基于多维度特征融合的深度学习骨龄评估模型

目的 通过基于特征提取的深度卷积神经网络,结合关键区域特征和人口学信息,评估儿童骨龄。方法 自动识别左手X线图像数据,对图像进行预处理,使用基于深度神经网络的X线图像分析方法,实现左手关节骨龄17个关键区域特征的自动提取,再将骨龄影像特征与临床大数据（人口统计、性别）融合训练骨龄评估模型,测试模型的评估效能。结果 使用基于深度学习的特征区域提取方法比传统图像分析方法可以更好地提取特征信息,结合临床信息从另一维度补充了骨龄发育信息。基于多维度数据特征融合的骨龄评估模型检测得到的骨龄平均绝对误差为0.455,优于传统方法和仅端到端的深度学习方法。结论 相较传统的机器学习特征提取方法,基于特征提取的深度卷积神经网络在骨龄回归模型上有更好的表现,结合人口和性别信息可进一步提升基于图像的骨龄预测准确率。     

基于迁移学习的全连接神经网络舌象分类方法

目的 针对深度学习在舌象分类中训练数据量大、训练设备要求高、训练时间长等问题,提出一种基于迁移学习的全连接神经网络小样本舌象分类方法。方法 应用经ImageNet海量数据集训练后的卷积Inception_v3网络提取舌象点、线等有效特征,再使用全连接神经网络对特征进行训练分类,将深度学习网络学习到的图像知识迁移到舌象识别任务中。利用舌象数据集进行训练、测试。结果 与典型舌象分类方法K最近邻（KNN）算法、支持向量机（SVM）算法和卷积神经网络（CNN）深度学习方法相比,本实验使用的两种方法（Inception_v3+2NN和Inception_v3+3NN）具有较高的舌象分类识别率,准确率分别达90.30%和93.98%,且样本训练时间明显缩短。结论 与KNN算法、SVM算法和CNN深度学习方法相比,基于迁移学习的全连接神经网络舌象分类方法可有效提高舌象分类的准确率、缩短网络的训练时间。     

Medical Image Analysis using Convolutional Neural Networks: A Review

The science of solving clinical problems by analyzing images generated in clinical practice is known as medical image analysis. The aim is to extract information in an affective and efficient manner for improved clinical diagnosis. The recent advances in the field of biomedical engineering have made medical image analysis one of the top research and development area. One of the reasons for this advancement is the application of machine learning techniques for the analysis of medical images. Deep learning is successfully used as a tool for machine learning, where a neural network is capable of automatically learning features. This is in contrast to those methods where traditionally hand crafted features are used. The selection and calculation of these features is a challenging task. Among deep learning techniques, deep convolutional networks are actively used for the purpose of medical image analysis. This includes application areas such as segmentation, abnormality detection, disease classification, computer aided diagnosis and retrieval. In this study, a comprehensive review of the current state-of-the-art in medical image analysis using deep convolutional networks is presented. The challenges and potential of these techniques are also highlighted.     

基于U型全卷积神经网络的腹部动脉CT图像分割算法研究

目的 为了精确分割腹部动脉血管,提出一种基于深度学习的全自动腹部动脉CT图像分割算法.方法 采用区域不平衡块生成方法 提取CT血管横断面、冠状面和矢状面图像特征,接着采用U型全卷积神经网络对块特征进行训练与分割,最后采用最大体素保留法获得三维血管分割图像.选用120例患者腹部CT血管图像进行网络训练和分割实验,分割结果...     

基于睡眠脑电信号和深度学习的抑郁症识别研究

目的 基于睡眠脑电信号,探索深度学习Vision Transformer(ViT)结合Transformer网络对抑郁症患者识别的有效性。方法 首先对28例抑郁症患者和37例正常对照的睡眠脑电信号进行预处理,并将信号转为图像格式,保留其频域及空间域特征信息,之后将图像输送到ViT-Transformer编码网络,分别学习抑郁症患者和正常对照的快速眼动（rapid eye movement, REM）睡眠期和非快速眼动（non-rapid eye movement, NREM）睡眠期的脑电信号特征,并对抑郁症进行识别。结果 基于ViT-Transformer网络,从不同脑电频率角度,发现delta、theta和beta波的组合对抑郁症识别具有比较好的结果。其中,REM期delta-theta-beta波组合的脑电信号特征对抑郁症识别的准确率达92.8%,精准率为93.8%,抑郁症患者的召回率为84.7%,F0.5值为0.917±0.074;NREM期delta-theta-beta波组合的脑电信号特征对抑郁症的识别准确率为91.7%,精准率为90.8%,召回率为85.2%,F0.5值为0...     

基于迁移学习的胃镜图像自动识别多分类系统的研究

目的通过迁移学习提高早期胃癌图像识别准确率。方法根据胃癌前病变概念收集5类胃镜图像,分别为早期和进展期胃癌图像783张、胃溃疡图像1042张、慢性胃炎图像1143张、胃息肉图像1096张和正常胃镜图像1763张,按6:2:2的比例分为训练集、验证集、测试集,通过从零训练模型ResNet34与微调迁移模型ResNet34、VGG16相比较。结果基于迁移学习的ResNet34模型识别准确率最高,验证集准确率95.64%,测试集准确率90.75%。结论ResNet34模型可较准确的实现常见胃镜图像识别,较传统深度学习方法对小数据集的医学图像有更好的泛化和特征提取能力。     

基于AI技术的腰椎X射线图像质量控制模型的构建与应用

目的 采用深度学习算法建立腰椎X射线摄影图像质量控制模型,通过该模型实时和回顾性评估临床图像。方法选取2018年1月至2021年2月在温州医科大学附属第一医院接受检查的1389例患者,搜集患者的正位、侧位和斜位腰椎X射线摄影图像。采用基于U-Net的全卷积神经网络对腰椎X射线图像中的解剖结构进行分割,利用该分割算法建立一种自动评价模型检测不合格图像。采用Dice相似系数（Dice similarity coefficient,DSC）评价模型性能,并对模型投入应用后的腰椎X射线摄影图像进行统计评价。结果 模型在验证集上的准确性为0.971～0.990(0.98±0.10)、敏感度为0.714～0.933(0.86±0.13)、特异性为0.995～1.000(0.99±0.12)。质控模型在2022年腰椎X射线摄影的优秀率为28.8%,中等率54.8%,不合格率16.4%。结论 基于人工智能的腰椎X射线图像质控模型实现腰椎解剖结构的精准分割,可对图像质量作出准确评价,有利于保证技师对腰椎X射线摄影操作的规范性。     

基于深度学习的白内障识别与分级

目的 利用深度学习方法自动提取眼底白内障特征,构建白内障自动分类器,并可视化分析深度网络中间层特征的逐层变换过程。方法 基于临床眼底图像,使用深度卷积神经网络（CNN）从输入数据的原始表示直接学习有用的特征,对比分析CNN自动提取的特征与预定义特征的性能表现。然后利用反卷积神经网络（DN）量化分析CNN各个中间层的特征,进一步研究输入图像中对CNN的预测贡献最大的像素集,探究CNN表征白内障的具体过程。结果 使用深度学习方法构建的分类器在四分类任务中达到0.818 6的平均准确率。与现有的预定义特征集相比,利用深度CNN自动提取的特征集能提供更好的白内障特征表示。CNN中间层特征呈现从低级抽象到高级抽象的分层变换,如梯度变化到边缘,然后到边缘状发散结构的组合,最后到血管和视神经盘信息的高级抽象,这种变换过程与临床检测白内障的诊断标准相吻合。结论 基于深度学习的分类器在性能表现上优于现有分类器。该方法对检测其他眼病也可能具有潜在的应用前景。     

Correlation Filters for Detection of Cellular Nuclei in Histopathology Images

Nuclei detection in histology images is an essential part of computer aided diagnosis of cancers and tumors. It is a challenging task due to diverse and complicated structures of cells. In this work, we present an automated technique for detection of cellular nuclei in hematoxylin and eosin stained histopathology images. Our proposed approach is based on kernelized correlation filters. Correlation filters have been widely used in object detection and tracking applications but their strength has not been explored in the medical imaging domain up till now. Our experimental results show that the proposed scheme gives state of the art accuracy and can learn complex nuclear morphologies. Like deep learning approaches, the proposed filters do not require engineering of image features as they can operate directly on histopathology images without significant preprocessing. However, unlike deep learning methods, the large-margin correlation filters developed in this work are interpretable, computationally efficient and do not require specialized or expensive computing hardware. Availability: A cloud based webserver of the proposed method and its python implementation can be accessed at the following URL: http://faculty.pieas.edu.pk/fayyaz/software.html#corehist.     

基于非局部能谱相似特征的基物质分解方法用于双能CT图像去噪

目的 提出一种非局部能谱相似特征引导的双能CT基物质分解方法（NSSD-Net）以抑制低剂量能谱CT基物质图像的相关性噪声。方法 首先构建模型驱动的双能CT迭代分解模型,采用迭代软阈值算法（ISTA）优化分解模型目标函数的求解过程,并利用深度学习技术将此过程展开为迭代分解网络的形式。然后构建非局部能谱相似特征引导的代价函数,约束网络的训练过程。利用双能CT真实病人数据所建立的基物质分解数据集进行评估。将NSSD-Net与2种传统模型驱动的基物质分解方法、1种基于数据驱动的基物质分解方法以及1种基于数据-模型耦合驱动的监督分解方法进行对比实验。结果 与传统模型驱动的基物质分解方法以及数据驱动的基物质分解方法相比,NSSD-Net方法在水和骨基物质分解结果中均获得最高的PNSR指标（31.383和31.444）、最高的SSIM指标（0.970和0.963）以及最低的RMSE指标（2.901和1.633）;与数据-模型耦合驱动的监督分解方法相比,NSSD-Net方法在水和骨基物质分解结果中均获得最高的SSIM指标;临床影像专家的主观图像质量评估结果显示,NSSD-Net方法在水和骨基物质分解结果中图像质量评分均最高（8.625和8.250）,与其他4种对比方法分解性能之间的差异具有统计学意义（P<0.001）。结论 本方法可以获得高质量的基物质分解结果,有效避免训练数据质量问题和模型不可解释问题。     

An Automated Self-Learning Quantification System to Identify Visible Areas in Capsule Endoscopy Images

Visibility in capsule endoscopic images is presently evaluated through intermittent analysis of frames selected by a physician. It is thus subjective and not quantitative. A method to automatically quantify the visibility on capsule endoscopic images has not been reported. Generally, when designing automated image recognition programs, physicians must provide a training image; this process is called supervised learning. We aimed to develop a novel automated self-learning quantification system to identify visible areas on capsule endoscopic images. The technique was developed using 200 capsule endoscopic images retrospectively selected from each of three patients. The rate of detection of visible areas on capsule endoscopic images between a supervised learning program, using training images labeled by a physician, and our novel automated self-learning program, using unlabeled training images without intervention by a physician, was compared. The rate of detection of visible areas was equivalent for the supervised learning program and for our automatic self-learning program. The visible areas automatically identified by self-learning program correlated to the areas identified by an experienced physician. We developed a novel self-learning automated program to identify visible areas in capsule endoscopic images.     

骨原发性恶性淋巴瘤影像学诊断与组织病理对照研究

目的 探讨骨原发恶性淋巴瘤(PLB)的影像学表现特点.方法 回顾性分析经手术或穿刺活检病理学证实的PLB 9例,其中男6例,女3例,年龄9～60岁,中位年龄26.5岁.9例中X线平片检查8例、CT检查5例、MRI检查7例.其中2例行X线平片和MR检查,2例CT和MR检查,4例具有X线、CT和MR资料.2例为穿刺活检证实;7例行手术切除和病理学检查证实,全部病例均做了常规的组织切片HE染色和免疫组化检查.结果 病灶位于骨盆4例、额骨1例、枕骨斜坡1例、脊柱1例、股骨上端2例.影像学表现:X线表现,病变骨组织外形基本正常4例,内部可见斑点状、大小不等的虫蚀状骨质破坏;4例表现为病变骨质轻度～中度膨胀性改变,局部骨质呈明显溶骨性破坏:CT表现骨髓腔内和骨皮质上可见大小不等的溶骨性破坏,病变骨质周围围绕明显的软组织肿块;MR表现病变区骨髓腔内及周围软组织肿块在T2WI上呈不均匀中度～明显高信号,T1WI上呈均匀等信号.增强扫描后骨髓腔内病灶和周围软组织肿块在CT和MRI上均呈中度～明显强化.病理结果B细胞型5例、T细胞型4例.结论 影像学上PLB以斑点状或渗透性溶骨性破坏为主,病变骨质外形可正常或呈膨胀性改变,伴有明显的周围软组织肿块,中块以病骨为中心生长并有明显强化为其特征.     

Multimodal,multitask, multiattention (M3) deep learning detection of reticular pseudodrusen: Toward automated and accessible classification of age-related macular degeneration

ObjectiveReticular pseudodrusen (RPD), a key feature of age-related macular degeneration (AMD), are poorly detected by human experts on standard color fundus photography (CFP) and typically require advanced imaging modalities such as fundus autofluorescence (FAF). The objective was to develop and evaluate the performance of a novel multimodal, multitask, multiattention (M3) deep learning framework on RPD detection.Materials and MethodsA deep learning framework (M3) was developed to detect RPD presence accurately using CFP alone, FAF alone, or both, employing >8000 CFP-FAF image pairs obtained prospectively (Age-Related Eye Disease Study 2). The M3 framework includes multimodal (detection from single or multiple image modalities), multitask (training different tasks simultaneously to improve generalizability), and multiattention (improving ensembled feature representation) operation. Performance on RPD detection was compared with state-of-the-art deep learning models and 13 ophthalmologists; performance on detection of 2 other AMD features (geographic atrophy and pigmentary abnormalities) was also evaluated.ResultsFor RPD detection, M3 achieved an area under the receiver-operating characteristic curve (AUROC) of 0.832, 0.931, and 0.933 for CFP alone, FAF alone, and both, respectively. M3 performance on CFP was very substantially superior to human retinal specialists (median F1 score = 0.644 vs 0.350). External validation (the Rotterdam Study) demonstrated high accuracy on CFP alone (AUROC, 0.965). The M3 framework also accurately detected geographic atrophy and pigmentary abnormalities (AUROC, 0.909 and 0.912, respectively), demonstrating its generalizability.ConclusionsThis study demonstrates the successful development, robust evaluation, and external validation of a novel deep learning framework that enables accessible, accurate, and automated AMD diagnosis and prognosis.     

弥漫性腱鞘巨细胞瘤的影像学分析

目的探讨弥漫性腱鞘巨细胞瘤的临床和影像学表现。方法收集18例经手术病理证实的弥漫性腱鞘巨细胞瘤患者的临床和影像学资料,进行回顾性分析。18例患者年龄19-52岁,平均29岁,均行X线平片及MRI检查。结果病变位于膝部6例,踝部6例,足部3例,手部3例。X线上病变均表现为局部软组织肿块,其内均未见明显钙化灶,15例可见邻近骨质的破坏。MRI上病变呈分叶状肿块,边缘欠清,T1WI呈等、低为主混杂信号,T2WI呈等低信号为主,高信号并存,所有病例均可见特征性的含铁血黄素沉着的低信号区域。增强扫描病灶呈明显不均匀强化。MRI上有12例病变累及邻近骨髓,3例造成邻近骨皮质压迫性骨质吸收,3例邻近骨质未见异常。结论弥漫性腱鞘巨细胞瘤的临床表现和影像学特点对其的临床诊断与治疗很有价值。     

Plaque Tissue Morphology-Based Stroke Risk Stratification Using Carotid Ultrasound: A Polling-Based PCA Learning Paradigm

Severe atherosclerosis disease in carotid arteries causes stenosis which in turn leads to stroke. Machine learning systems have been previously developed for plaque wall risk assessment using morphology-based characterization. The fundamental assumption in such systems is the extraction of the grayscale features of the plaque region. Even though these systems have the ability to perform risk stratification, they lack the ability to achieve higher performance due their inability to select and retain dominant features. This paper introduces a polling-based principal component analysis (PCA) strategy embedded in the machine learning framework to select and retain dominant features, resulting in superior performance. This leads to more stability and reliability. The automated system uses offline image data along with the ground truth labels to generate the parameters, which are then used to transform the online grayscale features to predict the risk of stroke. A set of sixteen grayscale plaque features is computed. Utilizing the cross-validation protocol (K = 10), and the PCA cutoff of 0.995, the machine learning system is able to achieve an accuracy of 98.55 and 98.83%corresponding to the carotidfar wall and near wall plaques, respectively. The corresponding reliability of the system was 94.56 and 95.63%, respectively. The automated system was validated against the manual risk assessment system and the precision of merit for same cross-validation settings and PCA cutoffs are 98.28 and 93.92%for the far and the near wall, respectively.PCA-embedded morphology-based plaque characterization shows a powerful strategy for risk assessment and can be adapted in clinical settings.     

单发型四肢长骨内生软骨瘤影像学诊断

目的：探讨单发型四肢长骨内生软骨瘤的临床和影像学特征。方法回顾性分析12例经手术病理证实的单发型四肢长骨内生软骨瘤的临床和影像资料。结果 X线平片12例,分为三种表现：2例仅表现为钙化,1例仅表现为骨质破坏,骨质破坏伴钙化9例。CT检查8例,所有病例均显示骨质破坏伴瘤内不同程度的钙化,CT显示小的骨质破坏及细微钙化更有价值。MRI检查7例,所有病灶境界清楚,平扫T1WI呈低信号为主,1例病灶内见斑片状高信号,T2WI均呈不均匀明显高信号,增强后4例病灶呈环形不均匀强化,3例呈斑片状不均匀强化。X线平片、CT及MRI图像所有病例均未见骨膜反应及软组织肿块。结论单发型长骨内生软骨瘤多具有典型的影像学特征,综合X线平片、CT和（或） MRI对本病大多数可做出定性诊断与鉴别诊断。     

基于深度学习的肝包虫疾病图像分类

目的:为了实现新疆高发病肝包虫病CT图像的正确分类,提出一种深度学习的肝包虫病CT图像的自动分类方法。方法:对单囊、多囊和肝囊肿CT图像使用深度学习的分类方法进行分类。首先,构建并优化ResNet-50网络模型,将肝包虫病图像分批次传入网络,然后用交叉熵作为损失函数,最后把网络结构加入对数据的批归一化处理,通过反向传播算法优化参数使损失函数最小化,最终选择训练所得的最优网络。结果:各类别的最佳分类准确率分别为单囊型78.33%、多囊型81.52%、肝囊肿型80.24%。结论:深度学习卷积神经网络的肝包虫病CT图像疾病分类方法可行、合理、且调整后的ResNet-50模型比较适合肝包虫病图像的分类,有望通过深度学习方法对肝包虫病提供辅助诊断及决策支持。     

原发性骨淋巴瘤的影像学表现分析

目的分析原发性骨淋巴瘤（PLB）的影像学表现，提高临床原发性骨淋巴瘤的影像诊断率方法17例经病理证实的原发性骨淋巴瘤，男性12例，女性5例，年龄7—80岁，中位年龄47．5岁，17例经X线及CT检查，12例MR检查。结果（1）发病部位：股骨6例次，髂骨、胸椎各3例次，胫骨、肱骨、腰椎各2例次，骶骨、胸骨各1例次，长管骨发病占59％。（2）病灶数目：17例PLB中，1例多骨受累，其余均为单骨受累；（3）病变特征：溶骨性骨质破坏，骨质硬化，轻度骨膜反应，明显软组织肿块，病理性骨折。结论（1）PLB影像学上缺乏特征性表现，须X线、CT及MRI结合观察，才能做出正确的诊断。（2）PLB与其它骨恶性肿瘤鉴别困难。但全身症状轻、病灶骨质破坏轻而软组织肿块大、骨膜反应少强烈提示PLB诊断。     

动脉瘤样骨囊肿的影像学诊断

目的评价影像学检查对动脉瘤样骨囊肿（ABC）的诊断价值。方法对16例经手术病理证实的ABc的影像学表现进行回顾性分析。16例均拍X线平片，9例CT扫描，6例MRI扫描。结果16例ABE发生于长管状骨9例，脊柱4例，短骨及扁骨3例。13例病变呈膨胀性骨质破坏，10例病变边缘可见硬化，5例合并病理性骨折，8例病变可见分隔，11例可见典型的液一液平面，3例病变周围可见软组织异常改变，其中2例软组织肿胀，1例软组织肿块，12例病变内密度和（或）信号显示不均匀。结论动脉瘤样骨囊肿具有一定的影像学特征，CT和MRI对ABC诊断较平片有优势，综合影像学检查能提高ABC诊断符合率。