Initial Commit

app.py

@@ -83,24 +83,45 @@ custom_symptom_weightage = {
     "Family Working in Public Exposed Places": 0.5,
 }
 
-# Create a bar chart to represent symptom weightage
-# def show_symptom_weightage(symptoms, weights):
-#     plt.figure(figsize=(10, 6))
-#     # Filter out the non-selected symptoms and their corresponding weights
-#     selected_symptoms = [symptom for symptom, weight in zip(symptoms, weights) if weight > 0]
-#     selected_weights = [weight for weight in weights if weight > 0]
-#     plt.bar(selected_symptoms, selected_weights)
-#     plt.xlabel("Symptom")
-#     plt.ylabel("Weight")
-#     plt.title("Weightage/Importance of Symptoms")
-#     plt.xticks(rotation=45, ha='right')
-#     plt.tight_layout()
-#     # Save the plot as an image and return the file path
-#     file_path = tempfile.NamedTemporaryFile(suffix=".png").name
-#     plt.savefig(file_path)
-#     plt.close()
-#     return file_path
-
+def grad_cam(model, image, class_index, layer_name):
+    grad_model = tf.keras.models.Model(
+        [model.inputs], [model.get_layer(layer_name).output, model.output]
+    )
+    with tf.GradientTape() as tape:
+        conv_output, predictions = grad_model(image)
+        loss = predictions[:, class_index]
+
+    grads = tape.gradient(loss, conv_output)
+    pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
+
+    conv_output = conv_output.numpy()[0]
+    pooled_grads = pooled_grads.numpy()
+    for i in range(conv_output.shape[-1]):
+        conv_output[:, :, i] *= pooled_grads[i]
+
+    heatmap = np.mean(conv_output, axis=-1)
+    heatmap = np.maximum(heatmap, 0)
+    heatmap /= np.max(heatmap)
+
+    return heatmap
+
+# Create a Grad CAM visualization
+def show_grad_cam(image, prediction_probabilities):
+    image = image[np.newaxis, ...]
+    predicted_class_index = np.argmax(prediction_probabilities)
+    predicted_class_name = class_index_mapping[predicted_class_index]
+    
+    heatmap = grad_cam(ensemble_models['InceptionResnetV2'], image, predicted_class_index, 'conv_7b_ac')
+    heatmap = cv2.resize(heatmap, (image.shape[2], image.shape[1]))
+    heatmap = np.uint8(255 * heatmap)
+    heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
+    superimposed_img = cv2.addWeighted(cv2.cvtColor(np.squeeze(image), cv2.COLOR_RGB2BGR), 0.6, heatmap, 0.4, 0)
+    
+    # Save the Grad CAM visualization as an image and return the file path
+    file_path = tempfile.NamedTemporaryFile(suffix=".png").name
+    cv2.imwrite(file_path, superimposed_img)
+    return file_path
+    
 # Create a bar chart to represent symptom weightage
 def show_symptom_weightage(symptoms, weights):
     selected_symptoms = [symptom for symptom, weight in zip(symptoms, weights) if weight > 0]
@@ -163,10 +184,10 @@ def predict_covid(*args):
     image = args[-1]
 
     if not any(symptoms) and image is None:
-        return "COVID-19 Negative", "100%", "COVID-19 Negative", "100%", "No inputs defined", None, None
+        return "COVID-19 Negative", "100%", "COVID-19 Negative", "100%", "No inputs defined", None, None, None
 
     if any(symptoms) and image is None:
-        return "NA", "NA", "NA", "NA", "Please select your Symptoms and provide your Chest X-Ray Scan", None, None    
+        return "NA", "NA", "NA", "NA", "Please select the Symptoms and provide the Chest X-Ray Scan", None, None, None    
         
     # Prepare the input data for the text-based model
     label_encoder = LabelEncoder()
@@ -229,7 +250,8 @@ def predict_covid(*args):
     image_prediction = None
     confidence = None
     reason = None
-
+    grad_cam_img = None
+    
     if image is not None:
         is_blurry = is_scan_blurry(image)
         if is_blurry:
@@ -240,6 +262,7 @@ def predict_covid(*args):
                 "NA",
                 "Your X-Ray scan does not meet the required standards. Please ensure that your scans are not blurry, pixelated, or disfigured",
                 None,
+                None,
                 None
             )
         
@@ -262,6 +285,9 @@ def predict_covid(*args):
 
         # Visualize the heatmap
         heatmap_img = show_heat_map(image, [prediction for prediction, _ in ensemble_predictions])
+
+        # Visualize the Grad CAM for the image-based model
+        grad_cam_img = show_grad_cam(image, [prediction for prediction, _ in ensemble_predictions])        
         
         # Provide reasoning for the prediction
         if ensemble_prediction == 0 and text_prediction == "COVID-19 Positive":
@@ -294,8 +320,9 @@ def predict_covid(*args):
     confidence = 0.0 if confidence is None else confidence
     heatmap_img = "" if heatmap_img is None else heatmap_img
     symptom_weightage_img = "" if symptom_weightage_img is None else symptom_weightage_img
+    grad_cam_img = "" if grad_cam_img is None else grad_cam_img
         
-    return text_prediction, f"{text_prob_positive:.2f}%", image_prediction, f"{confidence:.2f}%", reason, heatmap_img, symptom_weightage_img
+    return text_prediction, f"{text_prob_positive:.2f}%", image_prediction, f"{confidence:.2f}%", reason, symptom_weightage_img, heatmap_img, grad_cam_img
 
 # Create the input and output components
 symptom_components = [gr_components.Checkbox(label=label) for label in column_names[:-1]]
@@ -306,8 +333,9 @@ output_components = [
     gr_components.Label(label="Prediction based on X-Ray Scan"),
     gr_components.Label(label="X-Ray Scan Confidence (%)"),
     gr_components.Textbox(label="Final Prediction"),
+    gr_components.Image(label="Weightage assigned to Symptoms"),
     gr_components.Image(label="X-Ray Prediction Heatmap"),
-    gr_components.Image(label="Weightage assigned to Symptoms")
+    gr_components.Image(label="Grad CAM Visualization"),
 ]
 
 # Create the interface and launch