辅助医生能力成长与患者个体化治疗方案生成系统（下）

第六章 反馈学习与模型迭代6.1 医生修正数据记录# ==================== 反馈收集模块 ====================classFeedbackCollector:"""收集医生对AI方案的修正行为，形成训练样本"""def__init__(self,storage_path:str='feedback.db'):self.storage_path=storage_path self._init_db()def_init_db(self):importsqlite3 conn=sqlite3.connect(self.storage_path)c=conn.cursor()c.execute('''CREATE TABLE IF NOT EXISTS feedback (id INTEGER PRIMARY KEY AUTOINCREMENT, patient_id TEXT, original_plan TEXT, doctor_modification TEXT, modification_type TEXT, reason TEXT, timestamp DATETIME DEFAULT CURRENT_TIMESTAMP)''')conn.commit()conn.close()defrecord_feedback(self,patient_id:str,original_plan:Dict,modified_plan:Dict,mod_type:str,reason:str=''):"""记录一次修正"""importsqlite3,json conn=sqlite3.connect(self.storage_path)c=conn.cursor()c.execute('''INSERT INTO feedback (patient_id, original_plan, doctor_modification, modification_type, reason) VALUES (?, ?, ?, ?, ?)''',(patient_id,json.dumps(original_plan),json.dumps(modified_plan),mod_type,reason))conn.commit()conn.close()defget_training_samples(self,limit:int=1000)-List[Dict]:"""获取用于模型训练的样本"""importsqlite3,json conn=sqlite3.connect(self.storage_path)c=conn.cursor()c.execute('''SELECT patient_id, original_plan, doctor_modification, modification_type FROM feedback ORDER BY timestamp DESC LIMIT ?''',(limit,))rows=c.fetchall()conn.close()samples=[]forrowinrows:samples.append({'patient_id':row[0],'original':json.loads(row[1]),'modified':json.loads(row[2]),'type':row[3]})returnsamples6.2 闭环训练流程# ==================== 闭环训练管理器 ====================classContinuousLearningLoop:""" 管理从反馈数据到模型更新的闭环 1. 定期从反馈库抽取样本 2. 转换为监督学习格式 3. 增量训练ML模型 4. 更新规则库权重 """def__init__(self,predictor:MLDiseasePredictor,collector:FeedbackCollector):self.predictor=predictor self.collector=collectordefprepare_training_data(self)-tuple:"""从反馈数据构造训练集（疾病预测）"""samples=self.collector.get_training_samples()X_symptoms=[]X_patients=[]y_diseases=[]forsampleinsamples:ifsample['type']=='diagnosis_correction':# 原始输入症状和患者信息应从original_plan中提取（此处简化）# 实际需从完整病历获取，这里仅示意passreturnX_symptoms,X_patients,y_diseasesdefretrain_model(self):"""执行模型重训练"""X_s,X_p,y=self.prepare_training_data()iflen(y)50:# 最小样本量self.predictor.train(X_s,X_p,y)self.predictor.save_model('disease_model_v2.pkl')returnTruereturnFalsedefextract_rare_case_rules(self)-List[Dict]:""" 从反馈中挖掘罕见病例的处理规则 例如：某疾病在某特殊人群中用药调整 """# 使用关联规则挖掘或简单统计# 返回新规则列表供人工审核后加入规则库pass第七章 医生成长逻辑模块7.1 能力评估与学习路径推荐# ==================== 医生能力成长追踪 ====================classDoctorGrowthTracker:""" 追踪医生使用系统的行为，评估能力短板，推荐学习内容 """def__init__(self):self.doctor_profiles={}# 医生ID - 能力档案defupdate_profile(self,doctor_id:str,case_record:Dict):"""根据一例接诊更新医生能力档案"""ifdoctor_idnotinself.doctor_profiles:self.doctor_profiles[doctor_id]={'total_cases':0,'disease_exposure':{},# 疾病 - 处理例数'modification_rate':0.0,# 修改AI方案的比例'avg_confidence':0.0,'weak_areas':[]}profile=self.doctor_profiles[doctor_id]profile['total_cases']+=1disease=case_record.get('final_diagnosis')profile['disease_exposure'][disease]=profile['disease_exposure'