Radiology reports convey critical medical information to health care providers and patients. Unfortunately, they are often difficult for patients to comprehend, causing confusion and anxiety, thereby limiting patient engagement in health care decision-making. Large language models (LLMs) like ChatGPT (OpenAI) can create simplified, patient-friendly report summaries to increase accessibility, albeit with errors. We evaluated the accuracy and clarity of ChatGPT-generated summaries compared to original radiologist-assessed radiology reports, assessed patients' understanding and satisfaction with the summaries compared to the original reports, and compared the readability of the original reports and summaries using validated readability metrics. We anonymized 30 radiology reports created by neuroradiologists at our institution (6 brain magnetic resonance imaging, 6 brain computed tomography, 6 head and neck computed tomography angiography, 6 neck computed tomography, and 6 spine computed tomography). These anonymized reports were processed by ChatGPT to produce patient-centric summaries. Four board-certified neuroradiologists evaluated the ChatGPT-generated summaries on quality and accuracy compared to the original reports, and 4 patient volunteers separately evaluated the reports and summaries on perceived understandability and satisfaction. Readability was assessed using word count and validated readability scales. After reading the summary, patient confidence in understanding (98%, 116/118 vs 26%, 31/118) and satisfaction regarding the level of jargon/terminology (91%, 107/118 vs 8%, 9/118) and time taken to understand the content (97%, 115/118 vs 23%, 27/118) substantially improved. Ninety-two percent (108/118) of responses indicated the summary clarified patients' questions about the report, and 98% (116/118) of responses indicated patients would use the summary if available, with 67% (79/118) of responses indicating they would want access to both the report and summary, while 26% (31/118) of responses indicated only wanting the summary. Eighty-three percent (100/120) of radiologist responses indicated the summary represented the original report "extremely well" or "very well," with only 5% (6/120) of responses indicating it did so "slightly well" or "not well at all." Five percent (6/120) of responses indicated there was missing relevant medical information in the summary, 12% (14/120) reported instances of overemphasis of nonsignificant findings, and 18% (22/120) reported instances of underemphasis of significant findings. No fabricated findings were identified. Overall, 83% (99/120) of responses indicated that the summary would definitely/probably not lead patients to incorrect conclusions about the original report, with 10% (12/120) of responses indicating the summaries may do so. ChatGPT-generated summaries could significantly improve perceived comprehension and satisfaction while accurately reflecting most key information from original radiology reports. Instances of minor omissions and under-/overemphasis were noted in some summaries, underscoring the need for ongoing validation and oversight. Overall, these artificial intelligence-generated, patient-centric summaries hold promise for enhancing patient-centered communication in radiology.

Recently, deep neural networks have greatly advanced undersampled Magnetic Resonance Image (MRI) reconstruction, wherein most studies follow the one-anatomy-one-network fashion, i.e., each expert network is trained and evaluated for a specific anatomy. Apart from inefficiency in training multiple independent models, such convention ignores the shared de-aliasing knowledge across various anatomies which can benefit each other. To explore the shared knowledge, one naive way is to combine all the data from various anatomies to train an all-round network. Unfortunately, despite the existence of the shared de-aliasing knowledge, we reveal that the exclusive knowledge across different anatomies can deteriorate specific reconstruction targets, yielding overall performance degradation. Observing this, in this study, we present a novel deep MRI reconstruction framework with both anatomy-shared and anatomy-specific parameterized learners, aiming to "seek common ground while reserving differences" across different anatomies. Particularly, the primary anatomy-shared learners are exposed to different anatomies to model rich shared de-aliasing knowledge, while the efficient anatomy-specific learners are trained with their target anatomy for exclusive knowledge. Four different implementations of anatomy-specific learners are presented and explored on the top of our framework in two MRI reconstruction networks. Comprehensive experiments on brain, knee and cardiac MRI datasets demonstrate that three of these learners are able to enhance reconstruction performance via multiple anatomy collaborative learning. Extensive studies show that our strategy can also benefit multiple pulse sequence MRI reconstruction by integrating sequence-specific learners.

X-ray image based medical report generation achieves significant progress in recent years with the help of large language models, however, these models have not fully exploited the effective information in visual image regions, resulting in reports that are linguistically sound but insufficient in describing key diseases. In this paper, we propose a novel associative memory-enhanced X-ray report generation model that effectively mimics the process of professional doctors writing medical reports. It considers both the mining of global and local visual information and associates historical report information to better complete the writing of the current report. Specifically, given an X-ray image, we first utilize a classification model along with its activation maps to accomplish the mining of visual regions highly associated with diseases and the learning of disease query tokens. Then, we employ a visual Hopfield network to establish memory associations for disease-related tokens, and a report Hopfield network to retrieve report memory information. This process facilitates the generation of high-quality reports based on a large language model and achieves state-of-the-art performance on multiple benchmark datasets, including the IU X-ray, MIMIC-CXR, and Chexpert Plus. The source code and pre-trained models of this work have been released on https://github.com/Event-AHU/Medical_Image_Analysis.

Markerless tumor tracking in x-ray fluoroscopic images is an important technique for achieving precise dose delivery for moving lung tumors during radiation therapy. However, accurate tumor tracking is challenging due to the poor visibility of the target tumor overlapped by other organs such as rib bones. Dual-energy (DE) x-ray fluoroscopy can enhance tracking accuracy with improved tumor visibility by suppressing bones. However, DE x-ray imaging requires special hardware, limiting its clinical use. This study presents a deep learning-based DE subtraction (DES) synthesis method to avoid hardware limitations and enhance tracking accuracy. The proposed method employs a residual U-Net model trained on a simulated DES dataset from a digital phantom to synthesize DES from single-energy (SE) fluoroscopy. Experimental results using a digital phantom showed quantitative evaluation results of synthesis quality. Also, experimental results using clinical SE fluoroscopic images of ten lung cancer patients showed improved tumor tracking accuracy using synthesized DES images, reducing errors from 1.80 to 1.68 mm on average. The tracking success rate within a 25% movement range increased from 50.2% (SE) to 54.9% (DES). These findings indicate the feasibility of deep learning-based DES synthesis for markerless tumor tracking, offering a potential alternative to hardware-dependent DE imaging.

The onset of liver-related events (LREs) in fibrosis indicates a poor prognosis and worsens patients' quality of life, making the prediction and early detection of LREs crucial. The aim of this study was to develop a radiomics model using liver magnetic resonance imaging (MRI) to predict LRE risk in patients undergoing antiviral treatment for chronic fibrosis caused by hepatitis B virus (HBV). Patients with HBV-associated liver fibrosis and liver stiffness measurements ≥ 10 kPa were included. Feature selection and dimensionality reduction techniques identified discriminative features from three MRI sequences. Radiomics models were built using eight machine learning techniques and evaluated for performance. Shapley additive explanation and permutation importance techniques were applied to interpret the model output. A total of 222 patients aged 49 ± 10 years (mean ± standard deviation), 175 males, were evaluated, with 41 experiencing LREs. The radiomics model, incorporating 58 selected features, outperformed traditional clinical tools in prediction accuracy. Developed using a support vector machine classifier, the model achieved optimal areas under the receiver operating characteristic curves of 0.94 and 0.93 in the training and test sets, respectively, demonstrating good calibration. Machine learning techniques effectively predicted LREs in patients with fibrosis and HBV, offering comparable accuracy across algorithms and supporting personalized care decisions for HBV-related liver disease. Radiomics models based on liver multisequence MRI can improve risk prediction and management of patients with HBV-associated chronic fibrosis. In addition, it offers valuable prognostic insights and aids in making informed clinical decisions. Liver-related events (LREs) are associated with poor prognosis in chronic fibrosis. Radiomics models could predict LREs in patients with hepatitis B-associated chronic fibrosis. Radiomics contributes to personalized care choices for patients with hepatitis B-associated fibrosis.

<i>"Just Accepted" papers have undergone full peer review and have been accepted for publication in <i>Radiology: Artificial Intelligence</i>. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content</i>. Purpose To evaluate the technical feasibility of implementing homomorphic encryption in deep learning models for privacy-preserving CT image analysis of renal masses. Materials and Methods A privacy-preserving deep learning system was developed through three sequential technical phases: a reference CNN model (Ref-CNN) based on ResNet architecture, modification for encryption compatibility (Approx-CNN) by replacing ReLU with polynomial approximation and max-pooling with averagepooling, and implementation of fully homomorphic encryption (HE-CNN). The CKKS encryption scheme was used for its capability to perform arithmetic operations on encrypted real numbers. Using 12,446 CT images from a public dataset (3,709 renal cysts, 5,077 normal kidneys, and 2,283 kidney tumors), we evaluated model performance using area under the receiver operating characteristic curve (AUC) and area under the precision-recall curve (AUPRC). Results All models demonstrated high diagnostic accuracy with AUC ranging from 0.89-0.99 and AUPRC from 0.67-0.99. The diagnostic performance trade-off was minimal from Ref-CNN to Approx-CNN (AUC: 0.99 to 0.97 for normal category), with no evidence of differences between models. However, encryption significantly increased storage and computational demands: a 256 × 256-pixel image expanded from 65KB to 32MB, requiring 50 minutes for CPU inference but only 90 seconds with GPU acceleration. Conclusion This technical development demonstrates that privacy-preserving deep learning inference using homomorphic encryption is feasible for renal mass classification on CT images, achieving comparable diagnostic performance while maintaining data privacy through end-to-end encryption. ©RSNA, 2025.

To improve sensitivity and inter-reader consistency of prostate cancer localisation on micro-ultrasonography (MUS) by developing a deep learning model for automatic cancer segmentation, and to compare model performance with that of expert urologists. We performed an institutional review board-approved prospective collection of MUS images from patients undergoing magnetic resonance imaging (MRI)-ultrasonography fusion guided biopsy at a single institution. Patients underwent 14-core systematic biopsy and additional targeted sampling of suspicious MRI lesions. Biopsy pathology and MRI information were cross-referenced to annotate the locations of International Society of Urological Pathology Grade Group (GG) ≥2 clinically significant cancer on MUS images. We trained a no-new U-Net model - the Prostate Micro-Ultrasound Network (ProMUS-NET) - to localise GG ≥2 cancer on these image stacks in a fivefold cross-validation. Performance was compared vs that of six expert urologists in a matched sub-cohort. The artificial intelligence (AI) model achieved an area under the receiver-operating characteristic curve of 0.92 and detected more cancers than urologists (lesion-level sensitivity 73% vs 58%; patient-level sensitivity 77% vs 66%). AI lesion-level sensitivity for peripheral zone lesions was 86.2%. Our AI model identified prostate cancer lesions on MUS with high sensitivity and specificity. Further work is ongoing to improve margin overlap, to reduce false positives, and to perform external validation. AI-assisted prostate cancer detection on MUS has great potential to improve biopsy diagnosis by urologists.

Dual-energy X-ray absorptiometry (DXA) is the standard method for assessing areal bone mineral density (aBMD), diagnosing osteoporosis, and predicting fracture risk. However, DXA's availability is limited in resource-poor areas. This study aimed to develop an artificial intelligence (AI) system capable of estimating aBMD from standard radiographs. The study was part of the Vietnam Osteoporosis Study, a prospective population-based research involving 3783 participants aged 18 years and older. A total of 7060 digital radiographs of the frontal pelvis and lateral spine were taken using the FCR Capsula XLII system (Fujifilm Corp., Tokyo, Japan). aBMD at the femoral neck and lumbar spine was measured with DXA (Hologic Horizon, Hologic Corp., Bedford, MA, USA). An ensemble of seven deep-learning models was used to analyze the X-rays and predict bone mineral density, termed "xBMD". The correlation between xBMD and aBMD was evaluated using Pearson's correlation coefficients. The correlation between xBMD and aBMD at the femoral neck was strong ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mi>r</mi></math> = 0.90; 95% CI, 0.88-0.91), and similarly high at the lumbar spine ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mi>r</mi></math> = 0.87; 95% CI, 0.85-0.88). This correlation remained consistent across different age groups and genders. The AI system demonstrated excellent performance in identifying individuals at high risk for hip fractures, with area under the ROC curve (AUC) values of 0.96 (95% CI, 0.95-0.98) at the femoral neck and 0.97 (95% CI, 0.96-0.99) at the lumbar spine. These findings indicate that AI can accurately predict aBMD and identify individuals at high risk of fractures. This AI system could provide an efficient alternative to DXA for osteoporosis screening in settings with limited resources and high patient demand. An AI system developed to predict aBMD from X-rays showed strong correlations with DXA ( <math xmlns="http://www.w3.org/1998/Math/MathML"><mi>r</mi></math> = 0.90 at femoral neck; =  <math xmlns="http://www.w3.org/1998/Math/MathML"><mi>r</mi></math> 0.87 at lumbar spine) and high accuracy in identifying individuals at high risk for fractures (AUC = 0.96 at femoral neck; AUC = 0.97 at lumbar spine).

Retrieval-augmented learning based on radiology reports has emerged as a promising direction to improve performance on long-tail medical imaging tasks, such as rare disease detection in chest X-rays. Most existing methods rely on comparing high-dimensional text embeddings from models like CLIP or CXR-BERT, which are often difficult to interpret, computationally expensive, and not well-aligned with the structured nature of medical knowledge. We propose a novel, ontology-driven alternative for comparing radiology report texts based on clinically grounded concepts from the Unified Medical Language System (UMLS). Our method extracts standardised medical entities from free-text reports using an enhanced pipeline built on RadGraph-XL and SapBERT. These entities are linked to UMLS concepts (CUIs), enabling a transparent, interpretable set-based representation of each report. We then define a task-adaptive similarity measure based on a modified and weighted version of the Tversky Index that accounts for synonymy, negation, and hierarchical relationships between medical entities. This allows efficient and semantically meaningful similarity comparisons between reports. We demonstrate that our approach outperforms state-of-the-art embedding-based retrieval methods in a radiograph classification task on MIMIC-CXR, particularly in long-tail settings. Additionally, we use our pipeline to generate ontology-backed disease labels for MIMIC-CXR, offering a valuable new resource for downstream learning tasks. Our work provides more explainable, reliable, and task-specific retrieval strategies in clinical AI systems, especially when interpretability and domain knowledge integration are essential. Our code is available at https://github.com/Felix-012/ontology-concept-distillation

HONeYBEE (Harmonized ONcologY Biomedical Embedding Encoder) is an open-source framework that integrates multimodal biomedical data for oncology applications. It processes clinical data (structured and unstructured), whole-slide images, radiology scans, and molecular profiles to generate unified patient-level embeddings using domain-specific foundation models and fusion strategies. These embeddings enable survival prediction, cancer-type classification, patient similarity retrieval, and cohort clustering. Evaluated on 11,400+ patients across 33 cancer types from The Cancer Genome Atlas (TCGA), clinical embeddings showed the strongest single-modality performance with 98.5% classification accuracy and 96.4% precision@10 in patient retrieval. They also achieved the highest survival prediction concordance indices across most cancer types. Multimodal fusion provided complementary benefits for specific cancers, improving overall survival prediction beyond clinical features alone. Comparative evaluation of four large language models revealed that general-purpose models like Qwen3 outperformed specialized medical models for clinical text representation, though task-specific fine-tuning improved performance on heterogeneous data such as pathology reports.