Healthapp / app.py
mgbam's picture
Update app.py
55ef016 verified
import os
import json
import base64
import io
import ast
import logging
from abc import ABC, abstractmethod
from typing import Dict, List, Optional, Any
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import streamlit as st
import spacy
from scipy.stats import ttest_ind, f_oneway
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from statsmodels.tsa.seasonal import seasonal_decompose
from statsmodels.tsa.stattools import adfuller
from pydantic import BaseModel, Field
from Bio import Entrez # Ensure BioPython is installed
from dotenv import load_dotenv
import requests
import openai # Updated for OpenAI SDK v1.0.0+
from openai import OpenAIError, RateLimitError, BadRequestError, OpenAI
# ---------------------- Load Environment Variables ---------------------------
load_dotenv()
# ---------------------- Logging Configuration ---------------------------
logging.basicConfig(
filename='app.log',
filemode='a',
format='%(asctime)s - %(levelname)s - %(message)s',
level=logging.INFO
)
logger = logging.getLogger()
# ---------------------- Streamlit Page Configuration ---------------------------
st.set_page_config(page_title="AI Clinical Intelligence Hub", layout="wide")
# ---------------------- Initialize OpenAI SDK ---------------------------
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
PUB_EMAIL = os.getenv("PUB_EMAIL", "")
if not OPENAI_API_KEY:
st.error("OpenAI API key must be set as an environment variable (OPENAI_API_KEY).")
st.stop()
# Instantiate the OpenAI client
try:
client = OpenAI(api_key=OPENAI_API_KEY) # Instantiating the client right here
except Exception as e:
st.error(f"Failed to initialize OpenAI client: {e}")
logger.error(f"Failed to initialize OpenAI client: {e}")
st.stop()
# ---------------------- Load spaCy Model ---------------------------
try:
nlp = spacy.load("en_core_web_sm")
except OSError:
# Avoid using Streamlit commands before set_page_config()
import subprocess
import sys
subprocess.run([sys.executable, "-m", "spacy", "download", "en_core_web_sm"])
nlp = spacy.load("en_core_web_sm")
# ---------------------- Base Classes and Schemas ---------------------------
class ResearchInput(BaseModel):
"""Base schema for research tool inputs."""
data_key: str = Field(..., description="Session state key containing DataFrame")
columns: Optional[List[str]] = Field(None, description="List of columns to analyze")
class TemporalAnalysisInput(ResearchInput):
"""Schema for temporal analysis."""
time_col: str = Field(..., description="Name of timestamp column")
value_col: str = Field(..., description="Name of value column to analyze")
class HypothesisInput(ResearchInput):
"""Schema for hypothesis testing."""
group_col: str = Field(..., description="Categorical column defining groups")
value_col: str = Field(..., description="Numerical column to compare")
class ModelTrainingInput(ResearchInput):
"""Schema for model training."""
target_col: str = Field(..., description="Name of target column")
class DataAnalyzer(ABC):
"""Abstract base class for data analysis modules."""
@abstractmethod
def invoke(self, data: pd.DataFrame, **kwargs) -> Dict[str, Any]:
pass
# ---------------------- Concrete Analyzer Implementations ---------------------------
class AdvancedEDA(DataAnalyzer):
"""Comprehensive Exploratory Data Analysis."""
def invoke(self, data: pd.DataFrame, **kwargs) -> Dict[str, Any]:
try:
analysis = {
"dimensionality": {
"rows": len(data),
"columns": list(data.columns),
"memory_usage_MB": f"{data.memory_usage().sum() / 1e6:.2f} MB"
},
"statistical_profile": data.describe(percentiles=[.25, .5, .75]).to_dict(),
"temporal_analysis": {
"date_ranges": {
col: {
"min": data[col].min(),
"max": data[col].max()
} for col in data.select_dtypes(include='datetime').columns
}
},
"data_quality": {
"missing_values": data.isnull().sum().to_dict(),
"duplicates": data.duplicated().sum(),
"cardinality": {
col: data[col].nunique() for col in data.columns
}
}
}
return analysis
except Exception as e:
logger.error(f"EDA Failed: {str(e)}")
return {"error": f"EDA Failed: {str(e)}"}
class DistributionVisualizer(DataAnalyzer):
"""Distribution visualizations."""
def invoke(self, data: pd.DataFrame, columns: List[str], **kwargs) -> str:
try:
plt.figure(figsize=(12, 6))
for i, col in enumerate(columns, 1):
plt.subplot(1, len(columns), i)
sns.histplot(data[col], kde=True, stat="density")
plt.title(f'Distribution of {col}', fontsize=10)
plt.xticks(fontsize=8)
plt.yticks(fontsize=8)
plt.tight_layout()
buf = io.BytesIO()
plt.savefig(buf, format='png', dpi=300, bbox_inches='tight')
plt.close()
return base64.b64encode(buf.getvalue()).decode()
except Exception as e:
logger.error(f"Visualization Error: {str(e)}")
return f"Visualization Error: {str(e)}"
class TemporalAnalyzer(DataAnalyzer):
"""Time series analysis."""
def invoke(self, data: pd.DataFrame, time_col: str, value_col: str, **kwargs) -> Dict[str, Any]:
try:
ts_data = data.set_index(pd.to_datetime(data[time_col]))[value_col]
decomposition = seasonal_decompose(ts_data, period=365)
plt.figure(figsize=(12, 8))
decomposition.plot()
plt.tight_layout()
buf = io.BytesIO()
plt.savefig(buf, format='png')
plt.close()
plot_data = base64.b64encode(buf.getvalue()).decode()
stationarity_p_value = adfuller(ts_data)[1]
return {
"trend_statistics": {
"stationarity_p_value": stationarity_p_value,
"seasonality_strength": float(max(decomposition.seasonal))
},
"visualization": plot_data
}
except Exception as e:
logger.error(f"Temporal Analysis Failed: {str(e)}")
return {"error": f"Temporal Analysis Failed: {str(e)}"}
class HypothesisTester(DataAnalyzer):
"""Statistical hypothesis testing."""
def invoke(self, data: pd.DataFrame, group_col: str, value_col: str, **kwargs) -> Dict[str, Any]:
try:
groups = data[group_col].unique()
if len(groups) < 2:
return {"error": "Insufficient groups for comparison"}
group_data = [data[data[group_col] == g][value_col] for g in groups]
if len(groups) == 2:
stat, p = ttest_ind(*group_data)
test_type = "Independent t-test"
effect_size = self.calculate_cohens_d(group_data[0], group_data[1])
else:
stat, p = f_oneway(*group_data)
test_type = "ANOVA"
effect_size = None
return {
"test_type": test_type,
"test_statistic": stat,
"p_value": p,
"effect_size": effect_size,
"interpretation": self.interpret_p_value(p)
}
except Exception as e:
logger.error(f"Hypothesis Testing Failed: {str(e)}")
return {"error": f"Hypothesis Testing Failed: {str(e)}"}
@staticmethod
def calculate_cohens_d(x: pd.Series, y: pd.Series) -> Optional[float]:
"""Calculate Cohen's d for effect size."""
try:
mean_diff = abs(x.mean() - y.mean())
pooled_std = np.sqrt((x.var() + y.var()) / 2)
return mean_diff / pooled_std
except Exception as e:
logger.error(f"Error calculating Cohen's d: {str(e)}")
return None
@staticmethod
def interpret_p_value(p: float) -> str:
"""Interpret the p-value."""
if p < 0.001:
return "Very strong evidence against H0"
elif p < 0.01:
return "Strong evidence against H0"
elif p < 0.05:
return "Evidence against H0"
elif p < 0.1:
return "Weak evidence against H0"
else:
return "No significant evidence against H0"
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import LabelEncoder
class LogisticRegressionTrainer(DataAnalyzer):
"""Logistic Regression Model Trainer with Missing Value Handling and Target Encoding."""
def invoke(self, data: pd.DataFrame, target_col: str, columns: List[str], **kwargs) -> Dict[str, Any]:
try:
# Prevent data leakage by removing target_col from features if present
if target_col in columns:
columns.remove(target_col)
logger.warning(f"Removed target column '{target_col}' from feature list to prevent data leakage.")
X = data[columns].copy()
y = data[target_col].copy()
# Handle missing values in X
if X.isnull().values.any():
logger.info("Missing values detected in feature variables. Applying imputation.")
imputer = SimpleImputer(strategy='mean') # Choose strategy as needed
X_imputed = imputer.fit_transform(X)
X = pd.DataFrame(X_imputed, columns=columns)
logger.info("Imputation completed for feature variables.")
else:
logger.info("No missing values detected in feature variables.")
# Handle missing values in y
if y.isnull().values.any():
logger.info("Missing values detected in target variable. Dropping missing targets.")
# For classification, it's common to impute with the mode or drop missing targets
data = data.dropna(subset=[target_col])
y = data[target_col]
X = data[columns]
logger.info("Dropped rows with missing target values.")
else:
logger.info("No missing values detected in target variable.")
# Encode target if it's categorical and not numeric
if y.dtype == 'object' or y.dtype.name == 'category':
logger.info("Encoding categorical target variable.")
label_encoder = LabelEncoder()
y = label_encoder.fit_transform(y)
logger.info("Encoding completed.")
# Split the data
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
logger.info("Data split into training and testing sets.")
# Initialize and train the model
model = LogisticRegression(max_iter=1000, multi_class='auto', solver='lbfgs')
model.fit(X_train, y_train)
logger.info("Logistic Regression model training completed.")
# Make predictions and evaluate
y_pred = model.predict(X_test)
accuracy = accuracy_score(y_test, y_pred)
logger.info(f"Model accuracy on test set: {accuracy:.2%}")
return {
"model_type": "Logistic Regression",
"accuracy": accuracy,
"model_params": model.get_params()
}
except Exception as e:
logger.error(f"Logistic Regression Model Error: {str(e)}")
return {"error": f"Logistic Regression Model Error: {str(e)}"}
# ---------------------- Business Logic Layer ---------------------------
class ClinicalRule(BaseModel):
"""Defines a clinical rule."""
name: str
condition: str
action: str
severity: str # low, medium, or high
class ClinicalRulesEngine:
"""Executes rules against patient data."""
def __init__(self):
self.rules: Dict[str, ClinicalRule] = {}
def add_rule(self, rule: ClinicalRule):
self.rules[rule.name] = rule
def execute_rules(self, data: pd.DataFrame) -> Dict[str, Any]:
results = {}
for rule_name, rule in self.rules.items():
try:
# Using safe_eval instead of eval for security
rule_matched = self.safe_eval(rule.condition, {"df": data})
results[rule_name] = {
"rule_matched": rule_matched,
"action": rule.action if rule_matched else None,
"severity": rule.severity if rule_matched else None
}
except Exception as e:
logger.error(f"Error executing rule '{rule_name}': {str(e)}")
results[rule_name] = {
"rule_matched": False,
"error": str(e),
"severity": None
}
return results
@staticmethod
def safe_eval(expr, variables):
"""
Safely evaluate an expression using AST parsing.
Only allows certain node types to prevent execution of arbitrary code.
"""
allowed_nodes = (
ast.Expression, ast.BoolOp, ast.BinOp, ast.UnaryOp, ast.Compare,
ast.Call, ast.Name, ast.Load, ast.Constant, ast.Num, ast.Str,
ast.List, ast.Tuple, ast.Dict
)
try:
node = ast.parse(expr, mode='eval')
for subnode in ast.walk(node):
if not isinstance(subnode, allowed_nodes):
raise ValueError(f"Unsupported expression: {expr}")
return eval(compile(node, '<string>', mode='eval'), {"__builtins__": None}, variables)
except Exception as e:
logger.error(f"safe_eval error: {str(e)}")
raise ValueError(f"Invalid expression: {e}")
class ClinicalKPI(BaseModel):
"""Define a clinical KPI."""
name: str
calculation: str
threshold: Optional[float] = None
class ClinicalKPIMonitoring:
"""Calculates KPIs based on data."""
def __init__(self):
self.kpis: Dict[str, ClinicalKPI] = {}
def add_kpi(self, kpi: ClinicalKPI):
self.kpis[kpi.name] = kpi
def calculate_kpis(self, data: pd.DataFrame) -> Dict[str, Any]:
results = {}
for kpi_name, kpi in self.kpis.items():
try:
# Using safe_eval instead of eval for security
kpi_value = self.safe_eval(kpi.calculation, {"df": data})
status = self.evaluate_threshold(kpi_value, kpi.threshold)
results[kpi_name] = {
"value": kpi_value,
"threshold": kpi.threshold,
"status": status
}
except Exception as e:
logger.error(f"Error calculating KPI '{kpi_name}': {str(e)}")
results[kpi_name] = {"error": str(e)}
return results
@staticmethod
def evaluate_threshold(value: Any, threshold: Optional[float]) -> Optional[str]:
if threshold is None:
return None
try:
return "Above Threshold" if value > threshold else "Below Threshold"
except TypeError:
return "Threshold Evaluation Not Applicable"
@staticmethod
def safe_eval(expr, variables):
"""
Safely evaluate an expression using AST parsing.
Only allows certain node types to prevent execution of arbitrary code.
"""
allowed_nodes = (
ast.Expression, ast.BoolOp, ast.BinOp, ast.UnaryOp, ast.Compare,
ast.Call, ast.Name, ast.Load, ast.Constant, ast.Num, ast.Str,
ast.List, ast.Tuple, ast.Dict
)
try:
node = ast.parse(expr, mode='eval')
for subnode in ast.walk(node):
if not isinstance(subnode, allowed_nodes):
raise ValueError(f"Unsupported expression: {expr}")
return eval(compile(node, '<string>', mode='eval'), {"__builtins__": None}, variables)
except Exception as e:
logger.error(f"safe_eval error: {str(e)}")
raise ValueError(f"Invalid expression: {e}")
class DiagnosisSupport(ABC):
"""Abstract class for implementing clinical diagnoses."""
@abstractmethod
def diagnose(
self,
data: pd.DataFrame,
target_col: str,
columns: List[str],
diagnosis_key: str = "diagnosis",
**kwargs
) -> pd.DataFrame:
pass
class SimpleDiagnosis(DiagnosisSupport):
"""Provides a simple diagnosis example, based on the Logistic regression model."""
def __init__(self, client: OpenAI):
self.model_trainer: LogisticRegressionTrainer = LogisticRegressionTrainer()
self.client = client # Using the OpenAI client
def diagnose(
self,
data: pd.DataFrame,
target_col: str,
columns: List[str],
diagnosis_key: str = "diagnosis",
**kwargs
) -> pd.DataFrame:
try:
result = self.model_trainer.invoke(data, target_col=target_col, columns=columns)
if "accuracy" in result:
return pd.DataFrame({
diagnosis_key: [f"Model Accuracy: {result['accuracy']:.2%}"],
"model": [result["model_type"]]
})
else:
return pd.DataFrame({
diagnosis_key: [f"Diagnosis failed: {result.get('error', 'Unknown error')}"]
})
except Exception as e:
logger.error(f"Error during diagnosis: {str(e)}")
return pd.DataFrame({
diagnosis_key: [f"Error during diagnosis: {e}"]
})
class TreatmentRecommendation(ABC):
"""Abstract class for treatment recommendations."""
@abstractmethod
def recommend(
self,
data: pd.DataFrame,
condition_col: str,
treatment_col: str,
recommendation_key: str = "recommendation",
**kwargs
) -> pd.DataFrame:
pass
class BasicTreatmentRecommendation(TreatmentRecommendation):
"""A placeholder class for basic treatment recommendations."""
def recommend(
self,
data: pd.DataFrame,
condition_col: str,
treatment_col: str,
recommendation_key: str = "recommendation",
**kwargs
) -> pd.DataFrame:
if condition_col not in data.columns or treatment_col not in data.columns:
logger.warning(f"Condition or Treatment columns not found: {condition_col}, {treatment_col}")
return pd.DataFrame({
recommendation_key: ["Condition or Treatment columns not found!"]
})
treatment = data[data[condition_col] == "High"][treatment_col].to_list()
if treatment:
return pd.DataFrame({
recommendation_key: [f"Treatment recommended for High risk patients: {treatment}"]
})
else:
return pd.DataFrame({
recommendation_key: ["No treatment recommendation found!"]
})
# ---------------------- Medical Knowledge Base ---------------------------
class MedicalKnowledgeBase(ABC):
"""Abstract class for Medical Knowledge."""
@abstractmethod
def search_medical_info(self, query: str, pub_email: str = "") -> str:
pass
class SimpleMedicalKnowledge(MedicalKnowledgeBase):
"""Enhanced Medical Knowledge Class using OpenAI GPT-4."""
def __init__(self, nlp_model, client: OpenAI):
self.nlp = nlp_model # Using the loaded spaCy model
self.client = client # Using the OpenAI client
def search_medical_info(self, query: str, pub_email: str = "") -> str:
"""
Uses OpenAI's GPT-4 to fetch medical information based on the user's query.
"""
logger.info(f"Received medical query: {query}")
try:
# Preprocess the query (e.g., entity recognition)
doc = self.nlp(query.lower())
entities = [ent.text for ent in doc.ents]
processed_query = " ".join(entities) if entities else query.lower()
logger.info(f"Processed query: {processed_query}")
# Create a prompt for GPT-4
prompt = f"""
You are a medical assistant. Provide a comprehensive and accurate response to the following medical query:
Query: {processed_query}
Please ensure the information is clear, concise, and evidence-based.
"""
# Make the API request to OpenAI GPT-4
response = self.client.chat.completions.create(
messages=[
{"role": "system", "content": "You are a helpful medical assistant."},
{"role": "user", "content": prompt}
],
model="gpt-4", # Corrected model name
max_tokens=500,
temperature=0.7,
)
# Extract the answer from the response
answer = response.choices[0].message.content.strip() # Corrected access
logger.info("Successfully retrieved data from OpenAI GPT-4.")
# Fetch PubMed abstract related to the query
pubmed_abstract = self.fetch_pubmed_abstract(processed_query, pub_email)
# Format the response
return f"**Based on your query:** {answer}\n\n**PubMed Abstract:**\n\n{pubmed_abstract}"
except RateLimitError as e:
logger.error(f"Rate Limit Exceeded: {str(e)}")
return "Rate limit exceeded. Please try again later."
except BadRequestError as e:
logger.error(f"Bad Request: {str(e)}")
return f"Bad request: {str(e)}"
except OpenAIError as e:
logger.error(f"OpenAI API Error: {str(e)}")
return f"OpenAI API Error: {str(e)}"
except Exception as e:
logger.error(f"Medical Knowledge Search Failed: {str(e)}")
return f"Medical Knowledge Search Failed: {str(e)}"
def fetch_pubmed_abstract(self, query: str, email: str) -> str:
"""
Searches PubMed for abstracts related to the query.
"""
try:
if not email:
logger.warning("PubMed abstract retrieval skipped: Email not provided.")
return "No PubMed abstract available: Email not provided."
Entrez.email = email
handle = Entrez.esearch(db="pubmed", term=query, retmax=1, sort='relevance')
record = Entrez.read(handle)
handle.close()
logger.info(f"PubMed search for query '{query}' returned IDs: {record['IdList']}")
if record["IdList"]:
handle = Entrez.efetch(db="pubmed", id=record["IdList"][0], rettype="abstract", retmode="text")
abstract = handle.read()
handle.close()
logger.info(f"Fetched PubMed abstract for ID {record['IdList'][0]}")
return abstract
else:
logger.info(f"No PubMed abstracts found for query '{query}'.")
return "No abstracts found for this query on PubMed."
except Exception as e:
logger.error(f"Error searching PubMed: {e}")
return f"Error searching PubMed: {e}"
# ---------------------- Forecasting Engine ---------------------------
class ForecastingEngine(ABC):
"""Abstract class for forecasting."""
@abstractmethod
def predict(self, data: pd.DataFrame, **kwargs) -> pd.DataFrame:
pass
class SimpleForecasting(ForecastingEngine):
"""Simple forecasting engine."""
def predict(self, data: pd.DataFrame, period: int = 7, **kwargs) -> pd.DataFrame:
# Placeholder for actual forecasting logic
return pd.DataFrame({"forecast": [f"Forecast for the next {period} days"]})
# ---------------------- Insights and Reporting Layer ---------------------------
class AutomatedInsights:
"""Generates automated insights based on selected analyses."""
def __init__(self):
self.analyses: Dict[str, DataAnalyzer] = {
"EDA": AdvancedEDA(),
"temporal": TemporalAnalyzer(),
"distribution": DistributionVisualizer(),
"hypothesis": HypothesisTester(),
"model": LogisticRegressionTrainer()
}
def generate_insights(self, data: pd.DataFrame, analysis_names: List[str], **kwargs) -> Dict[str, Any]:
results = {}
for name in analysis_names:
analyzer = self.analyses.get(name)
if analyzer:
try:
results[name] = analyzer.invoke(data=data, **kwargs)
except Exception as e:
logger.error(f"Error in analysis '{name}': {str(e)}")
results[name] = {"error": str(e)}
else:
logger.warning(f"Analysis '{name}' not found.")
results[name] = {"error": "Analysis not found"}
return results
class Dashboard:
"""Handles the creation and display of the dashboard."""
def __init__(self):
self.layout: Dict[str, str] = {}
def add_visualisation(self, vis_name: str, vis_type: str):
self.layout[vis_name] = vis_type
def display_dashboard(self, data_dict: Dict[str, pd.DataFrame]):
st.header("Dashboard")
for vis_name, vis_type in self.layout.items():
st.subheader(vis_name)
df = data_dict.get(vis_name)
if df is not None:
if vis_type == "table":
st.table(df)
elif vis_type == "plot":
if len(df.columns) > 1:
fig = plt.figure()
sns.lineplot(data=df)
st.pyplot(fig)
else:
st.write("Please select a DataFrame with more than 1 column for plotting.")
else:
st.write("Data Not Found")
class AutomatedReports:
"""Manages automated report definitions and generation."""
def __init__(self):
self.report_definitions: Dict[str, str] = {}
def create_report_definition(self, report_name: str, definition: str):
self.report_definitions[report_name] = definition
def generate_report(self, report_name: str, data: Dict[str, pd.DataFrame]) -> Dict[str, Any]:
if report_name not in self.report_definitions:
return {"error": "Report name not found"}
report_content = {
"Report Name": report_name,
"Report Definition": self.report_definitions[report_name],
"Data": {df_name: df.to_dict() for df_name, df in data.items()}
}
return report_content
# ---------------------- Data Acquisition Layer ---------------------------
class DataSource(ABC):
"""Base class for data sources."""
@abstractmethod
def connect(self) -> None:
"""Connect to the data source."""
pass
@abstractmethod
def fetch_data(self, query: str, **kwargs) -> pd.DataFrame:
"""Fetch the data based on a specific query."""
pass
class CSVDataSource(DataSource):
"""Data source for CSV files."""
def __init__(self, file_path: io.BytesIO):
self.file_path = file_path
self.data: Optional[pd.DataFrame] = None
def connect(self):
self.data = pd.read_csv(self.file_path)
def fetch_data(self, query: str = None, **kwargs) -> pd.DataFrame:
if self.data is None:
raise Exception("No connection is made, call connect()")
return self.data
class DatabaseSource(DataSource):
"""Data source for SQL Databases."""
def __init__(self, connection_string: str, database_type: str):
self.connection_string = connection_string
self.database_type = database_type.lower()
self.connection = None
def connect(self):
if self.database_type == "sql":
# Placeholder for actual SQL connection logic
self.connection = "Connected to SQL Database"
else:
raise Exception(f"Database type '{self.database_type}' is not supported.")
def fetch_data(self, query: str, **kwargs) -> pd.DataFrame:
if self.connection is None:
raise Exception("No connection is made, call connect()")
# Placeholder for data fetching logic
return pd.DataFrame({"result": [f"Fetched data based on query: {query}"]})
class DataIngestion:
"""Handles data ingestion from various sources."""
def __init__(self):
self.sources: Dict[str, DataSource] = {}
def add_source(self, source_name: str, source: DataSource):
self.sources[source_name] = source
def ingest_data(self, source_name: str, query: str = None, **kwargs) -> pd.DataFrame:
if source_name not in self.sources:
raise Exception(f"Source '{source_name}' not found.")
source = self.sources[source_name]
source.connect()
return source.fetch_data(query, **kwargs)
class DataModel(BaseModel):
"""Defines a data model."""
name: str
kpis: List[str] = Field(default_factory=list)
dimensions: List[str] = Field(default_factory=list)
custom_calculations: Optional[Dict[str, str]] = None
relations: Optional[Dict[str, str]] = None # Example: {"table1": "table2"}
def to_json(self) -> str:
return json.dumps(self.dict())
@staticmethod
def from_json(json_str: str) -> 'DataModel':
return DataModel(**json.loads(json_str))
class DataModelling:
"""Manages data models."""
def __init__(self):
self.models: Dict[str, DataModel] = {}
def add_model(self, model: DataModel):
self.models[model.name] = model
def get_model(self, model_name: str) -> DataModel:
if model_name not in self.models:
raise Exception(f"Model '{model_name}' not found.")
return self.models[model_name]
# ---------------------- Main Streamlit Application ---------------------------
def main():
"""Main function to run the Streamlit app."""
st.title("🏥 AI-Powered Clinical Intelligence Hub")
# Initialize Session State
initialize_session_state()
# Sidebar for Data Management
with st.sidebar:
data_management_section()
# Main Content
if st.session_state.data:
col1, col2 = st.columns([1, 3])
with col1:
dataset_metadata_section()
with col2:
main_tabs_section()
def initialize_session_state():
"""Initialize necessary components in Streamlit's session state."""
if 'openai_client' not in st.session_state:
# Instantiate the OpenAI client only if it doesn't exist in session state
st.session_state.openai_client = client # The one created earlier
if 'data' not in st.session_state:
st.session_state.data = {} # Store pd.DataFrame under a name
if 'data_ingestion' not in st.session_state:
st.session_state.data_ingestion = DataIngestion()
if 'data_modelling' not in st.session_state:
st.session_state.data_modelling = DataModelling()
if 'clinical_rules' not in st.session_state:
st.session_state.clinical_rules = ClinicalRulesEngine()
if 'kpi_monitoring' not in st.session_state:
st.session_state.kpi_monitoring = ClinicalKPIMonitoring()
if 'forecasting_engine' not in st.session_state:
st.session_state.forecasting_engine = SimpleForecasting()
if 'automated_insights' not in st.session_state:
st.session_state.automated_insights = AutomatedInsights()
if 'dashboard' not in st.session_state:
st.session_state.dashboard = Dashboard()
if 'automated_reports' not in st.session_state:
st.session_state.automated_reports = AutomatedReports()
if 'diagnosis_support' not in st.session_state:
st.session_state.diagnosis_support = SimpleDiagnosis(client=st.session_state.openai_client)
if 'knowledge_base' not in st.session_state:
st.session_state.knowledge_base = SimpleMedicalKnowledge(nlp_model=nlp, client=st.session_state.openai_client)
if 'pub_email' not in st.session_state:
st.session_state.pub_email = PUB_EMAIL # Load PUB_EMAIL from environment variables
if 'treatment_recommendation' not in st.session_state:
st.session_state.treatment_recommendation = BasicTreatmentRecommendation()
def data_management_section():
"""Handles the data management section in the sidebar."""
st.header("⚙️ Data Management")
data_source_selection = st.selectbox("Select Data Source Type", ["CSV", "SQL Database"])
if data_source_selection == "CSV":
handle_csv_upload()
elif data_source_selection == "SQL Database":
handle_sql_database()
if st.button("Ingest Data"):
ingest_data_action()
def handle_csv_upload():
"""Handles CSV file uploads."""
uploaded_file = st.file_uploader("Upload research dataset (CSV)", type=["csv"])
if uploaded_file:
source_name = st.text_input("Data Source Name")
if source_name:
try:
csv_source = CSVDataSource(file_path=uploaded_file)
st.session_state.data_ingestion.add_source(source_name, csv_source)
st.success(f"Uploaded {uploaded_file.name} as '{source_name}'.")
except Exception as e:
st.error(f"Error loading dataset: {e}")
def handle_sql_database():
"""Handles SQL database connections."""
conn_str = st.text_input("Enter connection string for SQL DB")
if conn_str:
source_name = st.text_input("Data Source Name")
if source_name:
try:
sql_source = DatabaseSource(connection_string=conn_str, database_type="sql")
st.session_state.data_ingestion.add_source(source_name, sql_source)
st.success(f"Added SQL DB Source '{source_name}'.")
except Exception as e:
st.error(f"Error loading database source: {e}")
def ingest_data_action():
"""Performs data ingestion from the selected source."""
if st.session_state.data_ingestion.sources:
source_name_to_fetch = st.selectbox("Select Data Source to Ingest", list(st.session_state.data_ingestion.sources.keys()))
query = st.text_area("Optional Query to Fetch data")
if source_name_to_fetch:
with st.spinner("Ingesting data..."):
try:
data = st.session_state.data_ingestion.ingest_data(source_name_to_fetch, query)
st.session_state.data[source_name_to_fetch] = data
st.success(f"Ingested data from '{source_name_to_fetch}'.")
except Exception as e:
st.error(f"Ingestion failed: {e}")
else:
st.error("No data source added. Please add a data source.")
def dataset_metadata_section():
"""Displays metadata for the selected dataset."""
st.subheader("Dataset Metadata")
data_source_keys = list(st.session_state.data.keys())
selected_data_key = st.selectbox("Select Dataset", data_source_keys)
if selected_data_key:
data = st.session_state.data[selected_data_key]
metadata = {
"Variables": list(data.columns),
"Time Range": {
col: {
"min": data[col].min(),
"max": data[col].max()
} for col in data.select_dtypes(include='datetime').columns
},
"Size": f"{data.memory_usage().sum() / 1e6:.2f} MB"
}
st.json(metadata)
# Store the selected dataset key in session state for use in analysis
st.session_state.selected_data_key = selected_data_key
def main_tabs_section():
"""Creates and manages the main tabs in the application."""
analysis_tab, clinical_logic_tab, insights_tab, reports_tab, knowledge_tab = st.tabs([
"Data Analysis",
"Clinical Logic",
"Insights",
"Reports",
"Medical Knowledge"
])
with analysis_tab:
data_analysis_section()
with clinical_logic_tab:
clinical_logic_section()
with insights_tab:
insights_section()
with reports_tab:
reports_section()
with knowledge_tab:
medical_knowledge_section()
def data_analysis_section():
"""Handles the Data Analysis tab."""
selected_data_key = st.session_state.get('selected_data_key', None)
if not selected_data_key:
st.warning("Please select a dataset from the metadata section.")
return
data = st.session_state.data[selected_data_key]
analysis_type = st.selectbox("Select Analysis Mode", [
"Exploratory Data Analysis",
"Temporal Pattern Analysis",
"Comparative Statistics",
"Distribution Analysis",
"Train Logistic Regression Model"
])
if analysis_type == "Exploratory Data Analysis":
perform_eda(data)
elif analysis_type == "Temporal Pattern Analysis":
perform_temporal_analysis(data)
elif analysis_type == "Comparative Statistics":
perform_comparative_statistics(data)
elif analysis_type == "Distribution Analysis":
perform_distribution_analysis(data)
elif analysis_type == "Train Logistic Regression Model":
perform_logistic_regression_training(data)
def perform_eda(data: pd.DataFrame):
"""Performs Exploratory Data Analysis."""
analyzer = AdvancedEDA()
eda_result = analyzer.invoke(data=data)
st.subheader("Data Quality Report")
st.json(eda_result)
def perform_temporal_analysis(data: pd.DataFrame):
"""Performs Temporal Pattern Analysis."""
time_cols = data.select_dtypes(include='datetime').columns
num_cols = data.select_dtypes(include=np.number).columns
if len(time_cols) == 0:
st.warning("No datetime columns available for temporal analysis.")
return
time_col = st.selectbox("Select Temporal Variable", time_cols)
value_col = st.selectbox("Select Analysis Variable", num_cols)
if time_col and value_col:
analyzer = TemporalAnalyzer()
result = analyzer.invoke(data=data, time_col=time_col, value_col=value_col)
if "visualization" in result and result["visualization"]:
st.image(f"data:image/png;base64,{result['visualization']}", use_column_width=True)
st.json(result)
def perform_comparative_statistics(data: pd.DataFrame):
"""Performs Comparative Statistics."""
categorical_cols = data.select_dtypes(include=['category', 'object']).columns
numeric_cols = data.select_dtypes(include=np.number).columns
if len(categorical_cols) == 0:
st.warning("No categorical columns available for hypothesis testing.")
return
if len(numeric_cols) == 0:
st.warning("No numerical columns available for hypothesis testing.")
return
group_col = st.selectbox("Select Grouping Variable", categorical_cols)
value_col = st.selectbox("Select Metric Variable", numeric_cols)
if group_col and value_col:
analyzer = HypothesisTester()
result = analyzer.invoke(data=data, group_col=group_col, value_col=value_col)
st.subheader("Statistical Test Results")
st.json(result)
def perform_distribution_analysis(data: pd.DataFrame):
"""Performs Distribution Analysis."""
numeric_cols = data.select_dtypes(include=np.number).columns.tolist()
selected_cols = st.multiselect("Select Variables for Distribution Analysis", numeric_cols)
if selected_cols:
analyzer = DistributionVisualizer()
img_data = analyzer.invoke(data=data, columns=selected_cols)
if not img_data.startswith("Visualization Error"):
st.image(f"data:image/png;base64,{img_data}", use_column_width=True)
else:
st.error(img_data)
else:
st.info("Please select at least one numerical column to visualize.")
def perform_logistic_regression_training(data: pd.DataFrame):
"""Trains a Logistic Regression model."""
numeric_cols = data.select_dtypes(include=np.number).columns.tolist()
target_col = st.selectbox("Select Target Variable", data.columns.tolist())
selected_cols = st.multiselect("Select Feature Variables", numeric_cols)
if selected_cols and target_col:
analyzer = LogisticRegressionTrainer()
result = analyzer.invoke(data=data, target_col=target_col, columns=selected_cols)
st.subheader("Logistic Regression Model Results")
st.json(result)
else:
st.warning("Please select both target and feature variables for model training.")
def clinical_logic_section():
"""Handles the Clinical Logic tab."""
st.header("Clinical Logic")
# Clinical Rules Management
st.subheader("Clinical Rules")
rule_name = st.text_input("Enter Rule Name")
condition = st.text_area("Enter Rule Condition (use 'df' for DataFrame)",
help="Example: df['blood_pressure'] > 140")
action = st.text_area("Enter Action to be Taken on Rule Match")
severity = st.selectbox("Enter Severity for the Rule", ["low", "medium", "high"])
if st.button("Add Clinical Rule"):
if rule_name and condition and action and severity:
try:
rule = ClinicalRule(
name=rule_name,
condition=condition,
action=action,
severity=severity
)
st.session_state.clinical_rules.add_rule(rule)
st.success("Added Clinical Rule successfully.")
except Exception as e:
st.error(f"Error in rule definition: {e}")
else:
st.error("Please fill in all fields to add a clinical rule.")
# Clinical KPI Management
st.subheader("Clinical KPI Definition")
kpi_name = st.text_input("Enter KPI Name")
kpi_calculation = st.text_area("Enter KPI Calculation (use 'df' for DataFrame)",
help="Example: df['patient_count'].sum()")
threshold = st.text_input("Enter Threshold for KPI (Optional)", help="Leave blank if not applicable")
if st.button("Add Clinical KPI"):
if kpi_name and kpi_calculation:
try:
threshold_value = float(threshold) if threshold else None
kpi = ClinicalKPI(
name=kpi_name,
calculation=kpi_calculation,
threshold=threshold_value
)
st.session_state.kpi_monitoring.add_kpi(kpi)
st.success(f"Added KPI '{kpi_name}' successfully.")
except ValueError:
st.error("Threshold must be a numeric value.")
except Exception as e:
st.error(f"Error creating KPI: {e}")
else:
st.error("Please provide both KPI name and calculation.")
# Execute Clinical Rules and Calculate KPIs
selected_data_key = st.selectbox("Select Dataset for Clinical Logic", list(st.session_state.data.keys()))
if selected_data_key:
data = st.session_state.data[selected_data_key]
if st.button("Execute Clinical Rules"):
with st.spinner("Executing Clinical Rules..."):
result = st.session_state.clinical_rules.execute_rules(data)
st.json(result)
if st.button("Calculate Clinical KPIs"):
with st.spinner("Calculating Clinical KPIs..."):
result = st.session_state.kpi_monitoring.calculate_kpis(data)
st.json(result)
else:
st.warning("Please ingest data to execute clinical rules and calculate KPIs.")
def insights_section():
"""Handles the Insights tab."""
st.header("Automated Insights")
selected_data_key = st.selectbox("Select Dataset for Insights", list(st.session_state.data.keys()))
if not selected_data_key:
st.warning("Please select a dataset to generate insights.")
return
data = st.session_state.data[selected_data_key]
available_analyses = ["EDA", "temporal", "distribution", "hypothesis", "model"]
selected_analyses = st.multiselect("Select Analyses for Insights", available_analyses)
if st.button("Generate Automated Insights"):
if selected_analyses:
with st.spinner("Generating Insights..."):
results = st.session_state.automated_insights.generate_insights(
data, analysis_names=selected_analyses
)
st.json(results)
else:
st.warning("Please select at least one analysis to generate insights.")
# Diagnosis Support
st.subheader("Diagnosis Support")
target_col = st.selectbox("Select Target Variable for Diagnosis", data.columns.tolist())
numeric_cols = data.select_dtypes(include=np.number).columns.tolist()
selected_feature_cols = st.multiselect("Select Feature Variables for Diagnosis", numeric_cols)
if st.button("Generate Diagnosis"):
if target_col and selected_feature_cols:
with st.spinner("Generating Diagnosis..."):
result = st.session_state.diagnosis_support.diagnose(
data, target_col=target_col, columns=selected_feature_cols, diagnosis_key="diagnosis_result"
)
st.json(result)
else:
st.error("Please select both target and feature variables for diagnosis.")
# Treatment Recommendation
st.subheader("Treatment Recommendation")
condition_col = st.selectbox("Select Condition Column for Treatment Recommendation", data.columns.tolist())
treatment_col = st.selectbox("Select Treatment Column for Treatment Recommendation", data.columns.tolist())
if st.button("Generate Treatment Recommendation"):
if condition_col and treatment_col:
with st.spinner("Generating Treatment Recommendation..."):
result = st.session_state.treatment_recommendation.recommend(
data, condition_col=condition_col, treatment_col=treatment_col, recommendation_key="treatment_recommendation"
)
st.json(result)
else:
st.error("Please select both condition and treatment columns.")
def reports_section():
"""Handles the Reports tab."""
st.header("Automated Reports")
# Create Report Definition
st.subheader("Create Report Definition")
report_name = st.text_input("Report Name")
report_def = st.text_area("Report Definition", help="Describe the structure and content of the report.")
if st.button("Create Report Definition"):
if report_name and report_def:
st.session_state.automated_reports.create_report_definition(report_name, report_def)
st.success("Report definition created successfully.")
else:
st.error("Please provide both report name and definition.")
# Generate Report
st.subheader("Generate Report")
report_names = list(st.session_state.automated_reports.report_definitions.keys())
if report_names:
report_name_to_generate = st.selectbox("Select Report to Generate", report_names)
if st.button("Generate Report"):
with st.spinner("Generating Report..."):
report = st.session_state.automated_reports.generate_report(report_name_to_generate, st.session_state.data)
if "error" not in report:
st.header(f"Report: {report['Report Name']}")
st.markdown(f"**Definition:** {report['Report Definition']}")
for df_name, df_content in report["Data"].items():
st.subheader(f"Data: {df_name}")
st.dataframe(pd.DataFrame(df_content))
else:
st.error(report["error"])
else:
st.info("No report definitions found. Please create a report definition first.")
def medical_knowledge_section():
"""Handles the Medical Knowledge tab."""
st.header("Medical Knowledge")
query = st.text_input("Enter your medical question here:")
if st.button("Search"):
if query.strip():
with st.spinner("Searching..."):
result = st.session_state.knowledge_base.search_medical_info(
query, pub_email=st.session_state.pub_email
)
st.markdown(result)
else:
st.error("Please enter a medical question to search.")
if __name__ == "__main__":
main()