-the-art deep learning method for image-to-image translation, better than traditional CNN-based methods. However, due to the high computation cost and memory burden, it is largely limited to 2D applications. To address these challenges, we developed a novel 2.5D Multi-view Averaging Diffusion Model (MADM) for 3D image-to-image translation with application on NAC-LDPET to AC-SDPET translation. Specifically, MADM employs separate diffusion models for axial, coronal, and sagittal views, whose outputs are averaged in each sampling step to ensure the 3D generation quality from multiple views. To accelerate the 3D sampling process, we also proposed a strategy to use the CNN-based 3D generation as a prior for the diffusion model. Our experimental results on human patient studies suggested that MADM can generate high-quality 3D translation images, outperforming previous CNN-based and Diffusion-based baseline methods. The code is available at https://github.com/tianqic/MADM.

Evaluating large artery atherosclerosis is critical for predicting and preventing ischemic strokes. Ultrasonographic assessment of the carotid arteries is the preferred first-line examination due to its ease of use, noninvasive, and absence of radiation exposure. This study proposed an automated classification model for the common carotid artery (CCA), carotid bulb, internal carotid artery (ICA), and external carotid artery (ECA) to enhance the quantification of carotid artery examinations.&#xD;Approach: A total of 2,943 B-mode ultrasound images (CCA: 1,563; bulb: 611; ICA: 476; ECA: 293) from 288 patients were collected. Three distinct sets of embedding features were extracted from artificial intelligence networks including pre-trained DenseNet201, vision Transformer (ViT), and echo contrastive language-image pre-training (EchoCLIP) models using deep learning architectures for pattern recognition. These features were then combined in a support vector machine (SVM) classifier to interpret the anatomical structures in B-mode images.&#xD;Main results: After ten-fold cross-validation, the model achieved an accuracy of 82.3%, which was significantly better than using individual feature sets, with a p-value of <0.001.&#xD;Significance: The proposed model could make carotid artery examinations more accurate and consistent with the achieved classification accuracy. The source code is available at https://github.com/buddykeywordw/Artery-Segments-Recognition&#xD.

Accurately and efficiently estimating the cortical thickness from magnetic resonance images (MRIs) is crucial for neuroscientific studies and clinical applications with various large-scale datasets. Diffeomorphic registration-based cortical thickness estimation (DiReCT) is a prominent traditional method of calculating such measures directly from original MRIs by applying diffeomorphic registration on segmented tissues. However, it suffers from prolonged computational time and limited reproducibility, impediments to its application in large-scale studies or real-time environments. This paper proposes a framework for cortical thickness estimation using deep learning-based anatomy segmentation and diffeomorphic registration. The framework begins by applying a convolutional neural network (CNN) segmentation model to the original image, generating a segmentation map that accurately delineates the cortical boundaries. Subsequently, a pair of distance maps generated from the segmentation map is injected into an unsupervised learning-based registration network for fast and diffeomorphic registration. A novel algorithm based on diffeomorphisms of different time points is proposed to calculate the final thickness map. We systematically evaluated and compared our method with surface-based measures from FreeSurfer on two distinct datasets. The experimental results demonstrated a superior performance of the proposed method, surpassing the performance of DiReCT and DL+DiReCT in terms of time efficiency and consistency with FreeSurfer. Our code and pre-trained models are publicly available at: https://github.com/wujiong-hub/DL-CTE.git.

Inhalation injuries face a challenge in clinical diagnosis and grading due to the limitations of traditional methods, such as Abbreviated Injury Score (AIS), which rely on subjective assessments and show weak correlations with clinical outcomes. This study introduces a novel deep learning-based framework for grading inhalation injuries using bronchoscopy images with the duration of mechanical ventilation as an objective metric. To address the scarcity of medical imaging data, we propose enhanced StarGAN, a generative model that integrates Patch Loss and SSIM Loss to improve synthetic images' quality and clinical relevance. The augmented dataset generated by enhanced StarGAN significantly improved classification performance when evaluated using the Swin Transformer, achieving an accuracy of 77.78%, an 11.11% improvement over the original dataset. Image quality was assessed using the Fr\'echet Inception Distance (FID), where Enhanced StarGAN achieved the lowest FID of 30.06, outperforming baseline models. Burn surgeons confirmed the realism and clinical relevance of the generated images, particularly the preservation of bronchial structures and color distribution. These results highlight the potential of enhanced StarGAN in addressing data limitations and improving classification accuracy for inhalation injury grading.

Accurate segmentation of lung tumors is essential for advancing personalized medicine in non-small cell lung cancer (NSCLC). However, stage IV NSCLC presents significant challenges due to heterogeneous tumor morphology and the presence of associated conditions including infection, atelectasis and pleural effusion. The complexity of multicentric datasets further complicates robust segmentation across diverse clinical settings. In this study, we evaluate deep-learning-based approaches for automated segmentation of advanced-stage lung tumors using 3D architectures on 387 CT scans from the Deep-Lung-IV study. Through comprehensive experiments, we assess the impact of model design, HU windowing, and dataset size on delineation performance, providing practical guidelines for robust implementation. Additionally, we propose a confidence score using deep ensembles to quantify prediction uncertainty and automate the identification of complex cases that require further review. Our results demonstrate the potential of attention-based architectures and specific preprocessing strategies to improve segmentation quality in such a challenging clinical scenario, while emphasizing the importance of uncertainty estimation to build trustworthy AI systems in medical imaging. Code is available at: https://github.com/Sacha-Dedeken/SegStageIVNSCLC.

Each year, thousands of brain MRI scans are collected to study structural development in children and adolescents. However, the amygdala, a particularly small and complex structure, remains difficult to segment reliably, especially in developing populations where its volume is even smaller. To address this challenge, we developed AmygdalaGo-BOLT, a boundary-aware deep learning model tailored for human amygdala segmentation. It was trained and validated using 854 manually labeled scans from pediatric datasets, with independent samples used to ensure performance generalizability. The model integrates multiscale image features, spatial priors, and self-attention mechanisms within a compact encoder-decoder architecture to enhance boundary detection. Validation across multiple imaging centers and age groups shows that AmygdalaGo-BOLT closely matches expert manual labels, improves processing efficiency, and outperforms existing tools in accuracy. This enables robust and scalable analysis of amygdala morphology in developmental neuroimaging studies where manual tracing is impractical. To support open and reproducible science, we publicly release both the labeled datasets and the full source code.