Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
import tensorflow as tf
|
4 |
+
from tensorflow import keras
|
5 |
+
import tensorflow_hub as hub
|
6 |
+
from PIL import Image
|
7 |
+
|
8 |
+
# Load models
|
9 |
+
model_initial = keras.models.load_model(
|
10 |
+
"models/initial_model.h5", custom_objects={'KerasLayer': hub.KerasLayer}
|
11 |
+
)
|
12 |
+
model_tumor = keras.models.load_model(
|
13 |
+
"models/model_tumor.h5", custom_objects={'KerasLayer': hub.KerasLayer}
|
14 |
+
)
|
15 |
+
model_stroke = keras.models.load_model(
|
16 |
+
"models/model_stroke.h5", custom_objects={'KerasLayer': hub.KerasLayer}
|
17 |
+
)
|
18 |
+
model_alzheimer = keras.models.load_model(
|
19 |
+
"models/model_alzheimer.h5", custom_objects={'KerasLayer': hub.KerasLayer}
|
20 |
+
)
|
21 |
+
|
22 |
+
class CombinedDiseaseModel(tf.keras.Model):
|
23 |
+
def __init__(self, model_initial, model_alzheimer, model_tumor, model_stroke):
|
24 |
+
super(CombinedDiseaseModel, self).__init__()
|
25 |
+
self.model_initial = model_initial
|
26 |
+
self.model_alzheimer = model_alzheimer
|
27 |
+
self.model_tumor = model_tumor
|
28 |
+
self.model_stroke = model_stroke
|
29 |
+
self.disease_labels = ["Alzheimer's", 'No Disease', 'Stroke', 'Tumor']
|
30 |
+
|
31 |
+
self.sub_models = {
|
32 |
+
"Alzheimer's": model_alzheimer,
|
33 |
+
'Tumor': model_tumor,
|
34 |
+
'Stroke': model_stroke
|
35 |
+
}
|
36 |
+
|
37 |
+
def call(self, inputs):
|
38 |
+
initial_probs = self.model_initial(inputs, training=False)
|
39 |
+
main_disease_idx = tf.argmax(initial_probs, axis=1)
|
40 |
+
main_disease = self.disease_labels[main_disease_idx[0].numpy()]
|
41 |
+
main_disease_prob = initial_probs[0, main_disease_idx[0]].numpy()
|
42 |
+
|
43 |
+
if main_disease == 'No Disease':
|
44 |
+
sub_category = "No Disease"
|
45 |
+
sub_category_prob = main_disease_prob
|
46 |
+
else:
|
47 |
+
sub_model = self.sub_models[main_disease]
|
48 |
+
sub_category_pred = sub_model(inputs, training=False)
|
49 |
+
sub_category = tf.argmax(sub_category_pred, axis=1).numpy()[0]
|
50 |
+
sub_category_prob = sub_category_pred[0, sub_category].numpy()
|
51 |
+
|
52 |
+
if main_disease == "Alzheimer's":
|
53 |
+
sub_category_label = ['Very Mild', 'Mild', 'Moderate']
|
54 |
+
elif main_disease == 'Tumor':
|
55 |
+
sub_category_label = ['Glioma', 'Meningioma', 'Pituitary']
|
56 |
+
elif main_disease == 'Stroke':
|
57 |
+
sub_category_label = ['Ischemic', 'Hemorrhagic']
|
58 |
+
|
59 |
+
sub_category = sub_category_label[sub_category]
|
60 |
+
|
61 |
+
return f"The MRI image shows {main_disease} with a probability of {main_disease_prob*100:.2f}%.\nThe subcategory of {main_disease} is {sub_category} with a probability of {sub_category_prob*100:.2f}%."
|
62 |
+
|
63 |
+
|
64 |
+
# Initialize the combined model
|
65 |
+
cnn_model = CombinedDiseaseModel(
|
66 |
+
model_initial=model_initial,
|
67 |
+
model_alzheimer=model_alzheimer,
|
68 |
+
model_tumor=model_tumor,
|
69 |
+
model_stroke=model_stroke
|
70 |
+
)
|
71 |
+
|
72 |
+
|
73 |
+
def process_image(image):
|
74 |
+
image = image.resize((256, 256))
|
75 |
+
image.convert("RGB")
|
76 |
+
image_array = np.array(image) / 255.0
|
77 |
+
image_array = np.expand_dims(image_array, axis=0)
|
78 |
+
predictions = cnn_model(image_array)
|
79 |
+
return predictions
|
80 |
+
|
81 |
+
|
82 |
+
def gradio_interface(patient_info, query_type, image):
|
83 |
+
if image is not None:
|
84 |
+
image_response = process_image(image)
|
85 |
+
response = f"Patient Info: {patient_info}\nQuery Type: {query_type}\n{image_response}"
|
86 |
+
return response
|
87 |
+
else:
|
88 |
+
return "Please upload an image."
|
89 |
+
|
90 |
+
|
91 |
+
# Create Gradio app
|
92 |
+
iface = gr.Interface(
|
93 |
+
fn=gradio_interface,
|
94 |
+
inputs=[
|
95 |
+
gr.Textbox(
|
96 |
+
label="Patient Information",
|
97 |
+
placeholder="Enter patient details here...",
|
98 |
+
lines=5,
|
99 |
+
max_lines=10
|
100 |
+
),
|
101 |
+
gr.Textbox(
|
102 |
+
label="Query Type"
|
103 |
+
),
|
104 |
+
gr.Image(
|
105 |
+
type="pil",
|
106 |
+
label="Upload an Image",
|
107 |
+
)
|
108 |
+
],
|
109 |
+
outputs=gr.Textbox(label="Response", placeholder="The response will appear here..."),
|
110 |
+
title="Medical Diagnosis with MRI",
|
111 |
+
description="Upload MRI images and provide patient information for diagnosis.",
|
112 |
+
)
|
113 |
+
|
114 |
+
iface.launch()
|