Adding gender prediction

app.py

@@ -24,14 +24,21 @@ def preprocess_ecg(ecg,fs):
         pass
     return ecg
 
-def load_model(sample_frequency,recording_time, num_leads):
+def load_age_model(sample_frequency,recording_time, num_leads):
     cwd = os.getcwd()
-    weights = f"{cwd}/models/weights/model_weights_leadI.h5"
-    model = build_model((sample_frequency * recording_time, num_leads), 1)
+    weights = f"{cwd}/models/weights/model_weights_leadI_age.h5"
+    model = build_age_model((sample_frequency * recording_time, num_leads), 1)
     model.load_weights(weights)
     return model
 
 
+def load_gender_model(sample_frequency,recording_time, num_leads):
+    cwd = os.getcwd()
+    weights = f"{cwd}/models/weights/model_weights_leadI_gender.h5"
+    model = build_gender_model((sample_frequency * recording_time, num_leads), 1)
+    model.load_weights(weights)
+    return model
+
 def run(header_file, data_file):
     SAMPLE_FREQUENCY = 100
     TIME = 10
@@ -43,9 +50,11 @@ def run(header_file, data_file):
     shutil.copyfile(header_file.name, f"{demo_dir}/{hdr_basename}")
     data, fs = load_data(f"{demo_dir}/{hdr_basename.split('.')[0]}")
     ecg = preprocess_ecg(data,fs)
-    model = load_model(sample_frequency=SAMPLE_FREQUENCY,recording_time=TIME,num_leads=NUM_LEADS)
-    predicion = model.predict(np.expand_dims(ecg,0)).ravel()[0]
-    return str(round(predicion,1))
+    age_model = load_age_model(sample_frequency=SAMPLE_FREQUENCY,recording_time=TIME,num_leads=NUM_LEADS)
+    gender_model = load_gender_model(sample_frequency=SAMPLE_FREQUENCY,recording_time=TIME,num_leads=NUM_LEADS)
+    age_estimate = age_model.predict(np.expand_dims(ecg,0)).ravel()[0]
+    gender_prediction = gender_model.predict(np.expand_dims(ecg,0)).ravel()[0]
+    return str(round(age_estimate,1)), {"Male": 1- gender_prediction, "Female": gender_prediction}
 
 # Give credit to https://huggingface.co/spaces/Tej3/ECG_Classification/blob/main/app.py for interface
 
@@ -59,14 +68,14 @@ with gr.Blocks() as demo:
             header_file = gr.File(label = "header_file", file_types=[".hea"],)
             data_file = gr.File(label = "data_file", file_types=[".dat"])
         with gr.Column(scale=1):
-            output_age = gr.Textbox(label = "Predicted age")
-            #output_gender = gr.Textbox(label = "Predicted gender")
+            output_age = gr.Textbox(label = "Estimated age")
+            output_gender = gr.Label( label = "Predicted gender")
     #with gr.Row():
     #    ecg_graph = gr.Plot(label = "ECG Signal Visualisation")
     with gr.Row():    
         predict_btn = gr.Button("Predict")
         predict_btn.click(fn= run, inputs = [#pred_type, 
-                                            header_file, data_file], outputs=[output_age])
+                                            header_file, data_file], outputs=[output_age,output_gender])
     with gr.Row():    
         gr.Examples(examples=[[f"{CWD}/sample_data/ath_001.hea", f"{CWD}/sample_data/ath_001.dat"],\
     #                          [f"{CWD}/demo_data/test/00008_lr.hea", f"{CWD}/demo_data/test/00008_lr.dat", "sinusrhythmus linkstyp qrs(t) abnormal    inferiorer infarkt     alter unbest."], \

models/inception.py

@@ -70,7 +70,7 @@ def _shortcut_layer(input_tensor, out_tensor):
     return x
 
 
-def build_model(
+def build_age_model(
     input_shape: Tuple[int, int],
     nb_classes: int,
     depth: int = 6,
@@ -105,4 +105,43 @@ def build_model(
         metrics=[tf.keras.metrics.MeanSquaredError()],
     )
 
+    return model
+
+
+
+def build_gender_model(
+    input_shape: Tuple[int, int],
+    nb_classes: int,
+    depth: int = 6,
+    use_residual: bool = True,
+)-> tf.keras.models.Model:
+    """
+    Model proposed by HI Fawas et al 2019 "Finding AlexNet for Time Series Classification - InceptionTime"
+    """
+    input_layer = tf.keras.layers.Input(input_shape)
+
+    x = input_layer
+    input_res = input_layer
+
+    for d in range(depth):
+
+        x = _inception_module(x)
+
+        if use_residual and d % 3 == 2:
+            x = _shortcut_layer(input_res, x)
+            input_res = x
+
+    gap_layer = tf.keras.layers.GlobalAveragePooling1D()(x)
+
+    output_layer = tf.keras.layers.Dense(units=nb_classes, activation="sigmoid")(
+        gap_layer
+    )
+
+    model = tf.keras.models.Model(inputs=input_layer, outputs=output_layer)
+    model.compile(
+        loss=tf.keras.losses.BinaryCrossentropy(),
+        optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
+        metrics=[tf.keras.metrics.AUC(curve='ROC',name="AUROC")],
+    )
+
     return model

models/weights/model_weights_leadI_gender.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac1616b91eb8f740aaae4c44c08948b3b3d1469e628eb3c145b7e9670264f023
+size 1833768

sample_data/ath_008.hea

@@ -0,0 +1,15 @@
+ath_008 12 500 5000
+ath_008.dat 16 50000/mV 16 0 -11527 54167 0 I
+ath_008.dat 16 50000/mV 16 0 -18070 20408 0 II
+ath_008.dat 16 50000/mV 16 0 -15660 38252 0 III
+ath_008.dat 16 50000/mV 16 0 16820 59491 0 AVR
+ath_008.dat 16 50000/mV 16 0 4368 62574 0 AVL
+ath_008.dat 16 50000/mV 16 0 -17613 24434 0 AVF
+ath_008.dat 16 50000/mV 16 0 11148 20294 0 V1
+ath_008.dat 16 50000/mV 16 0 10557 25409 0 V2
+ath_008.dat 16 50000/mV 16 0 8134 4135 0 V3
+ath_008.dat 16 50000/mV 16 0 1343 20358 0 V4
+ath_008.dat 16 50000/mV 16 0 -12126 42898 0 V5
+ath_008.dat 16 50000/mV 16 0 -22817 49296 0 V6
+#SL12: Normal sinus rhythm, RSR' or QR pattern in V1 suggests right ventricular conduction delay, Borderline ECG
+#C: Normal sinus rhythm, Incomplete right bundle branch block, Normal ECG

sample_data/ath_013.hea

@@ -0,0 +1,15 @@
+ath_013 12 500 5000
+ath_013.dat 16 50000/mV 16 0 5460 51465 0 I
+ath_013.dat 16 50000/mV 16 0 -18724 34405 0 II
+ath_013.dat 16 50000/mV 16 0 -28034 29205 0 III
+ath_013.dat 16 50000/mV 16 0 9138 13588 0 AVR
+ath_013.dat 16 50000/mV 16 0 19917 20645 0 AVL
+ath_013.dat 16 50000/mV 16 0 -24408 31082 0 AVF
+ath_013.dat 16 50000/mV 16 0 21564 56921 0 V1
+ath_013.dat 16 50000/mV 16 0 20969 29226 0 V2
+ath_013.dat 16 50000/mV 16 0 18256 12273 0 V3
+ath_013.dat 16 50000/mV 16 0 -456 37253 0 V4
+ath_013.dat 16 50000/mV 16 0 -7056 1140 0 V5
+ath_013.dat 16 50000/mV 16 0 -14084 59880 0 V6
+#SL12: Marked sinus bradycardia, Right axis deviation, Abnormal ECG
+#C: Sinus bradycardia, Normal ECG