Upload 4 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+pneumonia_model_clean.keras filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,26 @@
+import tensorflow as tf
+import gradio as gr
+from preprocessing import preprocess_dicom
+
+# Load model once
+model = tf.keras.models.load_model("pneumonia_model_clean.keras")
+
+def predict(file):
+    img = preprocess_dicom(file.name)
+    prob = float(model.predict(img)[0][0])
+
+    if prob >= 0.5:
+        return f"PNEUMONIA (confidence: {prob:.2f})"
+    else:
+        return f"NORMAL (confidence: {1 - prob:.2f})"
+
+demo = gr.Interface(
+    fn=predict,
+    inputs=gr.File(label="Upload Chest X-ray (DICOM)"),
+    outputs=gr.Textbox(label="Prediction"),
+    title="Pneumonia Screening (DenseNet)",
+    description="DICOM-based pneumonia screening model"
+)
+
+if __name__ == "__main__":
+    demo.launch()

pneumonia_model_clean.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6135dbfa67bf67e9b7ef53d1c3d5db05bc3f701495a7659b65ddfc490cbcbd60
+size 30710690

preprocessing.py

@@ -0,0 +1,17 @@
+import numpy as np
+import pydicom
+import cv2
+
+def load_dicom_image(path):
+    ds = pydicom.dcmread(path)
+    img = ds.pixel_array.astype(np.float32)
+    return img
+
+def resize_image(img, target_size=(224, 224)):
+    return cv2.resize(img, target_size, interpolation=cv2.INTER_LINEAR)
+
+def normalize_image(img):
+    return img / 255.0
+
+def to_3channel(img):
+    return np.stack([img, img, img], axis=-1)

requirements.txt

@@ -0,0 +1,7 @@
+tensorflow>=2.16
+keras>=3.0
+gradio
+numpy
+pydicom
+opencv-python-headless
+