机器学习特征工程：特征选择及在肺癌CT分类中的优化应用

    # 安装依赖（若未安装）
    # pip install numpy pandas scikit-learn xgboost SimpleITK radiomics matplotlib
    
    
    python
    
    

    import SimpleITK as sitk
    import radiomics
    from radiomics import featureextractor
    import pandas as pd
    import numpy as np
    from sklearn.model_selection import train_test_split
    from sklearn.preprocessing import StandardScaler
    
    # 示例：加载CT影像和掩膜（需替换为实际路径）
    image_path = "path/to/ct_image.nii"
    mask_path = "path/to/mask_image.nii"
    image = sitk.ReadImage(image_path)
    mask = sitk.ReadImage(mask_path)
    
    # 初始化PyRadiomics提取器
    extractor = featureextractor.RadiomicsFeatureExtractor()
    settings = {'enableGLCM': True, 'enableShape': True}  # 启用GLCM和形状特征
    features = extractor.execute(image, mask)
    
    # 将特征转换为DataFrame
    feature_names = list(features.keys())
    feature_values = [features[name] for name in feature_names]
    X = np.array(feature_values).reshape(1, -1)  # 示例为单张影像
    
    # 假设已有多个样本的特征矩阵（替换为实际数据）
    # X: 特征矩阵（样本数 x 特征数），y: 标签（0:良性，1:恶性）
    # 这里使用模拟数据代替
    np.random.seed(42)
    X = np.random.rand(100, 50)  # 100个样本，50个特征
    y = np.random.randint(0, 2, 100)  # 随机标签
    
    # 数据标准化
    scaler = StandardScaler()
    X_scaled = scaler.fit_transform(X)
    
    # 划分训练集和测试集
    X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size=0.2, random_state=42)
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/p5YPgXWIxi7orF6h0H9cubTwy1LZ.png)

    from sklearn.feature_selection import SelectKBest, chi2
    import xgboost as xgb
    
    # 过滤法：卡方检验选择前10个特征
    # 确保特征非负（卡方检验要求）
    X_train_nonneg = np.maximum(X_train, 0)
    X_test_nonneg = np.maximum(X_test, 0)
    selector = SelectKBest(score_func=chi2, k=10)
    X_train_chi2 = selector.fit_transform(X_train_nonneg, y_train)
    X_test_chi2 = selector.transform(X_test_nonneg)
    
    # 获取选择的特征索引
    selected_features_chi2 = selector.get_support(indices=True)
    print("卡方检验选择的特征索引:", selected_features_chi2)
    
    # 嵌入法：XGBoost特征重要性
    model_xgb = xgb.XGBClassifier(random_state=42)
    model_xgb.fit(X_train, y_train)
    importance = model_xgb.feature_importances_
    top_k_idx = np.argsort(importance)[::-1][:10]
    print("XGBoost选择的特征索引:", top_k_idx)
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/T0SnecIFrd8ZMybGKkw7ANPBW9s3.png)

    from sklearn.ensemble import AdaBoostClassifier
    from sklearn.tree import DecisionTreeClassifier
    from sklearn.metrics import accuracy_score, classification_report
    
    # AdaBoost模型（使用过滤法特征）
    ada = AdaBoostClassifier(base_estimator=DecisionTreeClassifier(max_depth=1), n_estimators=50, random_state=42)
    ada.fit(X_train_chi2, y_train)
    y_pred_ada = ada.predict(X_test_chi2)
    print("AdaBoost准确率:", accuracy_score(y_test, y_pred_ada))
    print("AdaBoost分类报告:\n", classification_report(y_test, y_pred_ada))
    
    # XGBoost模型（使用嵌入法特征）
    X_train_xgb = X_train[:, top_k_idx]
    X_test_xgb = X_test[:, top_k_idx]
    model_xgb.fit(X_train_xgb, y_train)
    y_pred_xgb = model_xgb.predict(X_test_xgb)
    print("XGBoost准确率:", accuracy_score(y_test, y_pred_xgb))
    print("XGBoost分类报告:\n", classification_report(y_test, y_pred_xgb))
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/MAxL7USkTyO289pQiR3ePtY4wWFJ.png)

    from sklearn.model_selection import GridSearchCV
    
    # AdaBoost参数调优
    ada_param_grid = {
    'n_estimators': [50, 100, 200],
    'learning_rate': [0.01, 0.1, 1.0],
    'base_estimator__max_depth': [1, 2, 3]
    }
    ada = AdaBoostClassifier(base_estimator=DecisionTreeClassifier(random_state=42))
    ada_grid = GridSearchCV(ada, ada_param_grid, cv=5, scoring='accuracy', n_jobs=-1)
    ada_grid.fit(X_train_chi2, y_train)
    
    # 输出最优参数和性能
    print("AdaBoost最佳参数:", ada_grid.best_params_)
    print("AdaBoost最佳准确率:", ada_grid.best_score_)
    y_pred_ada = ada_grid.predict(X_test_chi2)
    print("AdaBoost测试集准确率:", accuracy_score(y_test, y_pred_ada))
    print("AdaBoost分类报告:\n", classification_report(y_test, y_pred_ada))
    
    # XGBoost参数调优
    xgb_param_grid = {
    'n_estimators': [50, 100, 200],
    'learning_rate': [0.01, 0.1, 0.3],
    'max_depth': [3, 5, 7],
    'reg_lambda': [0.1, 1.0, 10.0],
    'subsample': [0.6, 0.8, 1.0]
    }
    xgb_model = xgb.XGBClassifier(random_state=42)
    xgb_grid = GridSearchCV(xgb_model, xgb_param_grid, cv=5, scoring='accuracy', n_jobs=-1)
    xgb_grid.fit(X_train_xgb, y_train)
    
    # 输出最优参数和性能
    print("XGBoost最佳参数:", xgb_grid.best_params_)
    print("XGBoost最佳准确率:", xgb_grid.best_score_)
    y_pred_xgb = xgb_grid.predict(X_test_xgb)
    print("XGBoost测试集准确率:", accuracy_score(y_test, y_pred_xgb))
    print("XGBoost分类报告:\n", classification_report(y_test, y_pred_xgb))
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/FhJ0npxVlPQK1ErfS6OYRCZL5wvg.png)

    import tensorflow as tf
    from tensorflow.keras.applications import ResNet50
    from tensorflow.keras.applications.resnet50 import preprocess_input
    from tensorflow.keras.preprocessing.image import load_img, img_to_array
    
    # 模拟CT影像数据（替换为实际LIDC-IDRI影像）
    def load_ct_images(image_paths):
    images = []
    for path in image_paths:
        img = load_img(path, target_size=(224, 224))
        img = img_to_array(img)
        images.append(img)
    return np.array(images)
    
    # 模拟数据
    image_paths = ["path/to/ct_image_{}.png".format(i) for i in range(100)]
    X_images = np.random.rand(100, 224, 224, 3)
    y = np.random.randint(0, 2, 100)
    
    # 预处理影像
    X_images = preprocess_input(X_images)
    
    # 加载预训练ResNet50
    base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
    X_features = base_model.predict(X_images)
    X_features = X_features.reshape(X_features.shape[0], -1)
    
    # 数据标准化
    scaler = StandardScaler()
    X_features = scaler.fit_transform(X_features)
    
    # 划分训练集和测试集
    X_train, X_test, y_train, y_test = train_test_split(X_features, y, test_size=0.2, random_state=42)
    
    # 特征选择（嵌入法：XGBoost）
    xgb_model = xgb.XGBClassifier(random_state=42)
    xgb_model.fit(X_train, y_train)
    importance = xgb_model.feature_importances_
    top_k_idx = np.argsort(importance)[::-1][:100]
    X_train_selected = X_train[:, top_k_idx]
    X_test_selected = X_test[:, top_k_idx]
    
    # 训练XGBoost
    xgb_model.fit(X_train_selected, y_train)
    y_pred_xgb = xgb_model.predict(X_test_selected)
    print("XGBoost（CNN特征）准确率:", accuracy_score(y_test, y_pred_xgb))
    print("XGBoost（CNN特征）分类报告:\n", classification_report(y_test, y_pred_xgb))
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/NRHV2YWolPeFz8MgpTk7jJQwSZGr.png)

    import matplotlib.pyplot as plt
    
    # 模拟优化前后准确率
    models = ["AdaBoost (基线)", "AdaBoost (调优)", "XGBoost (基线)", "XGBoost (调优)", "XGBoost (CNN+调优)"]
    accuracies = [0.85, 0.90, 0.90, 0.95, 0.97]
    
    plt.figure(figsize=(10, 6))
    plt.bar(models, accuracies, color=['blue', 'lightblue', 'green', 'lightgreen', 'orange'])
    plt.xlabel("模型")
    plt.ylabel("准确率")
    plt.title("优化前后模型性能对比")
    plt.ylim(0, 1)
    for i, v in enumerate(accuracies):
    plt.text(i, v + 0.01, f"{v:.2f}", ha='center')
    plt.tight_layout()
    plt.show()
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/4q3WActsSyNaKOE7Rm2efj5lhMJw.png)

    from sklearn.metrics import roc_curve, auc
    
    # 计算ROC曲线（XGBoost with CNN特征）
    y_score = xgb_model.predict_proba(X_test_selected)[:, 1]
    fpr, tpr, _ = roc_curve(y_test, y_score)
    roc_auc = auc(fpr, tpr)
    
    plt.figure(figsize=(8, 6))
    plt.plot(fpr, tpr, label=f"ROC曲线 (AUC = {roc_auc:.2f})")
    plt.plot([0, 1], [0, 1], 'k--')
    plt.xlabel("假阳性率")
    plt.ylabel("真阳性率")
    plt.title("XGBoost（CNN特征）ROC曲线")
    plt.legend(loc="lower right")
    plt.grid(True)
    plt.tight_layout()
    plt.show()
    
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/pmU3zwnVeiYRsjJA1WG4h6yDQ8aX.png)