基于XGBoost对肺鳞癌和肺腺癌的分类预测

目的 对肺癌亚型肺鳞状细胞癌（肺鳞癌）和肺腺癌进行预测并找出分子标记。方法 通过研究两种不同癌症亚型中mRNA表达量,选取有差异有统计学意义的mRNA,利用极限梯度增强（extreme gradient boosting,XGBoost）算法构建模型,预测亚型分类,并比较其与逻辑回归分类模型和支持向量机分类模型的预测性能。结果 基于XBGoost模型的预测准确率为96.55%,曲线下面积为99.04%,优于逻辑回归分类模型和支持向量机分类模型。同时,找到11个基因作为两种亚型的分子标记。结论 肺癌两种亚型的在分子层面存在明显差异特征,将辅助临床医生进行疾病亚型预测。     

生物电阻抗测试频率对乳腺癌相关淋巴水肿定量评估的影响

目的 本研究通过对乳腺癌相关淋巴水肿患侧上肢不同频率生物电阻抗值的检测,分析生物电阻抗测试频率对淋巴水肿定量评估的影响,为生物电阻抗技术定量评估乳腺癌相关淋巴水肿临床应用提供依据。方法 选取乳腺癌相关淋巴水肿患者30例,采用多频生物电阻抗人体成分分析仪,检测受试者双上肢不同频率生物电阻值。比较健侧和患侧上肢平均周径、生物电阻抗值的差异,对双上肢平均周径差和不同频率阻抗差进行相关性分析。结果 （1）受试者患侧上肢周径和不同频率下阻抗值与健侧比较,差异有统计学意义（P<0.05）;（2）患侧和健侧上肢平均周径差与50 kHz下的阻抗差呈正相关（r=0.453,P=0.023）,低频率的生物电阻抗值与传统周径测量法的一致性好;（3）患侧50 kHz频率阻抗值较健侧阻抗值增加百分比显著高于肢体周径增加百分比（P<0.01）。结论 低频生物电阻抗值在淋巴水肿定量评估中有重要的应用价值。     

癫痫多导脑电非线性分析

目的 分析癫痫脑电信号,观察癫痫患者与健康成人脑电的非线性特征参数的差异值以利于识别。方法 采用多变量相空间重构对癫痫病人的EEG进行非线性分析,通过分析得到的相关维数估算值。相关维数D₂是定量刻画复杂非线性动力系统复杂性的最常用参数之一。结果和结论 相关维数能较好地识别癫痫脑电的一个特征参数。     

基于电子病历的重症老年患者急性肾损伤连续风险预测研究

目的 探索重症老年患者(≥60岁)急性肾损伤早期连续风险预测的可行性,促进机器学习在临床决策支持中的应用。具体实现以6 h为单位连续预测重症老年患者在未来48 h的急性肾损伤发病风险,并探索可实现何种程度的早期预测,以及比较当前数据和累积数据的预测效果。方法 基于重症监护医学信息数据库(Medical Information Mart for Intensive Care,MIMIC)-Ⅲ,应用逻辑回归、支持向量机、随机森林和轻量梯度提升机(light gradient boosting machine,LightGBM)建模预测。基于曲线下面积(area under curve,AUC)、精确度和召回率进行结果评估。结果 共11 261条重症老年患者记录纳入研究。基于当前6 h数据预测时,LightGBM的AUC达0.845~0.925,随机森林、支持向量机和逻辑回归的最高AUC均低于0.73。基于入重症监护病房最初6 h数据,LightGBM效果最好,AUC达0.845。LightGBM应用当前数据比累积数据获得更高的AUC、精确度和召回率,随机森林、支持向量机和逻辑回归反之。结论 利用LightGBM对重症老年患者进行急性肾损伤早期连续预测切实可行,仅基于重症监护病房前6 h数据的预测结果就可以达到24 h积累数据的预测效果。此外,不同模型对数据的接收能力和适用性不同,LightGBM在当前数据中表现优于累积数据,其他3种模型在累积数据中表现优于当前数据。     

基于半监督卷积神经网络进行人机不同步的识别

目的 基于半监督卷积神经网络(semi-supervised convolutional neural network, semi-CNN)构建人机不同步现象(patient-ventilator asynchrony, PVA)识别模型,评价其在压力支持通气(pressure support ventilation,PSV)模式下的诊断效能。方法 分析85例接受PSV通气脑损伤患者的机械通气数据,结合食道压监测数据进行人工标识。使用Transformer时间序列预测模型对已标识的正常或发生PVA的呼吸进行转化,转化后的数据输入semi-CNN模型判断是否发生PVA。在测试集中验证模型的准确性、灵敏度、特异度以及与专家标识结果的一致性。结果 初始训练集包含正常呼吸513次,异常呼吸69次,经过500次迭代后模型收敛。测试集包含正常呼吸48次,异常呼吸24次。在测试集中,Transformer联合semi-CNN模型识别PVA的准确率为0.92(0.83~0.97),灵敏度为0.79(0.58~0.93),特异度为0.98(0.89~1.00),Kappa值为0.80 (0.65~0.95),测试结果与专家人工标识结果具有高度一致性。结论 本研究提供了一种基于semi-CNN算法的PVA识别模型,其识别PVA的准确率和特异度高,识别结果与专家人工标识结果的一致性好,可用于临床实时PVA监测。     

面向慢性病人群的智能膳食评估系统

目的 基于移动终端拍摄的食物图像对慢性病患者的日常饮食进行智能营养评估。方法 构建基于人工智能的膳食评估系统,利用深度学习技术与图像处理方法,实现食物图像的智能分割、识别与营养素估算,使慢性病患者仅依据智能手机拍摄的食品图像即可得到食物的营养素信息。该系统同时支持172类中餐食谱与353种食材的细粒度识别,并在Vireo Food-172食谱数据集上得到了验证。 结果 基于卷积神经网络模型的食谱预测准确率为89.72%,食材评估指标微平均(micro-averaging, Micro-F1)提升至79.06%,宏平均(macro-averaging, Macro-F1)提升至 64.28%,在Vireo Food-172食谱数据集上取得了目前食材分类的最佳性能;基于食谱与食材识别结果对食物营养素进行估计,估计值与参考值误差均处于合理的范围内。结论 本系统可实现针对慢性病人群的智能膳食评估,便于患者进行每日饮食的自我监督,且有助于辅助营养师完成患者的日常饮食记录与评估,具有实用价值与研究意义。     

基于计算机视觉的糖尿病视网膜病变自动筛查系统

目的 本研究旨在利用计算机视觉相关技术自动识别眼底影像中糖尿病视网膜病变(diabetic retinopathy,DR,以下简称糖网)的病变特征,开发能够用于DR筛查的计算机自动筛查系统。方法 利用数学形态学和支撑向量机(support vector machine,SVM)分类技术设计出检测DR包括出血、渗出、微血管瘤等各类病变的算法,再根据DR的临床诊断标准,对眼底影像进行自动分级诊断,实现自动筛查。结果 利用建立完成的糖网自动筛查系统对国际Messidor数据库进行了筛查判断,以经过专家认证的诊断结果作为判定标准。在总共1 200张眼底图中,系统的判定灵敏度(sensitivity)为93.8%,特异度(specificity)为94.5%,检测时间为9.83 s。结论 基于计算机视觉算法开发的糖网自动筛查系统能准确、高效的完成眼科影像的糖网筛查工作,能大幅减少阅片医生的工作量和人为的主观性,具有很好的临床应用前景和社会效益。     

利用转录因子结合位点识别人类肿瘤特异性启动子研究

目的 研究应用计算机技术对人类肿瘤特异性启动子进行识别、预测。方法 通过收集肿瘤特异性启动子序列、转录因子结合位点序列、非肿瘤启动子序列3种数据集合,利用转录因子结合位点在不同序列集合中的密度求出各位点的对应密度比,确定识别特征,进行肿瘤特异性启动子的识别。结论 该方法具有较高的准确性,在保证对训练集合90%以上的识别率的情况下,对测试集合的识别率达到80%以上。     

MRI诊断140例腰椎智能网络自动检测分型MCs方法的比较

目的 探究使用Yolov5网络检测分类Modic改变(MCs)的性能,与基于Yolov5和Resnet34网络自动检测分类MCs方法进行比较。 方法 回顾性分析2020年6月至2021年6月接受MRI诊断且符合纳入和排除标准的MCs患者140例,其中男55例,女85例,25~89岁,平均(59.0±13.7)岁。在完成MRI影像的标注工作后,将标注后的MRI影像导入深度学习模型训练,使用医学数据常规增强和Mosaic数据增强进行数据扩充,降低训练数据集过少的因素;利用迁移学习的方法,解决网络在小数据集上过拟合的问题。采用平均精度(AP)、平均精度均值(mAP)、召回率、精确率、F1值等性能指标对两种方法诊断MCs进行评估并比较。 结果 Yolov5网络检测分类MCs时,mAP、召回率、精确率和F1值分别达到87.56%、82.05%、89.44%和0.845;Yolov5和Resnet34网络自动检测分类MCs时,召回率、精确率和F1值分别达到88.41%、88.68%和0.885。 结论 Yolov5网络可以帮助诊断腰椎MCs,使用Yolov5和Resnet34网络检测分类MCs时,模型诊断MCs的性能提升,进而表明Yolo系列网络可以为智能辅助诊断技术在脊柱领域的应用提供可能性。     

基于机器学习算法建立2型糖尿病患者冠心病辅助诊断模型

 目的 筛选2型糖尿病患者群合并冠心病危险因素并建立风险分类模型,为临床辅助诊断提供有价值的参考。方法 通过重庆医科大学大数据平台收集出院时间为2014年1月1日至2019年12月31日行冠状动脉造影术的2型糖尿病患者944例,根据造影结果分为2型糖尿病合并冠心病715例（T2DM-CAD组）和2型糖尿病非冠心病229例（T2DM组）。采用倾向得分匹配法（Propensity Score Matching,PSM）均衡组间混杂因素的影响,匹配后T2DM-CAD组389例,T2DM组221例。使用单因素分析与Logistic回归筛选冠心病发病的危险因素。采用贝叶斯优化（Bayesian Optimization,BO）算法优化支持向量机（Support Vector Machine,SVM）模型、随机森林（Random Forest,RF）模型、极限梯度上升（eXtreme Gradient Boosting,XGB）模型和Logistic回归模型,并比较4种分类模型的分类性能。结果 共收集缺失值<30%的指标35项,单因素分析筛选出有统计学差异的指标20项。逐步向前Logistic回归筛选出11项危险因素,包括心率、吸烟、糖尿病肾病、血肌酐、甘油三酯、脂蛋白a、白蛋白、总胆红素、谷草转氨酶、糖化血红蛋白和尿糖。基于危险因素建立的分类模型中优化后的RF模型性能在5折交叉验证（F1值=0.711,AUC=0.811） 以及验证集（F1值=0.752,AUC=0.810）中表现最优。结论 建立了参数优化RF模型,可用于判断2型糖尿病患者是否合并冠心病,具有良好性能。     

基于同轴相衬成像方法与支持向量机算法的早期骨性关节炎的鉴别诊断

目的 利用同轴相衬成像(in-line phase contrast imaging,IL-PCI)技术与支持向量机(support vector machine,SVM)算法对正常与早期骨性关节炎(osteoarthritis,OA)软骨组织建立分类模型.方法 研究样本分别来自接受人工膝关节置换手术及创伤性关节损伤患...     

Diagnosis of colon cancer with Fourier transform infrared spectroscopy on the malignant colon tissue samples

Background  Fourier transform infrared spectroscopy (FT-IR) combined with chemometrics discriminant analysis technology could improve diagnosis. The present study aimed to evaluate the effects of FT-IR on malignant colon tissue samples in diagnosis of colon cancer.

Methods  Principal component analysis (PCA) and support vector machine classification were used to discriminate FT-IR spectra from malignant and normal tissue. Colon tissues samples from 85 patients were used to demonstrate the procedure.

Results  For this set of colon spectral data, the  sensitivity and specificity of the support vector machine (SVM) classification were found both higher than 90%.

Conclusions  FT-IR provided important information about cancerous tissue, which could be used to discriminate malignant from normal tissues. The combination of PCA and SVM classification indicated that FT-IR has a potential clinical application in diagnosis of colon cancer.

     

基于PCA和KPCA特征抽取的SVM网络入侵检测方法

提出一种新颖的基于特征抽取的异常检测方法,应用主分量分析(PCA)和核主分量分析(KPCA)抽取入侵特征,再应用支持向量机(SVM)检测入侵。其中PCA对输入特征做线性变换,而KPCA通过核函数进行非线性变换。利用KDD 99数据集,将PCA-SVM、KPCA-SVM与SVM、PCR、KPCR进行比较,结果显示:在不降低分类器性能的情况下,特征抽取方法能对输入数据有效降维。在各种方法中,KPCA与SVM的结合能得到最优入侵检测性能。     

高强度运动前后的人体下肢电阻抗特性变化分析

目的研究高强度运动前后人体下肢电阻抗血流信号及频谱特性的变化情况,为用电阻抗参数评价人体运动功能及疲劳程度做基础。方法利用课题组自行研制的电阻抗快速测量系统,分别测量10名志愿者高强度运动前后双侧下肢的阻抗血流信号及1 kHz^1 MHz频段的电阻抗幅值变化情况。结果阻抗血流信号变化特性显示,运动后人体双侧下肢阻抗血流信号幅度增加22%±13%。阻抗频谱特性分析显示,100 kHz以下频段,运动对下肢阻抗幅值的影响不显著;在100 kHz以上频段,运动后的下肢电阻抗幅值显著降低,测量频率越高,下降的程度越明显。1 MHz时的变化率可达53%±14%。结论人体电阻抗与其运动功能状态有着密切的关系,因而电阻抗参数有望作为一项定性甚至定量区分组织疲劳状态的重要指标。     

Electrical impedance scanning in breast tumor imaging: correlation with the growth pattern of lesion

Background This study researched the electric impedance properties of breast tissue and demonstrated the different characteristic of electrical impedance scanning （EIS） images. Methods The impedance character of 40 malignant tumors, 34 benign tumors and some normal breast tissue from 69 patients undergoing breast surgery was examined by EIS in vivo measurement and mammography screening, with a series of frequencies set between 100 Hz-100 kHz in the ex vivo spectroscopy measurement. Results Of the 39 patients with 40 malignant tumors, 24 showed bright spots, 11 showed dark areas in EIS and 5 showed no specific image. Of the 30 patients with 34 benign tumors there were almost no specific abnormality shown in the EIS results. Primary ex vivo spectroscopy experiments showed that the resistivity of various breast tissue take the following pattern： adipose tissue〉cancerous tissue〉mammary gland and benign tumor tissue. Conclusions There are significant differences in the electrical impedance properties between cancerous tissue and healthy tissue. The impedivity of benign tumor is lower, and is at the same level with that of the mammary glandular tissue. The distinct growth pattern of breast lesions determined the different electrical impedance characteristics in the EIS results.     

以傅里叶变换红外光谱学检测结肠癌组织标本诊断结肠癌

Background The survival of cancer patients primarily depends on early detection of cancer. Fourier transform infrared spectroscopy (FT-IR) combined with chemometrics discriminant analysis technology could improve diagnosis. The present study aimed to evaluate the effects of FT-IR on Malignant Colon Tissue Samples on diagnosis of colorectal cancer. Methods Principal component analysis and support vector machine classification were used to discriminate FT-IR spectra from malignant and normal tissue. Colon tissues samples from 85 patients were used to demonstrate the procedure. Results For this set of colon spectral data, the sensitivity and specificity of the support vector machine (SVM) classification are both higher than 90%. Conclusions FT-IR provided important information about cancerous tissue, which could be used to discriminate malignant from normal tissue. The combination of principal component analysis (PCA) and SVM classification indicates that FT-IR has a potential clinical application in diagnosis of colon cancer.     

Automated Screening of Arrhythmia Using Wavelet Based Machine Learning Techniques

Arrhythmia is one of the preventive cardiac problems frequently occurs all over the globe. In order to screen such disease at early stage, this work attempts to develop a system approach based on registration, feature extraction using discrete wavelet transform (DWT), feature validation and classification of electrocardiogram (ECG). This diagnostic issue is set as a two-class pattern classification problem (normal sinus rhythm versus arrhythmia) where MIT-BIH database is considered for training, testing and clinical validation. Here DWT is applied to extract multi-resolution coefficients followed by registration using Pan Tompkins algorithm based R point detection. Moreover, feature space is compressed using sub-band principal component analysis (PCA) and statistically validated using independent sample t-test. Thereafter, the machine learning algorithms viz., Gaussian mixture model (GMM), error back propagation neural network (EBPNN) and support vector machine (SVM) are employed for pattern classification. Results are studied and compared. It is observed that both supervised classifiers EBPNN and SVM lead to higher (93.41% and 95.60% respectively) accuracy in comparison with GMM (87.36%) for arrhythmia screening.     

Study on Contribution of Biological Interpretable and Computer-Aided Features Towards the Classification of Childhood Medulloblastoma Cells

Diagnosis and Prognosis of brain tumour in children is always a critical case. Medulloblastoma is that subtype of brain tumour which occurs most frequently amongst children. Post-operation, the classification of its subtype is most vital for further clinical management. In this paper a novel approach of pathological subtype classification using biological interpretable and computer-aided textural features is forwarded. The classifier for accurate features prediction is built purely on the feature set obtained by segmentation of the ground truth cells from the original histological tissue images, marked by an experienced pathologist. The work is divided into five stages: marking of ground truth, segmentation of ground truth images, feature extraction, feature reduction and finally classification. Kmeans colour segmentation is used to segment out the ground truth cells from histological images. For feature extraction we used morphological, colour and textural features of the cells followed by feature reduction using Principal Component Analysis. Finally both binary and multiclass classification is done using Support Vector Method (SVM). The classification was compared using six different classifiers and performance was evaluated employing five-fold cross-validation technique. The accuracy achieved for binary and multiclass classification before applying PCA were 95.4 and 62.1% and after applying PCA were 100 and 84.9% respectively. The run-time analysis are also shown. Results reveal that this technique of cell level classification can be successfully adopted as architectural view can be confusing. Moreover it conforms substantially to the pathologist’s point of view regarding morphological and colour features, with the addition of computer assisted texture feature.