Update app.py

app.py

@@ -17,11 +17,38 @@ model = Conv5_FC3(input_size= [
 model.load_state_dict(checkpoint_state["model"])
 model.eval()
 
+def preprocess_nii(nii_file):
+    # Load NIfTI file
+    img = nib.load(nii_file)
+    data = img.get_fdata()  # numpy array (float64)
+
+    # Normalize intensities
+    data = (data - np.mean(data)) / (np.std(data) + 1e-8)
+
+    # Convert to tensor
+    tensor = torch.tensor(data, dtype=torch.float32).unsqueeze(0).unsqueeze(0)  
+    # Shape: [1, 1, D, H, W]
+
+    # Resize or pad to expected input shape
+    target_shape = (1, 1, 169, 208, 179)
+    tensor = F.interpolate(tensor, size=target_shape[2:], mode="trilinear", align_corners=False)
+
+    return tensor
+    
 def predict(input_image):
+    x = preprocess_nii(input_image)
     with torch.no_grad():
-        output = model(input_image.unsqueeze(0).to(model.device))
+        output = model(x)
     output = output.squeeze(0).cpu().float()
     return output[0]
 
-demo = gr.Interface(fn=predict, inputs="image", outputs="label")
+# Gradio app: file upload instead of image
+demo = gr.Interface(
+    fn=predict,
+    inputs=gr.File(type="file", label=".nii.gz MRI upload"),
+    outputs="label",
+    title="ClinicaDL MRI Classifier",
+    description="Upload a .nii.gz file to get the model's prediction."
+)
+
 demo.launch()