Create app.py

app.py

@@ -0,0 +1,208 @@
+import os
+import io
+import numpy as np
+import torch
+import torch.nn as nn
+import torchvision.models as models
+import tensorflow as tf
+from fastapi import FastAPI, UploadFile, File
+from PIL import Image
+import cv2
+import nibabel as nib
+
+app = FastAPI(title="Parkinson + DaTscan AI API")
+
+DEVICE = "cpu"
+
+# =========================
+# LABELS (BINARY CLASS)
+# =========================
+LABELS = ["No Parkinson's", "Parkinson's Disease"]
+
+# =========================
+# LOAD KERAS MODELS
+# =========================
+def load_keras(path):
+    return tf.keras.models.load_model(path, compile=False)
+
+model_121 = load_keras("densenet121_parkinsonsDATSCAN.keras")
+model_169 = load_keras("parkinsons_densenet169DATSCAN.keras")
+model_201 = load_keras("parkinsons_densenet201DATSCAN.keras")
+
+# optional fixed model (if better)
+model_fixed = load_keras("densenet121_parkinsonsDATSCAN_fixed.keras")
+
+
+# =========================
+# 3D CNN MODEL (PyTorch)
+# =========================
+def build_3dcnn():
+    model = nn.Sequential(
+        nn.Conv3d(1, 32, 3, padding=1),
+        nn.ReLU(),
+        nn.MaxPool3d(2),
+
+        nn.Conv3d(32, 64, 3, padding=1),
+        nn.ReLU(),
+        nn.MaxPool3d(2),
+
+        nn.AdaptiveAvgPool3d((4, 4, 4)),
+        nn.Flatten(),
+        nn.Linear(64 * 4 * 4 * 4, 128),
+        nn.ReLU(),
+        nn.Linear(128, 2)
+    )
+    return model.to(DEVICE).eval()
+
+model_3dcnn = build_3dcnn()
+
+
+# =========================
+# IMAGE PREPROCESSING (2D)
+# =========================
+def preprocess_2d(image_bytes):
+    image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+    image = np.array(image)
+    image = cv2.resize(image, (128, 128))
+    image = image / 255.0
+    image = np.expand_dims(image, axis=0)
+    return image
+
+
+# =========================
+# NIFTI PREPROCESSING (3D)
+# =========================
+def preprocess_3d(file_bytes):
+    temp_path = "temp.nii"
+    with open(temp_path, "wb") as f:
+        f.write(file_bytes)
+
+    volume = nib.load(temp_path).get_fdata()
+    volume = np.squeeze(volume)
+
+    if len(volume.shape) == 2:
+        volume = np.stack([volume] * 32, axis=-1)
+
+    depth = volume.shape[2]
+
+    slices = []
+    for i in np.linspace(0, depth - 1, 32).astype(int):
+        sl = volume[:, :, i]
+        sl = cv2.resize(sl, (64, 64))
+        slices.append(sl)
+
+    vol = np.stack(slices, axis=0)
+    vol = np.expand_dims(vol, axis=0)
+    vol = np.expand_dims(vol, axis=0)
+
+    return torch.tensor(vol, dtype=torch.float32).to(DEVICE)
+
+
+# =========================
+# KERAS SINGLE PREDICT
+# =========================
+def predict_keras(model, x):
+    pred = model.predict(x, verbose=0)[0]
+    return pred
+
+
+# =========================
+# ENSEMBLE LOGIC (4 MODELS)
+# =========================
+def ensemble_predict(image_bytes):
+
+    x = preprocess_2d(image_bytes)
+
+    p1 = predict_keras(model_121, x)
+    p2 = predict_keras(model_169, x)
+    p3 = predict_keras(model_201, x)
+    p4 = predict_keras(model_fixed, x)
+
+    preds = np.array([p1, p2, p3, p4])
+
+    avg = np.mean(preds, axis=0)
+
+    cls = int(np.argmax(avg))
+    confidence = float(avg[cls] * 100)
+
+    return {
+        "prediction": LABELS[cls],
+        "class_id": cls,
+        "confidence": round(confidence, 2),
+
+        "model_confidences": {
+            "DenseNet121": round(float(np.max(p1)) * 100, 2),
+            "DenseNet169": round(float(np.max(p2)) * 100, 2),
+            "DenseNet201": round(float(np.max(p3)) * 100, 2),
+            "Fixed121": round(float(np.max(p4)) * 100, 2),
+        },
+
+        "probabilities": {
+            LABELS[i]: round(float(avg[i]) * 100, 2)
+            for i in range(2)
+        }
+    }
+
+
+# =========================
+# 3D CNN PREDICTION
+# =========================
+def predict_3d(file_bytes):
+    x = preprocess_3d(file_bytes)
+
+    with torch.no_grad():
+        out = model_3dcnn(x)
+        probs = torch.softmax(out, dim=1)[0]
+
+    cls = int(torch.argmax(probs))
+    confidence = float(probs[cls] * 100)
+
+    return {
+        "prediction": LABELS[cls],
+        "class_id": cls,
+        "confidence": round(confidence, 2),
+        "probabilities": {
+            LABELS[i]: round(float(probs[i]) * 100, 2)
+            for i in range(2)
+        }
+    }
+
+
+# =========================
+# ROUTES
+# =========================
+@app.get("/")
+def home():
+    return {
+        "status": "running",
+        "models": ["121", "169", "201", "fixed", "3dcnn"]
+    }
+
+
+@app.post("/predict")
+async def predict(file: UploadFile = File(...)):
+    image_bytes = await file.read()
+    return ensemble_predict(image_bytes)
+
+
+@app.post("/predict/121")
+async def p121(file: UploadFile = File(...)):
+    x = preprocess_2d(await file.read())
+    return {"model": "121", "prob": predict_keras(model_121, x).tolist()}
+
+
+@app.post("/predict/169")
+async def p169(file: UploadFile = File(...)):
+    x = preprocess_2d(await file.read())
+    return {"model": "169", "prob": predict_keras(model_169, x).tolist()}
+
+
+@app.post("/predict/201")
+async def p201(file: UploadFile = File(...)):
+    x = preprocess_2d(await file.read())
+    return {"model": "201", "prob": predict_keras(model_201, x).tolist()}
+
+
+@app.post("/predict/3d")
+async def p3d(file: UploadFile = File(...)):
+    return predict_3d(await file.read())