main.py

# ==========================================
# IMPORTS
# ==========================================
from fastapi import FastAPI, File, UploadFile
from pydantic import BaseModel
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
import pandas as pd
import numpy as np
import joblib
import tensorflow as tf
from PIL import Image
import io

# ==========================================
# INITIALIZE APP
# ==========================================
app = FastAPI(
    title="Stroke Detection API (CT + Clinical Data)",
    description="Deep Learning (DenseNet121) + ML Logistic Regression",
    version="2.0"
)

# CORS setup
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# ==========================================
# LOAD MODELS
# ==========================================
logistic_model = joblib.load("stroke_logistic_regression_model.pkl")
preprocessor = joblib.load("preprocessor.pkl")
cnn_model = tf.keras.models.load_model("dense_final_finetuned.keras")

IMG_SIZE = (224, 224)

# ==========================================
# Pydantic Models
# ==========================================
class StrokeInput(BaseModel):
    age: float
    avg_glucose_level: float
    bmi: float
    hypertension: int
    heart_disease: int
    gender: str
    ever_married: str
    Residence_type: str
    work_type: str
    smoking_status: str

class StrokeOutput(BaseModel):
    stroke_prediction: int
    stroke_probability: float

# ==========================================
# HELPER FUNCTIONS
# ==========================================
def preprocess_image(image_bytes):
    img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
    img = img.resize(IMG_SIZE)
    img_arr = tf.keras.preprocessing.image.img_to_array(img) / 255.0
    img_arr = np.expand_dims(img_arr, axis=0)
    return img_arr

def predict_image_cnn(img_tensor, threshold=0.5):
    prob = cnn_model.predict(img_tensor)[0][0]
    label = "Stroke Detected" if prob >= threshold else "Normal Brain"
    return label, float(prob)

# ==========================================
# ENDPOINT 1: STRUCTURED DATA ML MODEL
# ==========================================
@app.post("/stroke-predict-struct", response_model=StrokeOutput)
def predict_stroke_struct(data: StrokeInput):

    df = pd.DataFrame([data.dict()])

    # Feature Engineering
    df['age_glu_interaction'] = df['age'] * df['avg_glucose_level']
    df['ht_hd_score'] = df['hypertension'] + df['heart_disease']

    df['work_type_simplified'] = df['work_type'].replace({
        'children': 'No_Work',
        'Never_worked': 'No_Work',
        'Private': 'Private',
        'Self-employed': 'Self_Employed',
        'Govt_job': 'Govt'
    })

    df['smoke_simplified'] = df['smoking_status'].replace({
        'formerly smoked': 'Former',
        'never smoked': 'Never',
        'smokes': 'Smoker',
        'Unknown': 'Unknown'
    })

    df['glucose_bin'] = pd.cut(
        df['avg_glucose_level'],
        bins=[0, 100, 140, np.inf],
        labels=['Normal', 'Prediabetic', 'High']
    )

    selected_features = [
        'age','avg_glucose_level','bmi','age_glu_interaction',
        'hypertension','heart_disease','ht_hd_score',
        'gender','ever_married','Residence_type',
        'work_type_simplified','smoke_simplified','glucose_bin'
    ]

    df = df[selected_features]

    processed = preprocessor.transform(df)
    prob = logistic_model.predict_proba(processed)[0][1]
    pred = logistic_model.predict(processed)[0]

    return {
        "stroke_prediction": int(pred),
        "stroke_probability": float(round(prob, 4))
    }

# ==========================================
# ENDPOINT 2: MRI IMAGE CNN MODEL
# ==========================================
@app.post("/stroke-predict-image")
async def predict_stroke_image(file: UploadFile = File(...)):
    image_bytes = await file.read()
    img_tensor = preprocess_image(image_bytes)

    label, prob = predict_image_cnn(img_tensor)

    return JSONResponse({
        "filename": file.filename,
        "prediction": label,
        "confidence_score": float(round(prob, 4))
    })