Ultra-wideband （UWB） microwave imaging is a promising method for breast cancer detection based on the large contrast of electric parameters between the malignant tumor and its surrounded normal breast organisms. In ...Ultra-wideband （UWB） microwave imaging is a promising method for breast cancer detection based on the large contrast of electric parameters between the malignant tumor and its surrounded normal breast organisms. In the case of multiple tumors being present, the conventional imaging approaches may be ineffective to detect all the tumors clearly. In this paper, a progressive processing method is proposed for detecting more than one tumor. The method is divided into three stages： primary detection, refocusing and image optimization. To test the feasibility of the approach, a numerical breast model is developed based on the realistic magnetic resonance image （MRI）. Two tumors are assumed embedded in different positions. Successful detection of a 3.6 mm-diameter tumor at a depth of 42 mm is achieved. The correct information of both tumors is shown in the reconstructed image, suggesting that the progressive processing method is promising for multi-tumor detection.展开更多
目的:探讨影响大型语言模型(large language models,LLMs)生成乳腺MRI报告性能的关键因素,并确定一种成本效益高的部署策略,以提升其在规模化影像诊断中的应用。方法:本研究回顾性整合三家医疗机构的中文放射报告,构建数据集。通过微调...目的:探讨影响大型语言模型(large language models,LLMs)生成乳腺MRI报告性能的关键因素,并确定一种成本效益高的部署策略,以提升其在规模化影像诊断中的应用。方法:本研究回顾性整合三家医疗机构的中文放射报告,构建数据集。通过微调多种LLMs(包括ChatGPT、Llama3、Qwen2.5、DeepSeek-R1-Distill-Llama3_8B),评估模型架构、参数规模及预训练数据对性能的影响。采用BLEU、ROUGE、Cosine Similarity及BERTScore评价生成诊断的文本质量,并通过BI-RADS分类任务评估其诊断推理能力。结果:实验表明,增加训练数据可显著提升Llama3_8B模型的性能:BLEU值从1.69×10^(-3)升至0.78,ROUGE-L从0.05升至0.90,BERTScore从0.52升至0.94,Cosine Similarity从0.04升至0.88。不同架构、推理能力及参数规模的模型在诊断准确率上差异较小(66%~67%)。模型在外部验证集的性能较内部验证集下降明显(如Llama3_8B的BERTScore从0.94降至0.71,准确率从66%降至22%)。报告分析显示,模型报告在完整度(4.56 vs.4.46)和正确性(4.33 vs.4.15)上优于人工报告。结论:经微调的LLMs在乳腺MRI印象生成任务中表现出色。本研究为模型微调提供了实践指导,为医疗机构提供可定制化的低成本部署方案,从而提升诊断效率并减轻放射科医师工作负荷。展开更多
This work presents an efficient method for volume rendering of glioma tumors from segmented 2D MRI Datasets with user interactive control, by replacing manual segmentation required in the state of art methods. The mos...This work presents an efficient method for volume rendering of glioma tumors from segmented 2D MRI Datasets with user interactive control, by replacing manual segmentation required in the state of art methods. The most common primary brain tumors are gliomas, evolving from the cerebral supportive cells. For clinical follow-up, the evaluation of the preoperative tumor volume is essential. Tumor portions were automatically segmented from 2D MR images using morphological filtering techniques. These segmented tumor slices were propagated and modeled with the software package. The 3D modeled tumor consists of gray level values of the original image with exact tumor boundary. Axial slices of FLAIR and T2 weighted images were used for extracting tumors. Volumetric assessment of tumor volume with manual segmentation of its outlines is a time-consuming process and is prone to error. These defects are overcome in this method. Authors verified the performance of our method on several sets of MRI scans. The 3D modeling was also done using segmented 2D slices with the help of medical software package called 3D DOCTOR for verification purposes. The results were validated with the ground truth models by the Radiologist.展开更多
.Abstracting eye models from MRI images is critical in advancing medical imaging, particularly for clinical diagnostics. Current methods often struggle with accuracy and efficiency, highlighting a gap this research ai....Abstracting eye models from MRI images is critical in advancing medical imaging, particularly for clinical diagnostics. Current methods often struggle with accuracy and efficiency, highlighting a gap this research aims to fill. This study investigates the application of machine learning methods, focusing on the U-net-based deep learning framework, to improve the accuracy of eye model extraction. The objectives include fitting measured eye data to models such as the Ellipsoid model, evaluating automated segmentation tools, and assessing the usability of machine learning-based extractions in clinical scenarios. We employed point cloud data of 202,872 points to fit eye models using ellipsoid, non-linear, and spherical fitting techniques. The fitting processes were optimized to ensure precision and reliability. We compared the performance of these models using mean squared error (MSE) as the primary metric. The non-linear model emerged as the most accurate, with a significantly lower MSE (1.186562) compared to the ellipsoid (781.0542) and spherical models. This finding indicates that the non-linear model provides a more detailed and precise representation of the eye’s geometry. These results suggest that machine learning methods, particularly non-linear models, can significantly enhance the accuracy and usability of eye model extraction in clinical diagnostics, offering a robust framework for future advancements in medical imaging.展开更多
基金supported by the National Natural Science Foundation of China(Grant No.61271323)the Open Project from State Key Laboratory of MillimeterWaves,China(Grant No.K200913)
文摘Ultra-wideband （UWB） microwave imaging is a promising method for breast cancer detection based on the large contrast of electric parameters between the malignant tumor and its surrounded normal breast organisms. In the case of multiple tumors being present, the conventional imaging approaches may be ineffective to detect all the tumors clearly. In this paper, a progressive processing method is proposed for detecting more than one tumor. The method is divided into three stages： primary detection, refocusing and image optimization. To test the feasibility of the approach, a numerical breast model is developed based on the realistic magnetic resonance image （MRI）. Two tumors are assumed embedded in different positions. Successful detection of a 3.6 mm-diameter tumor at a depth of 42 mm is achieved. The correct information of both tumors is shown in the reconstructed image, suggesting that the progressive processing method is promising for multi-tumor detection.
文摘This work presents an efficient method for volume rendering of glioma tumors from segmented 2D MRI Datasets with user interactive control, by replacing manual segmentation required in the state of art methods. The most common primary brain tumors are gliomas, evolving from the cerebral supportive cells. For clinical follow-up, the evaluation of the preoperative tumor volume is essential. Tumor portions were automatically segmented from 2D MR images using morphological filtering techniques. These segmented tumor slices were propagated and modeled with the software package. The 3D modeled tumor consists of gray level values of the original image with exact tumor boundary. Axial slices of FLAIR and T2 weighted images were used for extracting tumors. Volumetric assessment of tumor volume with manual segmentation of its outlines is a time-consuming process and is prone to error. These defects are overcome in this method. Authors verified the performance of our method on several sets of MRI scans. The 3D modeling was also done using segmented 2D slices with the help of medical software package called 3D DOCTOR for verification purposes. The results were validated with the ground truth models by the Radiologist.
文摘.Abstracting eye models from MRI images is critical in advancing medical imaging, particularly for clinical diagnostics. Current methods often struggle with accuracy and efficiency, highlighting a gap this research aims to fill. This study investigates the application of machine learning methods, focusing on the U-net-based deep learning framework, to improve the accuracy of eye model extraction. The objectives include fitting measured eye data to models such as the Ellipsoid model, evaluating automated segmentation tools, and assessing the usability of machine learning-based extractions in clinical scenarios. We employed point cloud data of 202,872 points to fit eye models using ellipsoid, non-linear, and spherical fitting techniques. The fitting processes were optimized to ensure precision and reliability. We compared the performance of these models using mean squared error (MSE) as the primary metric. The non-linear model emerged as the most accurate, with a significantly lower MSE (1.186562) compared to the ellipsoid (781.0542) and spherical models. This finding indicates that the non-linear model provides a more detailed and precise representation of the eye’s geometry. These results suggest that machine learning methods, particularly non-linear models, can significantly enhance the accuracy and usability of eye model extraction in clinical diagnostics, offering a robust framework for future advancements in medical imaging.
