Search

A study evaluated the diagnostic performance of eleven open-source large language models (LLMs) using 1,933 case reports from the Eurorad library. The LLMs provided differential diagnoses based on clinical history and imaging findings, with Llama-3-70B showing superior performance. The findings suggest that open-source LLMs have the potential to serve as decision support tools for radiological differential diagnosis in challenging cases. [Extracted from the article]

The article explores the association between cardiovascular risk factors and cerebral blood flow in a multi-ethnic population during mid-life. The study included participants of Dutch, Moroccan, and South-Asian Surinamese descent, examining cerebral blood flow and its spatial coefficient of variation in grey and white matter. The findings suggest differences in cerebral hemodynamics based on ethnicity and highlight the importance of investigating the longitudinal effects of midlife risk factors on cerebrovascular health, emphasizing the need for further research in this area. [Extracted from the article]

A preprint abstract from medrxiv.org discusses a study that examines the association between amyloid and tau pathology and subjective cognitive complaints in cognitively normal individuals. The study found that higher levels of tau pathology were associated with greater subjective cognitive complaints, even in the absence of marked differences in neuropsychological assessments of memory. These findings suggest that subjective cognitive complaints may serve as early indicators of accumulating cerebral tau pathology. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

A preprint abstract discusses the feasibility of using deuterium metabolic imaging (DMI) to study the metabolism of the abdomen using a routine MRI system. The study aimed to optimize the radiofrequency coil setup, the dose of deuterium-labeled glucose, and the data processing pipeline for quantitative characterization of DMI signals in the kidney and liver. Five healthy volunteers were imaged, and the results showed that DMI of the human abdomen is possible using a routine MRI system. The study suggests that the 2H-glucose dose can be reduced to minimize gastric signal without affecting the reliability of organ quantification. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

A recent study has developed and evaluated a 16-channel MRI coil array made from flexible copper foil sheets. The coils were manufactured using a die cut process and showed superior signal-to-noise ratio, lower noise correlation, and better parallel imaging performance compared to a commercially available flexible coil. Phantom testing demonstrated a 17-20% improvement in signal-to-noise ratio, while in vivo testing confirmed higher signal-to-noise ratio and enhanced image quality. The use of copper foil sheets allowed for rapid production of the coils without compromising quality, making them a cost-effective solution for improving the accessibility, adaptability, and performance of MRI technology. [Extracted from the article]

A recent study evaluated the diagnostic performance of open-source large language models (LLMs) in challenging radiological cases. The study used clinical and imaging descriptions from 4,049 case reports in the Eurorad library and assessed the accuracy of LLMs in providing differential diagnoses. The results showed that Llama-3-70B had the highest accuracy (75.1%), followed by Gemma-2-27B (63.9%) and Mixtral-8x-7B (61.5%). The study suggests that open-source LLMs have the potential to be effective decision support tools for radiological differential diagnosis in real-world cases. However, it is important to note that this study has not been peer-reviewed. [Extracted from the article]

This article discusses a novel method for assessing the volumetric pulsatility of cerebral microvasculature using high-resolution 4D CBV MRI at 7T. The study investigates the effect of aging on microvascular pulsatility and compares it to large arterial pulsatility. The researchers found that microvascular pulsatility was highest in the cerebrospinal fluid on the cortical surface and decreased towards the cortical layers and larger arteries. In the deep white matter, there was a significant increase in microvascular pulsatility in older participants compared to younger ones. This method provides the first in vivo measurement of microvascular volumetric pulsatility in the human brain, which has implications for cerebral microvascular health and its relationship with aging and neurodegenerative diseases. [Extracted from the article]

A study conducted in Uganda aimed to assess the impact of continuous medical education (CME) and clinical imaging guidelines (CIGs) on the appropriateness of computerized tomography (CT) utilization among children and young adults in resource-limited settings. The study found that after the intervention, there was an increase in the total number of CT examinations performed, particularly in public and private-for-profit hospitals. However, the proportion of inappropriate CT examinations also increased, highlighting the complexity of achieving consistent improvements across all examination types. The study suggests further research to explore the successful implementation of CIGs in resource-limited settings to optimize CT utilization and improve patient outcomes. [Extracted from the article]

A preprint abstract from medrxiv.org discusses a study on cross-modal functional plasticity after cochlear implantation. The study aimed to evaluate the plasticity of cross-modal functional connectivity (CMFC) in cochlear implant recipients. The experiment involved 22 post-lingually deafened adults with cochlear implants and 17 normal-hearing subjects. The results showed a negative correlation between CMFC and speech outcomes, indicating increased reliance on cross-modal processing in cochlear implant recipients. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

The article explores the potential of radiomic biomarkers to predict lesion-specific responses to systemic treatment in metastatic leiomyosarcoma patients. Topics discussed include the development of a radiomic model based on computed tomography features, the model's predictive accuracy compared to traditional methods, and the influence of tumor volume and drug type on prediction precision.

The article examines the effectiveness of pretrained machine learning models in interpreting various medical images, highlighting their potential to enhance diagnostic accuracy despite challenges in generalizability. Topics discussed include the evaluation of different convolutional neural networks (CNNs), their performance across multiple imaging modalities, and future directions for improving model generalizability and integration with patient data.

The article explores the effectiveness of arterial spin labeling (ASL) as a potential substitute for FDG-PET in diagnosing disorders of consciousness (DOC). Topics discussed include the comparative analysis of ASL and FDG-PET modalities, the use of SVM classifiers for distinguishing DOC patients from healthy controls, and the advantages of ASL in clinical settings.

The article focuses on the development of PixelPrint4D, a 3D printing technique for creating patient-specific deformable lung phantoms used in respiratory motion applications. Topics include the limitations of current respiratory motion phantoms, the capabilities of PixelPrint4D to replicate realistic lung structures and motions, and its potential to enhance testing for motion compensation technologies in CT imaging.

The article focuses on evaluating GPT-4's effectiveness in suggesting optimal brain MRI sequences for different medical conditions. Topics include the comparison of GPT-4's suggestions with established hospital protocols, the model's performance in English versus Japanese, and the need for further research to refine GPT-4's application in radiology.

The article discusses research by a team using cineCT, reported in the August 19, 2024, issue, on evaluating right ventricular myocardial work in patients with complex heart conditions. Topics include the use of cineCT for 3D regional assessment of right ventricular function, differences in myocardial work and strain among patients with end-stage heart failure, chronic thromboembolic pulmonary hypertension and the correlation between myocardial work and right ventricular ejection fraction.

The article discusses research by NewsRx correspondents and reported in the August 19, 2024 issue, on AI-based volumetric six-tissue body composition quantification from CT cardiac attenuation scans to enhance mortality prediction. Topics include the development of an AI-based method for measuring chest body composition from CTAC scans, its effectiveness in predicting all-cause mortality risk, and the clinical applicability of these measurements for improving risk stratification.

According to a preprint abstract, researchers have identified a topography in EEG microstates known as vertical topography (VT) that appears as a straight line separating positive and negative values from the nasion to the inion. The researchers suggest that VT may be partially caused by movements of the EEG cap and its metallic components, based on their analysis of human and phantom data. They caution that the presence of VT may affect the parameters of other microstates and distort the shapes of other topographies. This research has not yet undergone peer review. [Extracted from the article]

This article discusses a study that aims to develop a generalized signal model for dual-module velocity-selective arterial spin labeling (dm-VSASL) in medical imaging. The model is designed to integrate arbitrary saturation and inversion profiles and address the complexity of dm-VSASL. The study presents expressions for magnetization difference, arterial delivery functions, labeling efficiency, and cerebral blood flow (CBF) estimation error. The results show that a CBF-dependent estimation error can occur when using velocity-selective inversion (VSI) and velocity-selective saturation (VSS), but this error can be minimized by choosing a lower nominal labeling cutoff velocity. The proposed signal model will help researchers accurately assess and compare the performance of dm-VSASL implementations and improve the quantification of CBF measures. [Extracted from the article]

A recent preprint abstract discusses the development of a novel non-invasive positron emission tomography (PET) method to measure the permeability of the blood-brain barrier (BBB) in humans. The method uses dynamic imaging and kinetic modeling to estimate cerebral blood flow and tracer-specific BBB transport rate, allowing for the measurement of the molecular permeability-surface area (PS) product of the radiotracer. The study demonstrates the ability of the method to resolve BBB PS across different radiotracers, measure reductions in BBB PS in healthy aging, and suggest a potential association between decreased BBB PS and metabolic dysfunction-associated steatotic liver inflammation. This research provides new directions for studying the molecular permeability of the human BBB using available molecular PET tracers. [Extracted from the article]

A systematic review of studies exploring the use of natural language processing (NLP) in magnetic resonance imaging (MRI) has found that NLP techniques have the potential to enhance the utility of MRI. The review analyzed 26 studies published between April 2010 and January 2024, covering over 160,000 MRI reports. NLP applications in MRI included staging, quantification, and disease diagnosis, with neurology being the most frequently studied specialty. The review suggests that further prospective studies are needed to validate NLP algorithms in real-time clinical scenarios and across various radiology specialties. [Extracted from the article]

A preprint abstract from medrxiv.org discusses a deep learning (DL) approach to estimating skull porosity from T1-weighted MRI images for transcranial focused ultrasound (tFUS) planning. The traditional method of using CT scans for porosity estimates involves radiation, so the DL approach offers a radiation-free alternative. The study evaluates different DL approaches and finds that the best performing model achieves a mean absolute error of 6.9% in the test set, improving upon previous methods. The DL approach is robust to image noise and MRI acquisition parameters and yields near-optimal treatment planning and dose estimates for brain targets in the 200-1000 kHz frequency range. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

A preprint abstract from medrxiv.org discusses the use of diffusion magnetic resonance imaging (dMRI) to characterize microvasculature in cancer. The study introduces SpinFlowSim, an open-source simulator that generates realistic dMRI signals based on histology-reconstructed networks. The simulator was used to generate synthetic signals for liver biopsies containing cancerous and non-cancerous tissue, revealing complex flow patterns within the networks. The study suggests that realistic flow simulations like SpinFlowSim could contribute to the development of advanced dMRI methods for microvascular mapping in oncology. However, it is important to note that this preprint has not yet undergone peer review. [Extracted from the article]

A recent study evaluated the impact of a 10-minute tutorial on the use of a large language model (LLM)-based search engine on radiology residents' performance in LLM-assisted brain MRI differential diagnosis. The study found that radiology residents achieved the highest accuracy when using the LLM-based search engine following the tutorial, correctly diagnosing 62.5% of cases. The tutorial led to significantly higher performance and confidence levels, but did not result in any relevant differences in reading times. The study suggests that a structured LLM tutorial can improve performance and confidence in LLM-assisted brain MRI diagnosis among radiology residents. [Extracted from the article]

A quantitative review conducted by researchers aimed to assess the methodologies for combining radiomic features in the multi-metastatic setting. The study compared ten mathematical methods across three distinct datasets, encompassing a total of 16,850 lesions in 3,930 patients. The results showed variable performance among the methods, with no single method consistently outperforming others across all datasets. The study provides valuable insights into effective radiomic analysis strategies in multi-metastatic patients, filling a critical gap in radiomics research. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

This article discusses a study on quantitative MRI imaging in children and adults using a specific imaging sequence called MP2RAGE. The researchers aimed to optimize the sequence for producing T1-weighted images with uniform quality while also generating contrasts for fluid and white matter suppression (FLAWS). They developed an approach to calculate effective proton density (PD) and quantitative T1 relaxation maps using these images. The study found that the T1 values varied over development into adulthood, particularly in deep grey matter regions, while the effective PD maps did not show a significant difference between children and adults. The researchers concluded that their protocol allowed for high-resolution imaging in both children and adults within a clinically feasible duration. [Extracted from the article]

A preprint abstract from medrxiv.org discusses the development of a dentate nucleus (DN) segmentation tool using deep learning (DL) applied to brain quantitative susceptibility mapping (QSM) images. The study collected brain QSM images from healthy controls and individuals with cerebellar ataxia or multiple sclerosis. The researchers implemented a two-step approach, consisting of a localization model followed by DN segmentation, and achieved promising results. The model demonstrated generalizability across different datasets and can be used for biomarker discovery in observational, natural history, and treatment trials. However, it is important to note that this preprint has not yet undergone peer review. [Extracted from the article]

A study investigated the effectiveness of ComBat, a data-driven harmonization model, in reducing biases in regional amyloid PET measurements. The study used 113 head-to-head scan pairs from the Open Access Series of Imaging Studies 3 (OASIS-3) dataset and simulated longitudinal amyloid measurements from an anti-amyloid drug trial. The results showed that ComBat outperformed the Centiloid scale in harmonizing both global and regional amyloid estimates. The findings suggest that ComBat is a viable alternative to Centiloid for harmonizing regional amyloid PET analyses and improving the detection of differences between clinical trial groups. [Extracted from the article]

A recent study published in Medical Imaging Week discusses the development of an artificial intelligence (AI) system for detecting abnormalities in abdominal computed tomography (CT) scans. The researchers used free-text radiology reports instead of manual annotations to train a deep-learning-based pipeline. The system was able to accurately detect abnormalities in the liver, gallbladder, pancreas, spleen, and kidneys. The study found that using disease information from routine clinical free-text radiology reports can lead to the development of accurate anomaly detection models without the need for manual annotations. This approach has the potential to streamline diagnostic processes. The study was funded by the Japan Science and Technology Agency. [Extracted from the article]

This article discusses a study that aimed to differentiate between benign and malignant diseases in patients with solitary cavitary lesions detected on CT scans. The study found that non-malignant lesions accounted for almost half of the cases, with active pulmonary tuberculosis being the most common diagnosis. In the malignant group, primary lung cancer was the most prevalent subtype. The study identified significant differences in lesion size and cavity wall thickness between malignant and non-malignant cases, as well as the presence of perilesional consolidation or centrilobular nodules favoring non-malignant diagnoses. The study concludes that CT findings can provide valuable information in distinguishing between malignant and non-malignant solitary lung cavities. [Extracted from the article]

A proposed survey for patients prior to mammography is being explored as a tool to improve radiological reporting for breast cancer detection. The survey would gather important information about patient history and risk factors, which would aid radiologists in preparing reports. The survey includes questions about family history, gynecological-obstetric history, surgical history, and other relevant factors. This research is based on a semi-structured survey completed by 31 radiologists specializing in mammography. It is important to note that this preprint has not yet undergone peer review. [Extracted from the article]

A study was conducted to assess the performance of OpenAI's multimodal GPT-4 in interpreting radiological images. The study analyzed 230 emergency room diagnostic images using GPT-4V and compared its interpretations with those of senior radiologists. The results showed that GPT-4V accurately identified the imaging modality in all cases, the anatomical region in 87.1% of cases, and the pathology in 35.2% of cases. However, the model's performance varied across different modalities, indicating inconsistencies in its ability to interpret radiological images accurately. The study concludes that while AI integration in radiology offers promising avenues for diagnostic enhancement, the current capabilities of GPT-4V are not yet reliable for interpreting radiological images. [Extracted from the article]

This article discusses a study that aimed to address the challenge of inconsistent image conditions in Computer Tomography (CT) imaging equipment. The researchers developed three re-slice systems utilizing U-nets to generate intermediate slice images with the same width from existing CT images in the craniofacial area. The study utilized CT images from the Japanese Facial Bone Fracture CT Collection Project and compared the generated images to the original images using various calculations. The results showed that the developed re-slice systems with U-net models were effective in creating intermediate slice images. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

A preprint abstract from medrxiv.org discusses the use of multimodal surface coils for low field MRI imaging. The abstract explains that low field MRI is safer and more cost-effective than high field MRI, but it has a lower signal-to-noise ratio. The researchers propose a multimodal surface coil technique to improve the RF magnetic field efficiency and potentially enhance MR sensitivity. The prototype coils are built and tested, showing better performance than conventional surface coils. The abstract concludes that the proposed multimodal surface coil offers excellent low-frequency operation capability and reduced capacitance value compared to conventional surface coils. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

This article discusses the use of deep learning models, specifically DenseNet and EfficientNet, for the detection of aortic elongation on chest X-ray images. Aortic elongation can be caused by various factors and is associated with cardiovascular diseases. The study found that DenseNet achieved higher accuracy, precision, sensitivity, specificity, F1-score, and AUROC compared to EfficientNet. The researchers also employed explainability methods to gain insights into the decision-making process of the models and enhance their interpretability. This approach shows promise in aiding clinicians for timely and accurate diagnosis, potentially improving patient outcomes. However, it is important to note that this preprint has not been peer-reviewed. [Extracted from the article]

This article discusses a study that used magnetic resonance imaging (MRI) to assess the fatty acid composition of abdominal subcutaneous and visceral adipose tissue in the UK Biobank imaging cohort. The study found that there were differences in fatty acid composition between the two types of adipose tissue, with higher levels of saturated and polyunsaturated fatty acids and lower levels of monounsaturated fatty acids in visceral adipose tissue. The study also examined the relationship between adipose tissue fatty acid composition and various traits related to disease and body size. The findings highlight the importance of studying adipose tissue beyond its quantity. Please note that this article is a preprint and has not been peer-reviewed. [Extracted from the article]