General-purpose large language models (LLMs) are increasingly proposed for diagnostic and triage decision support, yet their reliability relative to humans remains unclear. We evaluated eight contemporary LLMs (ChatGPT-4, ChatGPT-o1, DeepSeek-V3, DeepSeek-R1, Gemini-2.0, Copilot, Grok-2, Llama-3.1) on 48 single-turn clinical vignettes spanning four triage levels (Emergent, 1-day, 1-week, Self-care). Models were tested without prompts and with structured prompts comprising exemplar cases. Primary outcomes were diagnostic and triage accuracy. Secondary measures included confusion matrices, over-triage, safety of advice, and the Capability Comparison Score (CCS). Structured prompting improved performance across models: mean diagnostic accuracy increased from 89.84% to 91.67%, and mean triage accuracy increased from 76.82% to 86.20%. The best diagnostic accuracy was 93.75% (ChatGPT-o1 and DeepSeek-R1; Grok-2 matched this when prompted). Prompting shifted models toward safety: safety of advice rose from 89.06% to 94.53%, accompanied by higher over-triage (from 53.15% to 65.62%). CCS values were numerically lower than accuracy but preserved rankings and conclusions (diagnosis CCS: from 49.54 to 50.46; triage CCS: from 47.66 to 52.34). Error analyses showed predominant over-triage, with rarer but clinically important under-triage. On concise, text-only vignettes, the diagnostic accuracy of advanced LLMs was high, in some cases nearing benchmarks set by physicians in prior studies, whereas triage remained a more significant challenge. Structured prompting provided a practical, training-free lever to enhance robustness. Future work should evaluate uncertainty-aware prompting and real-world, multi-turn/multi-modality cases to strengthen clinical reliability.

A novel recursive cascaded full‐resolution residual network (RCFRR‐Net) for abdominal four‐dimensional computed tomography (4D‐CT) image registration was proposed. The entire network was end‐to‐end and trained in the unsupervised approach, which meant that the deformation vector field, which presented the ground truth, was not needed during training. The network was designed by cascading three full‐resolution residual subnetworks with different architectures. The training loss consisted of the image similarity loss and the deformation vector field regularization loss, which were calculated based on the final warped image and the fixed image, allowing all cascades to be trained jointly and perform the progressive registration cooperatively. Extensive network testing was conducted using diverse datasets, including an internal 4D‐CT dataset, a public DIRLAB 4D‐CT dataset, and a 4D cone‐beam CT (4D‐CBCT) dataset. Compared with the iteration‐based demon method and two deep learning‐based methods (VoxelMorph and recursive cascaded network), the RCFRR‐Net achieved consistent and significant gains, which demonstrated that the proposed method had superior performance and generalization capability in medical image registration. The proposed RCFRR‐Net was a promising tool for various clinical applications.

Objective. In helical tomotherapy, image-guided radiotherapy employs megavoltage computed tomography (MVCT) for precise targeting. However, the high voltage of megavoltage radiation introduces substantial noise, significantly compromising MVCT image clarity. This study aims to enhance MVCT image quality using a deep learning-based denoising method. Approach. We propose an unpaired MVCT denoising network using a coupled generative adversarial network framework (DeCoGAN). Our approach assumes that a universal latent code within a shared latent space can reconstruct any given pair of images. By employing an encoder, we enforce this shared-latent space constraint, facilitating the conversion of low-quality (noisy) MVCT images into high-quality (denoised) counterparts. The network learns the joint distribution of images from both domains by leveraging samples from their respective marginal distributions, enhanced by adversarial training for effective denoising. Main Results. Compared to an analytical algorithm (BM3D) and three deep learning-based methods (RED-CNN, WGAN-VGG and CycleGAN), the proposed method excels in preserving image details and enhancing human visual perception by removing most noise and retaining structural features. Quantitative analysis demonstrates that our method achieves the highest peak signal-to-noise ratio and Structural Similarity Index Measurement values, indicating superior denoising performance. Significance. The proposed DeCoGAN method shows remarkable MVCT denoising performance, making it a promising tool in the field of radiation therapy.

This study aimed to analyze the associations and substitutability of multi-parametric MRI images in glioblastoma (Gb) using the radiomics method. Utilizing the University of Pennsylvania Health System Glioblastoma dataset from The Cancer Imaging Archive, we extracted quantitative features from T1-weighted, T2-weighted, T2 fluid attenuated inversion recovery (T2-FLAIR), and post-contrast T1-weighted (T1-Gd) images. Feature associations were assessed using Spearman rank correlation with Benjamini-Hochberg correction for multiple comparisons. The substitution analysis was subsequently performed by developing prognostic signatures based on individual MRI sequences and then evaluating the performance of substituted signatures, wherein features from one sequence were replaced by their counterparts from another. Discriminative power was evaluated by the area under the receiver-operating-characteristic curve (AUC). Significant feature associations were observed across different MRI sequences. The strongest correlation was identified between T2-weighted and T2-FLAIR images, where 93% of features were significantly and positively correlated (mean absolute correlation coefficient [CC]: 0.57 ± 0.21). A substantial association was also noted between T1-weighted and T1-Gd images, with 86% of features significantly correlated (mean absolute CC: 0.49 ± 0.23). The correlation between T2-weighted and T1-Gd images was less pronounced (75% of features; mean absolute CC: 0.44). In the substitution analysis for prognostication, a signature based on T1-weighted images achieved an AUC of 0.73 (95% CI, 0.60–0.84). Replacing its features with those from T2-weighted images resulted in a signature with a slightly lower AUC of 0.65 (95% CI, 0.51–0.77), a modest difference of 0.08 (95% CI, -0.05–0.21). Conversely, substituting features from a T2-weighted image-based signature with their T1-Gd counterparts resulted in a more substantial decrease in AUC (difference: 0.10, 95% CI, -0.05–0.25). Our radiomics analysis indicated potential substantial information redundancy among certain multi-parametric MRI sequences for Gb characterization, particularly between T2-weighted and T2-FLAIR images. Nevertheless, sequences providing unique pathophysiological contrast, such as T1-Gd, appeared to hold distinct prognostic value that was not substituted. While these findings suggested the feasibility of abbreviated multi-parametric MR protocols for specific radiomics applications, they simultaneously underscored that rigorous, task-specific validation was an indispensable prerequisite for any consideration of widespread clinical adoption.

Objective. Medical imaging offered a non-invasive window to visualize tumors, with radiomics transforming these images into quantitative data for tumor phenotyping. However, the intricate web linking imaging features, clinical endpoints, and tumor biology was mostly uncharted. This study aimed to unravel the connections between CT imaging features and clinical characteristics, including tumor histopathological grading, clinical stage, and endocrine symptoms, alongside immunohistochemical markers of tumor cell growth, such as the Ki-67 index and nuclear mitosis rate. Approach. We conducted a retrospective analysis of data from 137 patients with pancreatic neuroendocrine tumors who had undergone contrast-enhanced CT scans across two institutions. Our study focused on three clinical factors: pathological grade, clinical stage, and endocrine symptom status, in addition to two immunohistochemical markers: the Ki-67 index and the rate of nuclear mitosis. We computed both predefined (2D and 3D) and learning-based features (via sparse autoencoder, or SAE) from the scans. To unearth the relationships between imaging features, clinical factors, and immunohistochemical markers, we employed the Spearman rank correlation along with the Benjamini-Hochberg method. Furthermore, we developed and validated radiomics signatures to foresee these clinical factors. Main results. The 3D imaging features showed the strongest relationships with clinical factors and immunohistochemical markers. For the association with pathological grade, the mean absolute value of the correlation coefficient (CC) of 2D, SAE, and 3D features was 0.3318 ± 0.1196, 0.2149 ± 0.0361, and 0.4189 ± 0.0882, respectively. While for the association with Ki-67 index and rate of nuclear mitosis, the 3D features also showed higher correlations, with CC as 0.4053 ± 0.0786 and 0.4061 ± 0.0806. In addition, the 3D feature-based signatures showed optimal performance in clinical factor prediction. Significance. We found relationships between imaging features, clinical factors, and immunohistochemical markers. The 3D features showed higher relationships with clinical factors and immunohistochemical markers.