Deep learning has transformed computer vision but relies heavily on large labeled datasets and computational resources. Transfer learning, particularly fine-tuning pretrained models, offers a practical alternative; however, models pretrained on natural image datasets such as ImageNet may fail to capture domain-specific characteristics in medical imaging. This study introduces an unsupervised learning framework that extracts high-value dermatological features instead of relying solely on ImageNet-based pretraining. We employ a Variational Autoencoder (VAE) trained from scratch on a proprietary dermatological dataset, allowing the model to learn a structured and clinically relevant latent space. This self-supervised feature extractor is then compared to an ImageNet-pretrained backbone under identical classification conditions, highlighting the trade-offs between general-purpose and domain-specific pretraining. Our results reveal distinct learning patterns. The self-supervised model achieves a final validation loss of 0.110 (-33.33%), while the ImageNet-pretrained model stagnates at 0.100 (-16.67%), indicating overfitting. Accuracy trends confirm this: the self-supervised model improves from 45% to 65% (+44.44%) with a near-zero overfitting gap, whereas the ImageNet-pretrained model reaches 87% (+50.00%) but plateaus at 75% (+19.05%), with its overfitting gap increasing to +0.060. These findings suggest that while ImageNet pretraining accelerates convergence, it also amplifies overfitting on non-clinically relevant features. In contrast, self-supervised learning achieves steady improvements, stronger generalization, and superior adaptability, underscoring the importance of domain-specific feature extraction in medical imaging.

Medical image segmentation is critical for disease diagnosis, treatment planning, and prognosis assessment, yet the complexity and diversity of medical images pose significant challenges to accurate segmentation. While Convolutional Neural Networks capture local features and Vision Transformers excel in the global context, both struggle with efficient long-range dependency modeling. Inspired by Mamba's State Space Modeling efficiency, we propose ÆMMamba, a novel multi-scale feature extraction framework built on the Mamba backbone network. AÆMMamba integrates several innovative modules: the Efficient Fusion Bridge (EFB) module, which employs a bidirectional state-space model and attention mechanisms to fuse multi-scale features; the Edge-Aware Module (EAM), which enhances low-level edge representation using Sobel-based edge extraction; and the Boundary Sensitive Decoder (BSD), which leverages inverse attention and residual convolutional layers to handle cross-level complex boundaries. ÆMMamba achieves state-of-the-art performance across 8 medical segmentation datasets. On polyp segmentation datasets (Kvasir, ClinicDB, ColonDB, EndoScene, ETIS), it records the highest mDice and mIoU scores, outperforming methods like MADGNet and Swin-UMamba, with a standout mDice of 72.22 on ETIS, the most challenging dataset in this domain. For lung and breast segmentation, ÆMMamba surpasses competitors such as H2Former and SwinUnet, achieving Dice scores of 84.24 on BUSI and 79.83 on COVID-19 Lung. And on the LGG brain MRI dataset, ÆMMamba attains an mDice of 87.25 and an mIoU of 79.31, outperforming all compared methods. The source code will be released at https://github.com/xingbod/eMMamba.

With the growing capabilities of language models like ChatGPT to process text and images, this study evaluated their accuracy in detecting supernumerary teeth on periapical radiographs. A customized GPT-4V model (CGPT-4V) was also developed to assess whether domain-specific training could improve diagnostic performance compared to standard GPT-4V and GPT-4o models. One hundred eighty periapical radiographs (90 with and 90 without supernumerary teeth) were evaluated using GPT-4 V, GPT-4o, and a fine-tuned CGPT-4V model. Each image was assessed separately with the standardized prompt "Are there any supernumerary teeth in the radiograph above?" to avoid contextual bias. Three dental experts scored the responses using a three-point Likert scale for positive cases and a binary scale for negatives. Chi-square tests and ROC analysis were used to compare model performances (p < 0.05). Among the three models, CGPT-4 V exhibited the highest accuracy, detecting supernumerary teeth correctly in 91% of cases, compared to 77% for GPT-4o and 63% for GPT-4V. The CGPT-4V model also demonstrated a significantly lower false positive rate (16%) than GPT-4V (42%). A statistically significant difference was found between CGPT-4V and GPT-4o (p < 0.001), while no significant difference was observed between GPT-4V and CGPT-4V or between GPT-4V and GPT-4o. Additionally, CGPT-4V successfully identified multiple supernumerary teeth in radiographs where present. These findings highlight the diagnostic potential of customized GPT models in dental radiology. Future research should focus on multicenter validation, seamless clinical integration, and cost-effectiveness to support real-world implementation.

Computed Tomography serves as an indispensable tool in clinical workflows, providing non-invasive visualization of internal anatomical structures. Existing CT reconstruction works are limited to small-capacity model architecture and inflexible volume representation. In this work, we present X-GRM (X-ray Gaussian Reconstruction Model), a large feedforward model for reconstructing 3D CT volumes from sparse-view 2D X-ray projections. X-GRM employs a scalable transformer-based architecture to encode sparse-view X-ray inputs, where tokens from different views are integrated efficiently. Then, these tokens are decoded into a novel volume representation, named Voxel-based Gaussian Splatting (VoxGS), which enables efficient CT volume extraction and differentiable X-ray rendering. This combination of a high-capacity model and flexible volume representation, empowers our model to produce high-quality reconstructions from various testing inputs, including in-domain and out-domain X-ray projections. Our codes are available at: https://github.com/CUHK-AIM-Group/X-GRM.

Foundation models (FMs) such as CLIP and SAM have recently shown great promise in image segmentation tasks, yet their adaptation to 3D medical imaging-particularly for pathology detection and segmentation-remains underexplored. A critical challenge arises from the domain gap between natural images and medical volumes: existing FMs, pre-trained on 2D data, struggle to capture 3D anatomical context, limiting their utility in clinical applications like tumor segmentation. To address this, we propose an adaptation framework called TAGS: Tumor Adaptive Guidance for SAM, which unlocks 2D FMs for 3D medical tasks through multi-prompt fusion. By preserving most of the pre-trained weights, our approach enhances SAM's spatial feature extraction using CLIP's semantic insights and anatomy-specific prompts. Extensive experiments on three open-source tumor segmentation datasets prove that our model surpasses the state-of-the-art medical image segmentation models (+46.88% over nnUNet), interactive segmentation frameworks, and other established medical FMs, including SAM-Med2D, SAM-Med3D, SegVol, Universal, 3D-Adapter, and SAM-B (at least +13% over them). This highlights the robustness and adaptability of our proposed framework across diverse medical segmentation tasks.

To develop an integrative radiopathomic model based on deep learning to predict overall survival (OS) in locally advanced nasopharyngeal carcinoma (LANPC) patients. A cohort of 343 LANPC patients with pretreatment MRI and whole slide image (WSI) were randomly divided into training (n = 202), validation (n = 91), and external test (n = 50) sets. For WSIs, a self-attention mechanism was employed to assess the significance of different patches for the prognostic task, aggregating them into a WSI-level representation. For MRI, a multilayer perceptron was used to encode the extracted radiomic features, resulting in an MRI-level representation. These were combined in a multimodal fusion model to produce prognostic predictions. Model performances were evaluated using the concordance index (C-index), and Kaplan-Meier curves were employed for risk stratification. To enhance model interpretability, attention-based and Integrated Gradients techniques were applied to explain how WSIs and MRI features contribute to prognosis predictions. The radiopathomics model achieved high predictive accuracy in predicting the OS, with a C-index of 0.755 (95 % CI: 0.673-0.838) and 0.744 (95 % CI: 0.623-0.808) in the training and validation sets, respectively, outperforming single-modality models (radiomic signature: 0.636, 95 % CI: 0.584-0.688; deep pathomic signature: 0.736, 95 % CI: 0.684-0.810). In the external test, similar findings were observed for the predictive performance of the radiopathomics, radiomic signature, and deep pathomic signature, with their C-indices being 0.735, 0.626, and 0.660 respectively. The radiopathomics model effectively stratified patients into high- and low-risk groups (P < 0.001). Additionally, attention heatmaps revealed that high-attention regions corresponded with tumor areas in both risk groups. n: The radiopathomics model holds promise for predicting clinical outcomes in LANPC patients, offering a potential tool for improving clinical decision-making.

Exploring the trustworthiness of deep learning models is crucial, especially in critical domains such as medical imaging decision support systems. Conformal prediction has emerged as a rigorous means of providing deep learning models with reliable uncertainty estimates and safety guarantees. However, conformal prediction results face challenges due to the backbone model's struggles in domain-shifted scenarios, such as variations in different sources. To aim this challenge, this paper proposes a novel framework termed Conformal Ensemble of Vision Transformers (CE-ViTs) designed to enhance image classification performance by prioritizing domain adaptation and model robustness, while accounting for uncertainty. The proposed method leverages an ensemble of vision transformer models in the backbone, trained on diverse datasets including HAM10000, Dermofit, and Skin Cancer ISIC datasets. This ensemble learning approach, calibrated through the combined mentioned datasets, aims to enhance domain adaptation through conformal learning. Experimental results underscore that the framework achieves a high coverage rate of 90.38\%, representing an improvement of 9.95\% compared to the HAM10000 model. This indicates a strong likelihood that the prediction set includes the true label compared to singular models. Ensemble learning in CE-ViTs significantly improves conformal prediction performance, increasing the average prediction set size for challenging misclassified samples from 1.86 to 3.075.

To perform a systematic review on the impact of deep learning (DL)-based triage for reducing diagnostic delays and improving patient outcomes in peer-reviewed and pre-print publications. A search was conducted of primary research studies focused on DL-based worklist optimization for diagnostic imaging triage published on multiple databases from January 2018 until July 2024. Extracted data included study design, dataset characteristics, workflow metrics including report turnaround time and time-to-treatment, and patient outcome differences. Further analysis between clinical settings and integration modality was investigated using nonparametric statistics. Risk of bias was assessed with the risk of bias in non-randomized studies-of interventions (ROBINS-I) checklist. A total of 38 studies from 20 publications, involving 138,423 images, were analyzed. Workflow interventions concerned pulmonary embolism (n = 8), stroke (n = 3), intracranial hemorrhage (n = 12), and chest conditions (n = 15). Patients in the post DL-triage group had shorter median report turnaround times: a mean difference of 12.3 min (IQR: -25.7, -7.6) for pulmonary embolism, 20.5 min (IQR: -32.1, -9.3) for stroke, 4.3 min (IQR: -8.6, 1.3) for intracranial hemorrhage and 29.7 min (IQR: -2947.7, -18.3) for chest diseases. Sub-group analysis revealed that reductions varied per clinical environment and relative prevalence rates but were the highest when algorithms actively stratified and reordered the radiological worklist, with reductions of -43.7% in report turnaround time compared to -7.6% from widget-based systems (p < 0.01). DL-based triage systems had comparable report turnaround time improvements, especially in outpatient and high-prevalence settings, suggesting that AI-based triage holds promise in alleviating radiology workloads. Question Can DL-based triage address lengthening imaging report turnaround times and improve patient outcomes across distinct clinical environments? Findings DL-based triage improved report turnaround time across disease groups, with higher reductions reported in high-prevalence or lower acuity settings. Clinical relevance DL-based workflow prioritization is a reliable tool for reducing diagnostic imaging delay for time-sensitive disease across clinical settings. However, further research and reliable metrics are needed to provide specific recommendations with regards to false-negative examinations and multi-condition prioritization.

Medical image segmentation plays an important role in various clinical applications, but existing models often struggle with the computational inefficiencies and challenges posed by complex medical data. State Space Sequence Models (SSMs) have demonstrated promise in modeling long-range dependencies with linear computational complexity, yet their application in medical image segmentation remains hindered by incompatibilities with image tokens and autoregressive assumptions. Moreover, it is difficult to achieve a balance in capturing both local fine-grained information and global semantic dependencies. To address these challenges, we introduce SAMA-UNet, a novel architecture for medical image segmentation. A key innovation is the Self-Adaptive Mamba-like Aggregated Attention (SAMA) block, which integrates contextual self-attention with dynamic weight modulation to prioritise the most relevant features based on local and global contexts. This approach reduces computational complexity and improves the representation of complex image features across multiple scales. We also suggest the Causal-Resonance Multi-Scale Module (CR-MSM), which enhances the flow of information between the encoder and decoder by using causal resonance learning. This mechanism allows the model to automatically adjust feature resolution and causal dependencies across scales, leading to better semantic alignment between the low-level and high-level features in U-shaped architectures. Experiments on MRI, CT, and endoscopy images show that SAMA-UNet performs better in segmentation accuracy than current methods using CNN, Transformer, and Mamba. The implementation is publicly available at GitHub.

Computed Tomography serves as an indispensable tool in clinical workflows, providing non-invasive visualization of internal anatomical structures. Existing CT reconstruction works are limited to small-capacity model architecture, inflexible volume representation, and small-scale training data. In this paper, we present X-GRM (X-ray Gaussian Reconstruction Model), a large feedforward model for reconstructing 3D CT from sparse-view 2D X-ray projections. X-GRM employs a scalable transformer-based architecture to encode an arbitrary number of sparse X-ray inputs, where tokens from different views are integrated efficiently. Then, tokens are decoded into a new volume representation, named Voxel-based Gaussian Splatting (VoxGS), which enables efficient CT volume extraction and differentiable X-ray rendering. To support the training of X-GRM, we collect ReconX-15K, a large-scale CT reconstruction dataset containing around 15,000 CT/X-ray pairs across diverse organs, including the chest, abdomen, pelvis, and tooth etc. This combination of a high-capacity model, flexible volume representation, and large-scale training data empowers our model to produce high-quality reconstructions from various testing inputs, including in-domain and out-domain X-ray projections. Project Page: https://github.com/CUHK-AIM-Group/X-GRM.