Meta-Radiology最新文献

Recent years have shown great merits in utilizing neuroimaging data to understand brain structural and functional changes, as well as its relationship to different neurodegenerative diseases and other clinical phenotypes. Brain networks, derived from different neuroimaging modalities, have attracted increasing attention due to their potential to gain system-level insights to characterize brain dynamics and abnormalities in neurological conditions. Traditional methods aim to pre-define multiple topological features of brain networks and relate these features to different clinical measures or demographical variables. With the enormous successes in deep learning techniques, graph learning methods have played significant roles in brain network analysis. In this survey, we first provide a brief overview of neuroimaging-derived brain networks. Then, we focus on presenting a comprehensive overview of both traditional methods and state-of-the-art deep-learning methods for brain network mining. Major models, and objectives of these methods are reviewed within this paper. Finally, we discuss several promising research directions in this field.

Background and purpose

Computed tomography (CT) is highly sensitive to lung-related abnormalities as a non-invasive method and has become an essential tool for screening and diagnosing Coronavirus disease 2019 (COVID-19). To reduce the stress of work for physicians and speed up diagnosis, we propose a novel automatic diagnosis pipeline for COVID-19 based on high-dimensional radiomic features extracted from multimodal CT scans (multi-geometric and multiscale).

Materials and methods

There are 746 CT scans involved in this study, where 349 CT scans are COVID-19 positive and 397 CT scans are COVID-19 negative. All of them are from the public dataset. We first construct a transfer learning-based auto-segmentation model with a morphological post-processing block to improve the lung region segmentation. Then the radiomics feature extraction is guided by the proposed multi-modal CT scans strategy. In addition, our automatic diagnosis pipeline is driven by a well-designed loss function. We also explain the diagnosis capability from the related theory of linear subspace spanned by multi-modal radiomics features.

Results

Under the 10-fold cross-validation strategy, our approach can achieve an improvement in diagnostic performance of 5. 77%, 7. 78%, 7. 74%, 7. 78%, 7. 45% compared to the radiomic features extracted from the original CT scans, and diagnosis performance is promoted to 91.53%, 86.46%, 86.47%, 86.46%, 86.95% in terms of AUC, Acc, F1, Recall and Precision in public datasets.

Conclusions

We demonstrate a statistically significant improvement of the proposed statistical learning method compared to the state-of-the-art machine learning-based diagnosis approaches. Thanks to theoretical support and excellent diagnostic performance, our method can be deployed in clinical auxiliary diagnosis, releasing the overstretched medical resources.

Large Language Models (LLMs) especially when combined with Generative Pre-trained Transformers (GPT) represent a groundbreaking in natural language processing. In particular, ChatGPT, a state-of-the-art conversational language model with a user-friendly interface, has garnered substantial attention owing to its remarkable capability for generating human-like responses across a variety of conversational scenarios. This survey offers an overview of ChatGPT, delving into its inception, evolution, and key technology. We summarize the fundamental principles that underpin ChatGPT, encompassing its introduction in conjunction with GPT and LLMs. We also highlight the specific characteristics of GPT models with details of their impressive language understanding and generation capabilities. We then summarize applications of ChatGPT in a few representative domains. In parallel to the many advantages that ChatGPT can provide, we discuss the limitations and challenges along with potential mitigation strategies. Despite various controversial arguments and ethical concerns, ChatGPT has drawn significant attention from research industries and academia in a very short period. The survey concludes with an envision of promising avenues for future research in the field of ChatGPT. It is worth noting that knowing and addressing the challenges faced by ChatGPT will mount the way for more reliable and trustworthy conversational agents in the years to come.

Objectives

The purpose of this study is to investigate the diagnostic performance of the peritumoral brain zone (PBZ) in differentiating glioma grades. This is accomplished by comparing the relative standardized uptake values (rSUV) and relative cerebral blood flow (rCBF) obtained from hybrid ¹⁸F-fluoro-2-deoxy-d-glucose positron emission tomography/magnetic resonance imaging (¹⁸F-FDG PET/MRI) within different regions of interest, including the solid portion (SP) and the PBZ.

Methods

Twenty-four patients with gliomas who underwent preoperative ¹⁸F-PET/MRI were enrolled in this study. The maximum standardized uptake values (SUV_max) and relative maximum cerebral blood flow (rCBF_max) were obtained from the FDG-PET and ASL data, respectively. The relative SUV_max (rSUV_max) was calculated by standardizing against the contralateral normal-appearing brain cortex. Data from the solid portion (SP) of tumor and the peritumoral brain zone (PBZ) at distance of 5 ​mm, 10 ​mm, 15 ​mm, and 20 ​mm from the SP margin were recorded. Logistic regression was used to generate receiver-operating characteristic (ROC) curves. The areas under the ROC curves (AUCs) were calculated and compared to analyze the diagnostic utility of each parameter.

Results

In comparison to low-grade glioma (LGG), high-grade glioma (HGG) exhibited significantly higher rSUV_max and rCBF_max values in both the SP and the proximal PBZ (P ​< ​0.05). Among the various distance parameters and their combinations, the single parameter rSUV_max-SP demonstrated the highest diagnostic efficacy with an AUC of 0.788 (P ​< ​0.05). However, the AUC of rSUV_max-SP did not show a significantly improvement when combined with PBZs (P ​> ​0.05). When combining PBZs and SP with rSUV_max and rCBF_max, the rSUV_max and rCBF_max values of SP to PBZ 20 ​mm exhibited superior performance compared to single parameters and smaller regions of interest, with an AUC of 0.848. The sensitivity and specificity were determined as 73.8% and 83.6%, respectively.

Conclusion

The combination of rSUV_max and rCBF_max in the SP and PBZ, based on hybrid PET/MRI, proves to be superior to using parameters solely in the SP when it comes to differentiating between HGG and LGG. Expanding the study appropriately and incorporating the use of multiple parameters can offer more valuable diagnostic information, which holds potential for clinical applications.

This paper presents a comprehensive survey of ChatGPT-related (GPT-3.5 and GPT-4) research, state-of-the-art large language models (LLM) from the GPT series, and their prospective applications across diverse domains. Indeed, key innovations such as large-scale pre-training that captures knowledge across the entire world wide web, instruction fine-tuning and Reinforcement Learning from Human Feedback (RLHF) have played significant roles in enhancing LLMs' adaptability and performance. We performed an in-depth analysis of 194 relevant papers on arXiv, encompassing trend analysis, word cloud representation, and distribution analysis across various application domains. The findings reveal a significant and increasing interest in ChatGPT-related research, predominantly centered on direct natural language processing applications, while also demonstrating considerable potential in areas ranging from education and history to mathematics, medicine, and physics. This study endeavors to furnish insights into ChatGPT's capabilities, potential implications, ethical concerns, and offer direction for future advancements in this field.

Introduction

Functional connectivity in the brain is often studied with blood oxygenation level dependent (BOLD) resting state functional magnetic resonance imaging (rsfMRI), but the BOLD signal is several steps removed from neuronal activity. Arterial spin labeling (ASL), particularly pulsed ASL (PASL), has also the capacity to measure the blood-flow changes in response to activity. In this paper, we investigated the feasibility of extracting major brain networks from PASL data, in contrast with rsfMRI analsyis.

Materials and methods

In this retrospective study, we analyzed a cohort dataset that consists of 21 mild traumatic brain injury (mTBI) patients and 29 healthy controls, which was collected in a previous study. By extracting 10 major brain networks from the data of both PASL and rsfMRI, we contrasted their similarities and differences in the 10 networks extracted from both modalities.

Results

Our data demonstrated that PASL could be used to extract all 10 major brain networks. Eight out of 10 networks demonstrated over 60 ​% similarity to rsfMRI data. Meanwhile, there are similar but not identical changes in networks detected between mTBI patients and healthy controls with both modalities. Notably, the PASL-extracted default mode network (DMN), other than the rsfMRI-extracted DMN, includes some regions known to be associated with the DMN in other studies. It demonstrated that PASL data can be analyzed to identify resting state networks with reasonable reliability, even without rsfMRI data.

Conclusion

Our analysis provides an opportunity to extract functional connectivity information in heritage datasets in which ASL but not BOLD was collected.

Purpose

The difficulty of recanalization for intracranial atherosclerosis–related large vessel occlusion (ICAS-LVO) is closely related to the severity of stenosis. This study sought to investigate the characteristics of enhancement based on high-resolution magnetic resonance imaging (HR-MRI) so as to judge the severity of stenosis.

Methods

Sixty-two patients with symptomatic ICAS-LVOs who underwent endovascular treatment were prospectively recruited for HR-MRI (33 patients with severe stenosis and 29 without). The diagnostic agreements in locating occlusion segments were assessed between HR-MRI and angiographic assessment. The severity of atherosclerotic stenosis was evaluated by enhancement grade and quantitative enhancement index. Univariate and multivariate analyses were used to identify the parameters associated with the severity of stenosis.

Results

HR-MRI showed good agreement with angiographic assessments for evaluating the occlusion site (κ ​= ​0.97) and length (concordance correlation coefficient ​= ​0.70). Compared with patients without severe stenosis, patients with severe stenosis exhibited higher enhancement index (0.69 versus 0.19; p ​< ​0.001) of occlusion segments. In multivariate analysis, the enhancement index was an independent factor associated with the severity of stenosis (OR ​= ​2.92; 95% CI, 1.60–5.34, p ​< ​0.001). The enhancement index had an AUC of 0.89, with a sensitivity of 76.0% and a specificity of 86.0%. The model fit improved when including the enhancement index (AUC ​= ​0.93 versus 0.72). All of patients with severe stenosis required additional rescue treatments, which have a longer procedural time (104.0 versus 91.0 ​min; p ​= ​0.002).

Conclusion

Higher enhancement index of occlusion segments was associated with the severe atherosclerotic stenosis.

The growing armamentarium of artificial intelligence (AI) tools cleared by the United States Food and Drug Administration mostly target a narrow, single imaging modality or data source of information. While imaging technologies continue evolving rapidly, it is recognized that multimodal data provides synergistic information and enables better performance than what is achievable when these modalities are used separately. Deep learning approaches can integrate multimodal data, including not only imaging but also non-imaging modalities such as electronic medical records (EMRs) and genetic profiles. Such convergence advances clinical applications and research for improved effectiveness, especially the prediction of disease risks. This new avenue could address concerns over justification of imaging scans, clinical context-based interpretation of examinations, effectiveness of single modal and multimodal data to influence clinical decision making, as well as prediction of personalized disease risk. In this new era of radiology AI, the paradigm is being shifted from imaging alone AI analytics to multimodal artificial general intelligence (AGI). The heterogeneity of the data and the non-intuitive nature of certain modalities pose major challenges for developing multimodal large AI models and at the same time bring enormous opportunities.

The use of AI systems in healthcare for the early screening of diseases is of great clinical importance. Deep learning has shown great promise in medical imaging, but the reliability and trustworthiness of AI systems limit their deployment in real clinical scenes, where patient safety is at stake. Uncertainty estimation plays a pivotal role in producing a confidence evaluation along with the prediction of the deep model. This is particularly important in medical imaging, where the uncertainty in the model's predictions can be used to identify areas of concern or to provide additional information to the clinician. In this paper, we review the various types of uncertainty in deep learning, including aleatoric uncertainty and epistemic uncertainty. We further discuss how they can be estimated in medical imaging. More importantly, we review recent advances in deep learning models that incorporate uncertainty estimation in medical imaging. Finally, we discuss the challenges and future directions in uncertainty estimation in deep learning for medical imaging. We hope this review will ignite further interest in the community and provide researchers with an up-to-date reference regarding applications of uncertainty estimation models in medical imaging.