医疗诊断决策系统的社会影响

    输入层(图像数据) 
||

    卷积层 - 池化层
||

    卷积层 - 池化层 
||

    全连接层
||

    输出层(疾病分类)

    咳嗽 -> 可能是肺部疾病的症状 
     -> 肺部疾病有肺癌/肺炎/...
     -> 患者曾有肺结节,与肺癌症状"咳血"相符合
     -> 查看患者胸部CT影像...

    import pandas as pd
    
    # 读取电子病历数据
    records = pd.read_csv('medical_records.csv')
    
    # 数据清洗
    records = records.dropna(subset=['symptoms', 'lab_tests']) 
    records = records[~records['ct_scan'].isna()]
    
    # 特征工程
    records['age_group'] = pd.cut(records['age'], bins=[0,18,35,55,75,120], labels=['<18','18-35','35-55','55-75','>75'])
    records['bmi'] = records['weight'] / (records['height']/100)**2 
    
    features = ['age_group', 'gender', 'bmi', 'symptoms', 'lab_tests']
    labels = records['pneumonia']
    
    # 分割数据集
    from sklearn.model_selection import train_test_split
    X_train, X_test, y_train, y_test = train_test_split(records[features], labels, test_size=0.2)
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/kqsztYWiwPoalXC4JQevFfGZMm89.png)

    import tensorflow as tf
    from tensorflow.keras.layers import *
    
    # 文本输入分支
    text_input = Input(shape=(max_len,), name='text')
    x = Embedding(vocab_size, embed_dim)(text_input)
    x = Bidirectional(GRU(units, return_sequences=True))(x)
    x = Attention()([x, x]) 
    x = tf.reduce_mean(x, axis=1)
    text_branch = Model(inputs=text_input, outputs=x)
    
    # 数值输入分支 
    numerical_input = Input(shape=(n_numerical,), name='numerical')
    x = numerical_branch = Dense(units, activation='relu')(numerical_input)
    
    # CT影像输入分支
    image_input = Input(shape=(img_width, img_height, 3), name='ct')  
    x = image_branch = ConvolutionBlock()(image_input)
    
    # 融合多分支输出  
    merged = Concatenate()([text_branch.output, numerical_branch, image_branch])
    x = Dropout(0.2)(merged)
    x = Dense(64, activation='relu')(x)
    output = Dense(1, activation='sigmoid')(x)
    
    model = Model(inputs=[text_branch.input, numerical_input, image_input], outputs=output)
    model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
    
    # 训练模型
    model.fit([X_train['symptoms'], X_train[numerical_cols], X_train['ct_scan']], 
          y_train, 
          validation_data=([X_val['symptoms'], X_val[numerical_cols], X_val['ct_scan']], y_val),
          epochs=20)
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/RECeoqX8Gl32MwUctbL7DPQYuskA.png)

    from reasoning import KGReasoner
    
    # 加载知识图谱
    reasoner = KGReasoner(kg_path='med_knowledgegraph.pkl')
    
    # 输入病历信息   
    patient_data = {
    '症状描述': '...', 
    '体征检查': {...},
    'CT报告': '...',
    '分类模型输出': [0.85] # 疾病概率
    }    
    
    # 执行知识推理,得到结构化诊断报告
    diagnosis_report = reasoner.explain(patient_data)
    print(diagnosis_report)
    
    python
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/DOnVh39rYz6BHGSUgENeCyj0baZu.png)

    诊断报告
    
    疾病名称: 肺炎
    置信度: 85%
    
    症状证据:
      - 发热(39.5℃), 权重0.95 
      - 咳嗽(黄色...浓痰), 权重0.9
      ...
    体征证据:
      - 血常规示中性粒细胞比例 81.2%
      ...  
    影像证据: 
      - CT示双肺可见斑片状...浸润
    关键證據途徑:
      规则1: 肺部感染 & 发热 & ...症状 -> 可能为肺炎
      规则2: CT表现合并症状 -> 确诊为肺炎   
      ...  
    相关病例:
      - 案例1: 63岁男性...
      - 案例2: ...
    辅助治疗建议:
      - 抗生素A方案, 可治愈概率0.77
      - 抗生素B方案, 可治愈概率0.63 (不建议)    
    
    
![](https://ad.itadn.com/c/weblog/blog-img/images/2025-08-18/SbwAq1maWTLIRuhfXClN8Y9cUEkM.png)