In recent years, the advances of large language models and autonomous agents have revolutionized the healthcare field, facilitating diagnosis and improving treatment results. However, most existing AI systems rely on pre-trained knowledge and predefined pipelines, which struggle to learn dynamically from the interactive chat session history that contains patient outcomes and past failures. To address this limitation, we propose VIBEMed, a multi-agent framework with a built-in self-evolution mechanism and architecture-level safety sandbox for robust clinical decision support. The system integrates three specialized agents, including a Clinical Diagnostic Agent (CDA) for hypothesis generation, a Therapeutic Execution Agent (TEA) for treatment planning, and a Clinical Evolution Manager Agent (CEMA) that distills longitudinal clinical feedback into reusable knowledge, transforming multimodal patient information into personalized medical decisions. Through self-evolution mechanism, the framework enables iterative updates across memory, model behavior, and decision strategies, allowing the system to improve over time. Experimental results show that VIBEMed demonstrates superior performance through its evolving mechanism in complex clinical cases, particularly in tasks that require integrated decision-making and longitudinal planning. The framework also supports reliable end-to-end decisions in challenging scenarios such as oncology treatment planning, highlighting its feasibility in real-world clinical contexts. Overall, VIBEMed provides a practical path beyond static AI systems toward adaptive, experience-driven clinical decision support, demonstrating the value of combining multi-agent collaboration with continuous evolution for advancing precision medicine.

Large language models (LLMs) have emerged as transformative tools in medicine, with strong capabilities in language understanding, reasoning, and structured information extraction. Radiation oncology is particularly well suited for LLM integration due to its data-intensive workflows, reliance on structured guidelines, and documentation burden. This review summarizes recent applications, including domain-specific fine-tuning for decision support, automated nomenclature standardization, registry curation using autonomous LLM agents, and protocol-aware radiotherapy plan evaluation using modular retrieval-augmented generation (RAG). Additional applications include patient safety analysis through incident classification and root cause analysis, electronic health record (EHR)-integrated communication, CT simulation order summarization, daily readiness briefings, and patient education systems. Emerging multimodal approaches enable context-aware contouring, while early studies show LLMs can assist treatment planning by interpreting dosimetric feedback. Together, these advances highlight a shift toward clinically grounded, auditable, and workflow-integrated AI systems that enhance efficiency, safety, and patient engagement.

Purpose:To investigate the value of radiomics features extracted from plain and enhanced spectral CT-derived metrics in differentiating osteoblastic bone metastasis (OBM) and bone island (BI) in newly diagnosed cancer patients. Methods:From January to November 2020, 51 newly diagnosed cancer patients with 204 bone lesions (OBM = 116, BI = 88) receiving spectral CT were retrospectively enrolled. 40-140 keV mono-energy images were generated from plain CT and contrast-enhanced CT, and material-decomposition images, including water (calcium) and calcium (water) substrate density images from plain CT and Iodine (calcium) substrate density images from contrast-enhanced CT. Radiomics features were extracted from the manually segmented lesions, including shape feature set, material-separation feature set, plain spectral CT feature set, and enhanced spectral CT feature set. U-test and LASSO analysis were sequentially used to select the most relevant features. The shape model, material-separation model, plain CT model, contrast-enhanced CT model, and combined model were built using Random Forest with model performance evaluated using ROC analysis and compared using the Delong test. Results:After feature selection, four features were selected for the shape set, seven features for the material-separation set, seven features for the plain spectral CT set, and nine features for the enhanced spectral CT set. The AUC of the shape model was significantly smaller than that of the other four models (all P < 0.05). The combined model (AUC = 0.874, 95%CI: 0.821-0.916) outperformed the material-separation model (AUC = 0.828, 95%CI: 0.769-0.877, P = 0.005), the plain spectral CT model (AUC = 0.820 95%CI: 0.760-0.870, P = 0.005) and the enhanced spectral CT model (AUC = 0.838, 95%CI: 0.780-0.886, P = 0.005). Conclusion:The radiomics features derived from spectral CT metrics will enhance the differentiation of de novo OBM and BI in newly diagnosed cancer patients.

PurposeThere is a lack of research evaluating the clinical performance of virtual contrast-enhanced MRI (VCE-MRI). This study aims to assess the clinical utility of an established VCE-MRI technique in NPC patients and to establish patient selection criteria.Materials and methodsWe retrospectively collected data from 333 NPC patients across six institutions (2012-2023). VCE-MRI was synthesized from T1-weighted and T2-weighted MRI for each patient using the multimodality-guided synergistic neural network (MMgSN-Net), which was pre-trained on a large cohort of 1682 NPC patients from 14 institutions. Three experienced radiologists independently assessed image quality using a 5-point Likert scale, with scores below 4 deemed clinically unacceptable. The association between assessment results and patient clinical characteristics of corresponding patients (tumor diameter, T-stage, shape, gender, age) was analyzed to stratify the patients suitable for MMgSN-Net-based VCE-MRI.ResultsClinically acceptability of VCE-MRI significantly decreased with larger tumor diameters (p=0.001), advanced T-stage (T1: 100%, T2: 95.1%, T3: 90.0%, T4: 69.6%; p<0.001), and complex tumor shape (regular: 94.8% vs. complex: 74.0%; p<0.001). Age and gender had no significant impact (p=0.327, 0.773). 13 unacceptable patients were found due to severe VCE-MRI artifacts. VCE-MRI achieved high clinical acceptability in T1/T2 patients (109/114, 95.6%) and T3/T4 patients with regular tumor shapes (115/120, 95.8%). In contrast, acceptability significantly decreased to 73.0% (65/89) for T3/T4 tumors with complex shapes. The T-staging result discordance rate between VCE-MRI and CE-MRI is 8.4%, with overstaging in 15 cases.ConclusionMMgSN-Net-based VCE-MRI demonstrates favorable clinical feasibility in selected patient subgroups, T1/T2-stage NPC and T3/T4-stage NPC with regular tumor morphology, potentially reducing GBCAs administration in 67.3% of eligible patients when applying these stratification criteria.

Background and Purpose: Precise delineation of pelvic organs-at-risk (OARs) is crucial for high-dose-rate brachytherapy (HDR-BT) in cervical cancer treatment. While deep learning methods have shown promise in automatic delineation, substantial and complex organ deformations pose significant challenges. This study presents a novel approach to address these issues. Materials and Methods: We introduce a coarse-to-refine strategy for annotation, utilizing limited existing data to expedite the process. Combined with deformation-based data augmentation, we incorporate this information into a three-dimensional attention U-Net (C2FAU-Net). The study included 100 cervical cancer patients, with OARs annotated by experienced oncologists. The dataset was divided into 80 patients for training, 10 for validation, and 10 for testing. To assess the delineation performance, we employed the volumetric dice similarity coefficient (DSC), 95th percentile Hausdorff Distance (HD95), average symmetric surface distance (ASSD), precision, and recall. We compared the time consumed by manual delineation versus artificial intelligence (AI)-assisted delineation. Dosimetric parameters were compared using different contours to evaluate the clinical impact of the automated approach. Results: Our method achieved an average DSC of 89.7%, HD95 of 3.61 mm, and an ASSD of 1.02 mm in the test cohort. The AI-assisted method significantly reduced the manual delineation time from 17.85 ± 3.84 min to 7.54 ± 4.95 min. No significant difference was observed in ΔD2cc, ΔD1cc, ΔD0.1cc, and ΔDmax for bladder, rectum, and sigmoid when comparing contours generated by C2FAU-Net to those created manually. Conclusion: We introduce an effective automatic delineation framework for pelvic OARs, enhancing efficiency within the HDR-BT workflow and potentially improving treatment outcomes.

In modern radiation therapy for head and neck cancers, the treatment related toxicities remain a significant clinical challenge. This review critically evaluates the evolution of data-driven approaches in predicting patient outcomes in head and neck cancer patients treated with radiation therapy. Three transformative methodological advances are reviewed: radiomics, AI-based algorithms, and causal inference frameworks. The integration of linear energy transfer in patient outcomes study, which has uncovered critical mechanisms behind unexpected toxicity, was also introduced for proton therapy. While radiomics has transformed medical image analysis through comprehensive quantitative characterization, AI models have demonstrated markedly superior predictive capabilities over traditional approaches, offering promising avenues for personalized radiation therapy with reduced toxicity profiles. However, the field faces significant challenges in translating statistical correlations from real-world data into interventional clinical insights. We highlight how causal inference methods can bridge this gap by providing a rigorous framework for identifying treatment effects. Looking ahead, we envision that combining these complementary approaches, especially the interventional prediction models, will enable more personalized treatment strategies, ultimately improving both tumor control and quality of life for head and neck cancer patients treated with radiation therapy.

We proposed Med-VLM (Medical Vision-language Model), an innovative approach that leverages textual descriptions of organs to enhance segmentation accuracy in medical images. Existing medical image segmentation methods face several challenges: (1) Current medical segmentation models often fail to effectively incorporate valuable prior knowledge, such as detailed descriptions of organ locations and characteristics. (2) Most text-visual models prioritize target identification, rather than focusing on enhancing overall accuracy. (3) While some approaches attempt to use prior knowledge for accuracy enhancement, they often fall short in effectively incorporating pre-trained models. To overcome these limitations, Med-VLM introduced several key innovations: low-rank adaptation, authoritative descriptions, BioBERT weights, and a feature mixer. We conducted a comprehensive evaluation of MedVLM using three authoritative medical image datasets, covering the segmentation of various human body parts. Our method demonstrated superior performance compared to existing state-of-the-art approaches, including Lvit, Med-SAM, SAM, and nnUnet. We designed a series of ablation experiments, which systematically assessed the contribution of each component of Med-VLM, providing insights into the model's performance characteristics.

Proton therapy offers significant advantages due to its unique physical and biological properties, particularly the Bragg peak, enabling precise dose delivery to tumors while sparing healthy tissues. However, the clinical implementation is challenged by the oversimplification of the relative biological effectiveness (RBE) as a fixed value of 1.1, which does not account for the complex interplay between dose, linear energy transfer (LET), and biological endpoints. Lack of heterogeneity control or the understanding of the complex interplay may result in unexpected adverse events and suboptimal patient outcomes. On the other hand, expanding our knowledge of variable tumor RBE and LET optimization may provide a better management strategy for radioresistant tumors. This review examines recent advancements in LET calculation methods, including analytical models and Monte Carlo simulations. The integration of LET into plan evaluation is assessed to enhance plan quality control. LET-guided robust optimization demonstrates promise in minimizing high-LET exposure to organs at risk, thereby reducing the risk of adverse events. Dosimetric seed spot analysis is discussed to show its importance in revealing the true LET-related effect upon the adverse event initialization by finding the lesion origins and eliminating the confounding factors from the biological processes. Dose-LET volume histograms (DLVH) are discussed as effective tools for correlating physical dose and LET with clinical outcomes, enabling the derivation of clinically relevant dose-LET volume constraints without reliance on uncertain RBE models. Based on DLVH, the dose-LET volume constraints (DLVC)-guided robust optimization is introduced to upgrade conventional dose-volume constraints-based robust optimization, which optimizes the joint distribution of dose and LET simultaneously. In conclusion, translating the advances in LET-related research into clinical practice necessitates a better understanding of the LET-related biological mechanisms and the development of clinically relevant LET-related volume constraints directly derived from the clinical outcomes. Future research is needed to refine these models and conduct prospective trials to assess the clinical benefits of LET-guided optimization on patient outcomes.

Background:Postoperative radiotherapy is standard for high-risk cervical cancer, but acute toxicities-particularly gastrointestinal and hematologic-remain clinically relevant. Patient positioning may influence organ dose exposure and setup accuracy, yet its multidimensional clinical impact is poorly characterized. Methods:This retrospective cohort study evaluated patients with cervical cancer treated with postoperative volumetric modulated arc therapy between 2019 and 2022. Propensity score matching (2:1) produced a balanced matched cohort of prone and supine treatments for comparative analyses. Primary endpoints included pelvic organ dose-volume parameters, interfractional setup error, and grade ≥2 hematologic and gastrointestinal toxicities, evaluated using multivariable logistic regression and linear mixed-effects models. Results:In this single-center retrospective cohort (n = 168), propensity score matching (2:1) yielded 112 balanced patients (prone n = 70; supine n = 42). After matching, target coverage was comparable between positions (PTV_D95: 45.52 Gy vs 45.54 Gy, p = 0.24). The prone group showed higher low-dose exposure in bowel bag and rectum at V5-V15 (e.g., V10 difference -9.84%, 95% CI -17.07 to 1.08; adjusted p = 0.040). Setup error was similar across all axes (p > 0.05). The supine group had significantly higher incidence of leukopenia (92.9% vs 71.4%; p = 0.0073), with prone positioning associated with reduced hematologic toxicity (OR = 14.40, 95% CI 1.60-129.74; p = 0.017). Conversely, diarrhea occurred more often in the prone group (44.3% vs 26.2%, p = 0.070), and supine positioning was protective in multivariable analysis (OR = 0.42, 95% CI 0.17-0.97; p = 0.047). Conclusion:These findings suggest prone positioning may be preferable for patients with limited hematopoietic reserve, while supine positioning may benefit those with gastrointestinal vulnerability. Positioning choice should be individualized based on toxicity risk and functional anatomy to optimize safety in postoperative cervical cancer radiotherapy.

Purpose Automating quality assurance (QA) for contours generated by automatic algorithms is critical in radiotherapy treatment planning. Manual QA is tedious, time-consuming, and prone to subjective experiences. Automatic segmentation reduces physician workload and improves consistency. However, an effective QA process for these automatic contours remains an unmet need in clinical practice. Materials and Methods The patient data used in this study was derived from the AAPM Thoracic Auto-Segmentation Challenge dataset, including left and right lungs, heart, esophagus, and spinal cord. Two groups of organ-at-risk (OAR) were generated. A ResNet-152 network was used as a feature extractor, and a one-class support vector machine (OC-SVM) was employed to classify contours as ‘high’ or ‘low’ quality. To evaluate the generalizability, we generated low-quality contours using translation and resizing techniques and assessed correlations between detection limits and metrics such as volume, Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and mean surface distance (MSD). Results The proposed OC-SVM model outperformed binary classifiers n metrics such as balanced accuracy and area under the receiver operating characteristic curve (AUC) . It demonstrated superior performance in detecting various types of contour errors while maintaining high interpretability. Strong correlations were observed between detection limits and contour metrics. Conclusion Our proposed model integrates an attention mechanism with a one-class classification framework to automate QA for OAR delineations. This approach effectively detects diverse types of contour errors with high accuracy, significantly reducing the burden on physicians during radiotherapy planning.

We present the Radiation Oncology NLP Database (ROND), the first dedicated Natural Language Processing (NLP) dataset for radiation oncology, an important medical specialty that has received limited attention from the NLP community in the past. With the advent of Artificial General Intelligence (AGI), there is an increasing need for specialized datasets and benchmarks to facilitate research and development. ROND is specifically designed to address this gap in the domain of radiation oncology, a field that offers many opportunities for NLP exploration. It encompasses various NLP tasks including Logic Reasoning, Text Classification, Named Entity Recognition (NER), Question Answering (QA), Text Summarization, and Patient-Clinician Conversations, each with a distinct focus on radiation oncology concepts and application cases. In addition, we have developed an instruction-tuning dataset consisting of over 20k instruction pairs (based on ROND) and trained a large language model, CancerChat. This serves to demonstrate the potential of instruction-tuning large language models within a highly-specialized medical domain. The evaluation results in this study could serve as baseline results for future research. ROND aims to stimulate advancements in radiation oncology and clinical NLP by offering a platform for testing and improving algorithms and models in a domain-specific context. The ROND dataset is a joint effort of multiple U.S. health institutions. The data is available at https://github.com/zl-liu/Radiation-Oncology-NLP-Database.

The delineation of tumor target and organs-at-risk is critical in the radiotherapy treatment planning. Automatic segmentation can be used to reduce the physician workload and improve the consistency. However, the quality assurance of the automatic segmentation is still an unmet need in clinical practice. The patient data used in our study was a standardized dataset from AAPM Thoracic Auto-Segmentation Challenge. The OARs included were left and right lungs, heart, esophagus, and spinal cord. Two groups of OARs were generated, the benchmark dataset manually contoured by experienced physicians and the test dataset automatically created using a software AccuContour. A resnet-152 network was performed as feature extractor, and one-class support vector classifier was used to determine the high or low quality. We evaluate the model performance with balanced accuracy, F-score, sensitivity, specificity and the area under the receiving operator characteristic curve. We randomly generated contour errors to assess the generalization of our method, explored the detection limit, and evaluated the correlations between detection limit and various metrics such as volume, Dice similarity coefficient, Hausdorff distance, and mean surface distance. The proposed one-class classifier outperformed in metrics such as balanced accuracy, AUC, and others. The proposed method showed significant improvement over binary classifiers in handling various types of errors. Our proposed model, which introduces residual network and attention mechanism in the one-class classification framework, was able to detect the various types of OAR contour errors with high accuracy. The proposed method can significantly reduce the burden of physician review for contour delineation.

Purpose Hippocampal-avoidance whole-brain radiotherapy (HA-WBRT) planning can present challenges. This study examines the influence of head tilt angles on the dosimetric characteristics of target and organs at risk (OARs), aiming to identify the optimal tilt angle that yields optimal dosimetric outcomes using tomotherapy (TOMO). Methods Eight patients diagnosed with brain metastases underwent CT scans at five tilt angles: [0°, 10°), [10°, 20°), [20°, 30°), [30°, 40°), and [40°, 45°]. Treatment plans were generated using TOMO and volumetric modulated arc therapy (VMAT). Dosimetric parameters including conformity index (CI), homogeneity index (HI), D 2cc , D 98% , and D mean of PTV, as well as D max , and D mean of OARs were analyzed. Furthermore, a comparison was made between the dosimetric parameters of TOMO and VMAT plans. Finally, delivery efficiency of TOMO plans were assessed. Results For the PTV, [40°, 45°] tilt angle demonstrated significantly better conformity, homogeneity, lower D 2cc , and lower D mean for the PTV. Regarding the OARs, the [40°, 45°] head tilt angle demonstrated significantly lower D max and D mean in hippocampus, eyes, optic chiasm, and optic nerves. The [40°, 45°] tilt angle also showed significantly lower D max for brainstem and cochleas, as well as a lower D mean for lens. In the [40°,45°] tilt angle for HA-WBRT, TOMO showed superior performance over VMAT for the PTV. TOMO achieved lower D max for brainstem, cochleas, optic nerves, and optic chiasm, as well as a lower D mean for hippocampus. Furthermore, a significant correlation was found between delivery time and the PTV projection length in the sagittal plane. Conclusion The TOMO plan utilizing a tilt angle range of [40°, 45°] demonstrated superior PTV conformity and uniformity, along with enhanced OARs sparing. Furthermore, it exhibited a dosimetric advantage over VMAT for PTV and most OARs at the same angle range.

We propose TG-LMM (Text-Guided Large Multi-Modal Model), a novel approach that leverages textual descriptions of organs to enhance segmentation accuracy in medical images. Existing medical image segmentation methods face several challenges: current medical automatic segmentation models do not effectively utilize prior knowledge, such as descriptions of organ locations; previous text-visual models focus on identifying the target rather than improving the segmentation accuracy; prior models attempt to use prior knowledge to enhance accuracy but do not incorporate pre-trained models. To address these issues, TG-LMM integrates prior knowledge, specifically expert descriptions of the spatial locations of organs, into the segmentation process. Our model utilizes pre-trained image and text encoders to reduce the number of training parameters and accelerate the training process. Additionally, we designed a comprehensive image-text information fusion structure to ensure thorough integration of the two modalities of data. We evaluated TG-LMM on three authoritative medical image datasets, encompassing the segmentation of various parts of the human body. Our method demonstrated superior performance compared to existing approaches, such as MedSAM, SAM and nnUnet.

The delineation of tumor target and organs-at-risk (OARs) is critical in the radiotherapy treatment planning. It is also tedious, time-consuming and prone to subjective experiences. Automatic segmentation can be used to reduce the physician’s workload. However, the quality assurance of the segmentation is an unmet need in clinical practice. In this study, we developed an automatic model that detects the errors of the contouring using one-class classifier. The OARs included left and right lungs, heart, esophagus, and spinal cord. Each data includes the ground truth, which is manually contoured by experienced doctor, and contour generated by a contouring software. We used three metrics to determine whether the contour of an OAR is "high" or "low" quality. A resnet-152 network performed as a feature extractor, and a one class support vector machine determines the quality of the contour. We generated certain contour errors to evaluate the generalizability of this method. Furthermore, to enhance the interpretability of this method, we conducted a set of experiments to assess its detection limit and discussed the correlation between this limit and metrics such as volume, DSC, HD95, and MSD. The proposed method showed significant improvement over binary classifiers in handling various types of errors. The relationship between the detection limit and multiple factors of the OARs indicates that our method is highly interpretable. Moreover, the model's fast execution speed can significantly reduce the burden on physicians.

PURPOSE:The aim of this work was to provide a method to evaluate the yield of DNA double-strand breaks (DSBs) for carbon ions, overcoming the bias in existing methods due to the nonrandom distribution of DSBs. METHODS AND MATERIALS:A previously established biophysical program based on the radiation track structure and a multilevel chromosome model was used to simulate DNA damage induced by x-rays and carbon ions. The fraction of activity retained (FAR) as a function of absorbed dose or particle fluence was obtained by counting the fraction of DNA fragments larger than 6 Mbp. Simulated FAR curves for the 250 kV x-rays and carbon ions at various energies were compared with measurements using constant-field gel electrophoresis. The doses or fluences at the FAR of 0.7 based on linear interpolation were used to estimate the simulation error for the production of DSBs. RESULTS:The relative difference of doses at the FAR of 0.7 between simulation and experiment was -8.5% for the 250 kV x-rays. The relative differences of fluences at the FAR of 0.7 between simulations and experiments were -17.5%, -42.2%, -18.2%, -3.1%, 10.8%, and -14.5% for the 34, 65, 130, 217, 2232, and 3132 MeV carbon ions, respectively. In comparison, the measurement uncertainty was about 20%. Carbon ions produced remarkably more DSBs and DSB clusters per unit dose than x-rays. The yield of DSBs for carbon ions, ranging from 10 to 16 Gbp-1Gy-1, increased with linear energy transfer (LET) but plateaued in the high-LET end. The yield of DSB clusters first increased and then decreased with LET. This pattern was similar to the relative biological effectiveness for cell survival for heavy ions. CONCLUSIONS:The estimated yields of DSBs for carbon ions increased from 10 Gbp-1Gy-1 in the low-LET end to 16 Gbp-1Gy-1 in the high-LET end with 20% uncertainty.

This paper presents RadOnc-GPT, a large language model specialized for radiation oncology through advanced tuning methods. RadOnc-GPT was finetuned on a large dataset of radiation oncology patient records from the Mayo Clinic in Arizona. The model employs instruction tuning on three key tasks - generating radiotherapy treatment regimens, determining optimal radiation modalities, and providing diagnostic descriptions/ICD codes based on patient diagnostic details. Evaluations conducted by comparing RadOnc-GPT outputs to general large language model outputs showed higher ROUGE scores in these three tasks. The study demonstrated the potential of using large language models fine-tuned using domain-specific knowledge like RadOnc-GPT to achieve transformational capabilities in highly specialized healthcare fields such as radiation oncology. However, our model's clinical relevance requires confirmation, and it specializes in only the aforementioned three specific tasks and lacks broader applicability. Furthermore, its evaluation through ROUGE scores might not reflect the true semantic and clinical accuracy - challenges we intend to address in future research.

The emergence of artificial general intelligence (AGI) is transforming radiation oncology. As prominent vanguards of AGI, large language models (LLMs) such as GPT-4 and PaLM 2 can process extensive texts and large vision models (LVMs) such as the Segment Anything Model (SAM) can process extensive imaging data to enhance the efficiency and precision of radiation therapy. This paper explores full-spectrum applications of AGI across radiation oncology including initial consultation, simulation, treatment planning, treatment delivery, treatment verification, and patient follow-up. The fusion of vision data with LLMs also creates powerful multimodal models that elucidate nuanced clinical patterns. Together, AGI promises to catalyze a shift towards data-driven, personalized radiation therapy. However, these models should complement human expertise and care. This paper provides an overview of how AGI can transform radiation oncology to elevate the standard of patient care in radiation oncology, with the key insight being AGI's ability to exploit multimodal clinical data at scale.

To investigate measurements derived from plain and enhanced spectral CT in differentiating osteoblastic bone metastasis (OBM) from bone island (BI). From January to November 2020, 73 newly diagnosed cancer patients with 201 bone lesions (OBM = 92, BI = 109) having received spectral CT were retrospectively enrolled. Measurements including CT values of 40–140 keV, slope of the spectral curve, effective atomic number (Zeff), water (calcium) density, calcium (water) density, and Iodine (calcium) density were derived from manually segmented lesions on plain and enhanced spectral CT, and then analyzed using Student t-test and Pearson’s correlation. Multivariate analysis was performed to build models (plain spectral model, enhanced spectral CT model, and combined model) for the discrimination of OBM and BI with performance evaluated using receiver operator characteristics curve and DeLong test. All features were significantly different between the BI group and OBM group (all p < 0.05), highly correlated with the corresponding features between plain and enhanced spectral CT both in OBM (r: 0.392–0.763) and BI (r: 0.430–0.544). As for the model performance, the combined model achieved the best performance (AUC = 0.925, 95 • We intend to investigate plain and enhanced spectral CT measurements in differentiating OBM from BI. • Both plain and enhanced spectral CT help in discriminating OBM and BI in newly diagnosed cancer patients. • Enhanced spectral CT measurements further improve plain spectral CT measurements-based differential diagnosis.

Rapid detection of pathogenic bacteria within a few minutes is the key to control infec-tious disease. However, rapid detection of pathogenic bacteria in clinical samples is quite a challenging task due to the complex matrix, as well as the low abundance of bacteria in real samples. Herein, we employ a label-free single-particle imaging approach to address this challenge. By tracking the scattering intensity variation of single particles in free solu-tion, the morphological heterogeneity can be well identified with particle size smaller than the diffraction limit, facilitating the morphological identification of single bacteria from a complex matrix in a label-free manner. Furthermore, the manipulation of convec-tion in free solution enables the rapid screening of low-abundance bacteria in a small field of view, which significantly improves the sensitivity of single-particle detection. As a proof of concept demonstration, we are able to differentiate the group B streptococci (GBS)-positive samples within 10 min from vaginal swabs without using any biological reagents. This is the most rapid and low-cost method to the best of our knowledge. We believe that such a single-particle imaging approach will find wider applications in clinical diagnosis and disease control due to its high sensitivity, rapidity, simplicity, and low cost.