To compare the diagnostic performance of three proprietary large language models (LLMs)-Claude, GPT, and Gemini-in structuring free-text Japanese radiology reports for intracranial hemorrhage and skull fractures, and to assess the impact of three different prompting approaches on model accuracy. In this retrospective study, head CT reports from the Japan Medical Imaging Database between 2018 and 2023 were collected. Two board-certified radiologists established the ground truth regarding intracranial hemorrhage and skull fractures through independent review and consensus. Each radiology report was analyzed by three LLMs using three prompting strategies-Standard, Chain of Thought, and Self Consistency prompting. Diagnostic performance (accuracy, precision, recall, and F1-score) was calculated for each LLM-prompt combination and compared using McNemar's tests with Bonferroni correction. Misclassified cases underwent qualitative error analysis. A total of 3949 head CT reports from 3949 patients (mean age 59 ± 25 years, 56.2% male) were enrolled. Across all institutions, 856 patients (21.6%) had intracranial hemorrhage and 264 patients (6.6%) had skull fractures. All nine LLM-prompt combinations achieved very high accuracy. Claude demonstrated significantly higher accuracy for intracranial hemorrhage than GPT and Gemini, and also outperformed Gemini for skull fractures (p < 0.0001). Gemini's performance improved notably with Chain of Thought prompting. Error analysis revealed common challenges including ambiguous phrases and findings unrelated to intracranial hemorrhage or skull fractures, underscoring the importance of careful prompt design. All three proprietary LLMs exhibited strong performance in structuring free-text head CT reports for intracranial hemorrhage and skull fractures. While the choice of prompting method influenced accuracy, all models demonstrated robust potential for clinical and research applications. Future work should refine the prompts and validate these approaches in prospective, multilingual settings.

To evaluate the performance of large language models (LLMs) in automatically generating whole-organ MRI score (WORMS)-based structured MRI reports and predicting osteoarthritis (OA) severity for the knee. A total of 160 consecutive patients suspected of OA were included. Knee MRI reports were reviewed by three radiologists to establish the WORMS reference standard for 39 key features. GPT-4o and GPT-4o-mini were prompted using in-context knowledge (ICK) and chain-of-thought (COT) to generate WORMS-based structured reports from original reports and to automatically predict the OA severity. Four Orthopedic surgeons reviewed original and LLM-generated reports to conduct pairwise preference and difficulty tests, and their review times were recorded. GPT-4o demonstrated perfect performance in extracting the laterality of the knee (accuracy = 100%). GPT-4o outperformed GPT-4o mini in generating WORMS reports (Accuracy: 93.9% vs 76.2%, respectively). GPT-4o achieved higher recall (87.3% s 46.7%, p < 0.001), while maintaining higher precision compared to GPT-4o mini (94.2% vs 71.2%, p < 0.001). For predicting OA severity, GPT-4o outperformed GPT-4o mini across all prompt strategies (best accuracy: 98.1% vs 68.7%). Surgeons found it easier to extract information and gave more preference to LLM-generated reports over the original reports (both p < 0.001) while spending less time on each report (51.27 ± 9.41 vs 87.42 ± 20.26 s, p < 0.001). GPT-4o generated expert multi-feature, WORMS-based reports from original free-text knee MRI reports. GPT-4o with COT achieved high accuracy in categorizing OA severity. Surgeons reported greater preference and higher efficiency when using LLM-generated reports. The perfect performance of generating WORMS-based reports and the high efficiency and ease of use suggest that integrating LLMs into clinical workflows could greatly enhance productivity and alleviate the documentation burden faced by clinicians in knee OA. GPT-4o successfully generated WORMS-based knee MRI reports. GPT-4o with COT prompting achieved impressive accuracy in categorizing knee OA severity. Greater preference and higher efficiency were reported for LLM-generated reports.

Bone fracture risk assessment for osteoporotic patients is essential for implementing early countermeasures and preventing discomfort and hospitalization. Current methodologies, such as Fracture Risk Assessment Tool (FRAX), provide a risk assessment over a 5- to 10-year period rather than evaluating the bone's current health status. The database was collected by Ajou University Medical Center from 2017 to 2021. It included 9,260 patients, aged 55 to 99, comprising 242 femur fracture (FX) cases and 9,018 non-fracture (NFX) cases. To model the association of the bone's current health status with prevalent FXs, three prediction algorithms-extreme gradient boosting (XGB), support vector machine, and multilayer perceptron-were trained using two-dimensional dual-energy X-ray absorptiometry (2D-DXA) analysis results and subsequently benchmarked. The XGB classifier, which proved most effective, was then further refined using synthetic data generated by the adaptive synthetic oversampler to balance the FX and NFX classes and enhance boundary sharpness for better classification accuracy. The XGB model trained on raw data demonstrated good prediction capabilities, with an area under the curve (AUC) of 0.78 and an F1 score of 0.71 on test cases. The inclusion of synthetic data improved classification accuracy in terms of both specificity and sensitivity, resulting in an AUC of 0.99 and an F1 score of 0.98. The proposed methodology demonstrates that current bone health can be assessed through post-processed results from 2D-DXA analysis. Moreover, it was also shown that synthetic data can help stabilize uneven databases by balancing majority and minority classes, thereby significantly improving classification performance.

Lung cancer is a leading cause of cancer-related deaths globally, where early detection and accurate diagnosis are critical for improving survival rates. While deep learning, particularly convolutional neural networks (CNNs), has revolutionized medical image analysis by detecting subtle patterns indicative of early-stage lung cancer, its adoption faces challenges. These models are often computationally expensive and require significant resources, making them unsuitable for resource constrained environments. Additionally, their lack of transparency hinders trust and broader adoption in sensitive fields like healthcare. Knowledge distillation addresses these challenges by transferring knowledge from large, complex models (teachers) to smaller, lightweight models (students). We propose a knowledge distillation-based approach for lung cancer detection, incorporating explainable AI (XAI) techniques to enhance model transparency. Eight CNNs, including ResNet50, EfficientNetB0, EfficientNetB3, and VGG16, are evaluated as teacher models. We developed and trained a lightweight student model, Distilled Custom Student Network (DCSNet) using ResNet50 as the teacher. This approach not only ensures high diagnostic performance in resource-constrained settings but also addresses transparency concerns, facilitating the adoption of AI-driven diagnostic tools in healthcare.

Cardiac image segmentation is an important step in many cardiac image analysis and modeling tasks such as motion tracking or simulations of cardiac mechanics. While deep learning has greatly advanced segmentation in clinical settings, there is limited work on pre-clinical imaging, notably in porcine models, which are often used due to their anatomical and physiological similarity to humans. However, differences between species create a domain shift that complicates direct model transfer from human to pig data. Recently, foundation models trained on large human datasets have shown promise for robust medical image segmentation; yet their applicability to porcine data remains largely unexplored. In this work, we investigate whether foundation models can generate sufficiently accurate pseudo-labels for pig cardiac CT and propose a simple self-training approach to iteratively refine these labels. Our method requires no manually annotated pig data, relying instead on iterative updates to improve segmentation quality. We demonstrate that this self-training process not only enhances segmentation accuracy but also smooths out temporal inconsistencies across consecutive frames. Although our results are encouraging, there remains room for improvement, for example by incorporating more sophisticated self-training strategies and by exploring additional foundation models and other cardiac imaging technologies.

Introduction: Timely identification of intracranial hemorrhage (ICH) subtypes on non-contrast computed tomography is critical for prognosis prediction and therapeutic decision-making, yet remains challenging due to low contrast and blurring boundaries. This study evaluates the performance of zero-shot multi-modal large language models (MLLMs) compared to traditional deep learning methods in ICH binary classification and subtyping. Methods: We utilized a dataset provided by RSNA, comprising 192 NCCT volumes. The study compares various MLLMs, including GPT-4o, Gemini 2.0 Flash, and Claude 3.5 Sonnet V2, with conventional deep learning models, including ResNet50 and Vision Transformer. Carefully crafted prompts were used to guide MLLMs in tasks such as ICH presence, subtype classification, localization, and volume estimation. Results: The results indicate that in the ICH binary classification task, traditional deep learning models outperform MLLMs comprehensively. For subtype classification, MLLMs also exhibit inferior performance compared to traditional deep learning models, with Gemini 2.0 Flash achieving an macro-averaged precision of 0.41 and a macro-averaged F1 score of 0.31. Conclusion: While MLLMs excel in interactive capabilities, their overall accuracy in ICH subtyping is inferior to deep networks. However, MLLMs enhance interpretability through language interactions, indicating potential in medical imaging analysis. Future efforts will focus on model refinement and developing more precise MLLMs to improve performance in three-dimensional medical image processing.

In neuroimaging research, the utilization of multi-centric analyses is crucial for obtaining sufficient sample sizes and representative clinical populations. Data harmonization techniques are typically part of the pipeline in multi-centric studies to address systematic biases and ensure the comparability of the data. However, most multi-centric studies require centralized data, which may result in exposing individual patient information. This poses a significant challenge in data governance, leading to the implementation of regulations such as the GDPR and the CCPA, which attempt to address these concerns but also hinder data access for researchers. Federated learning offers a privacy-preserving alternative approach in machine learning, enabling models to be collaboratively trained on decentralized data without the need for data centralization or sharing. In this paper, we present Fed-ComBat, a federated framework for batch effect harmonization on decentralized data. Fed-ComBat extends existing centralized linear methods, such as ComBat and distributed as d-ComBat, and nonlinear approaches like ComBat-GAM in accounting for potentially nonlinear and multivariate covariate effects. By doing so, Fed-ComBat enables the preservation of nonlinear covariate effects without requiring centralization of data and without prior knowledge of which variables should be considered nonlinear or their interactions, differentiating it from ComBat-GAM. We assessed Fed-ComBat and existing approaches on simulated data and multiple cohorts comprising healthy controls (CN) and subjects with various disorders such as Parkinson's disease (PD), Alzheimer's disease (AD), and autism spectrum disorder (ASD). The results of our study show that Fed-ComBat performs better than centralized ComBat when dealing with nonlinear effects and is on par with centralized methods like ComBat-GAM. Through experiments using synthetic data, Fed-ComBat demonstrates a superior ability to reconstruct the target unbiased function, achieving a 35% improvement (RMSE=0.5952) compared to d-ComBat (RMSE=0.9162) and a 12% improvement compared to our proposal to federate ComBat-GAM, d-ComBat-GAM (RMSE=0.6751). Additionally, Fed-ComBat achieves comparable results to centralized methods like ComBat-GAM for MRI-derived phenotypes without requiring prior knowledge of potential nonlinearities.

This study aimed to develop a BI-RADS network (DL-UM) via integrating ultrasound (US) and mammography (MG) images and explore its performance in improving breast lesion diagnosis and management when collaborating with radiologists, particularly in cases with discordant US and MG Breast Imaging Reporting and Data System (BI-RADS) classifications. We retrospectively collected image data from 1283 women with breast lesions who underwent both US and MG within one month at two medical centres and categorised them into concordant and discordant BI-RADS classification subgroups. We developed a DL-UM network via integrating US and MG images, and DL networks using US (DL-U) or MG (DL-M) alone, respectively. The performance of DL-UM network for breast lesion diagnosis was evaluated using ROC curves and compared to DL-U and DL-M networks in the external testing dataset. The diagnostic performance of radiologists with different levels of experience under the assistance of DL-UM network was also evaluated. In the external testing dataset, DL-UM outperformed DL-M in sensitivity (0.962 vs. 0.833, P = 0.016) and DL-U in specificity (0.667 vs. 0.526, P = 0.030), respectively. In the discordant BI-RADS classification subgroup, DL-UM achieved an AUC of 0.910. The diagnostic performance of four radiologists improved when collaborating with the DL-UM network, with AUCs increased from 0.674-0.772 to 0.889-0.910, specificities from 52.1%-75.0 to 81.3-87.5% and reducing unnecessary biopsies by 16.1%-24.6%, particularly for junior radiologists. Meanwhile, DL-UM outputs and heatmaps enhanced radiologists' trust and improved interobserver agreement between US and MG, with weighted kappa increased from 0.048 to 0.713 (P < 0.05). The DL-UM network, integrating complementary US and MG features, assisted radiologists in improving breast lesion diagnosis and management, potentially reducing unnecessary biopsies.

The application of artificial intelligence (AI) technology has realized the transformation of people's production and lifestyle, and also promoted the rapid development of the medical field. At present, the application of intelligence in the medical field is increasing. Using its advanced methods and technologies of AI, this paper aims to realize the integration of medical imaging-aided diagnosis system and AI, which is helpful to analyze and solve the loopholes and errors of traditional artificial diagnosis in the diagnosis of pulmonary nodules. Drawing on the principles and rules of image segmentation methods, the construction and optimization of a medical image-aided diagnosis system is carried out to realize the precision of the diagnosis system in the diagnosis of pulmonary nodules. In the diagnosis of pulmonary nodules carried out by traditional artificial and medical imaging-assisted diagnosis systems, 231 nodules with pathology or no change in follow-up for more than two years were also tested in 200 cases. The results showed that the AI software detected a total of 881 true nodules with a sensitivity of 99.10% (881/889). The radiologists detected 385 true nodules with a sensitivity of 43.31% (385/889). The sensitivity of AI software in detecting non-calcified nodules was significantly higher than that of radiologists (99.01% vs 43.30%, P < 0.001), and the difference was statistically significant.

Deep learning, particularly convolutional neural networks (CNNs), has proven valuable for brain tumor classification, aiding diagnostic and therapeutic decisions in medical imaging. Despite their accuracy, these models are vulnerable to adversarial attacks, compromising their reliability in clinical settings. In this research, we utilized a VGG16-based CNN model to classify brain tumors, achieving 96% accuracy on clean magnetic resonance imaging (MRI) data. To assess robustness, we exposed the model to Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks, which reduced accuracy to 32% and 13%, respectively. We then applied a multi-layered defense strategy, including adversarial training with FGSM and PGD examples and feature squeezing techniques such as bit-depth reduction and Gaussian blurring. This approach improved model resilience, achieving 54% accuracy on FGSM and 47% on PGD adversarial examples. Our results highlight the importance of proactive defense strategies for maintaining the reliability of AI in medical imaging under adversarial conditions.