Spaces:
Build error
Build error
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,295 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from numpy import load
|
3 |
+
from numpy import expand_dims
|
4 |
+
from matplotlib import pyplot
|
5 |
+
from PIL import Image, ImageDraw, ImageFont
|
6 |
+
import numpy as np
|
7 |
+
import os
|
8 |
+
import os,sys
|
9 |
+
sys.path.insert(0,"..")
|
10 |
+
from glob import glob
|
11 |
+
import torch
|
12 |
+
import torchvision
|
13 |
+
import sys
|
14 |
+
import torch.nn.functional as F
|
15 |
+
import torchxrayvision as xrv
|
16 |
+
import pydicom as dicom
|
17 |
+
import PIL # optional
|
18 |
+
import pandas as pd
|
19 |
+
import matplotlib.pyplot as plt
|
20 |
+
import os
|
21 |
+
import cv2
|
22 |
+
import skimage
|
23 |
+
from skimage.transform import rescale, resize, downscale_local_mean
|
24 |
+
import operator
|
25 |
+
import mols2grid
|
26 |
+
import streamlit.components.v1 as components
|
27 |
+
from rdkit import Chem
|
28 |
+
from rdkit.Chem.Descriptors import ExactMolWt
|
29 |
+
from chembl_webresource_client.new_client import new_client
|
30 |
+
|
31 |
+
### Title
|
32 |
+
st.markdown("<h1 style='text-align: center;'>Chest Anomaly Identifier</h1>",unsafe_allow_html=True)
|
33 |
+
### Description
|
34 |
+
st.markdown("""<p style='text-align: center;'>The goal of this application is mainly to help doctors to interpret
|
35 |
+
Chest X-Ray Images, being able to find medical compounds in a quick way to deal with Chest's anomalies found</p>""",unsafe_allow_html=True)
|
36 |
+
|
37 |
+
### Image
|
38 |
+
st.image("doctors.jpg")
|
39 |
+
|
40 |
+
### Uploder
|
41 |
+
# st.markdown("""<p style='text-align: center;'>The goal of this application is mainly to help doctors to interpret
|
42 |
+
# Chest X-Ray Images, being able to find medical compounds in a quick way to deal with Chest's anomalies found</p>""",unsafe_allow_html=True)
|
43 |
+
uploaded_file = st.file_uploader("Choose an X-Ray image to detect anomalies of the chest (the file must be a dicom extension or jpg)")
|
44 |
+
|
45 |
+
|
46 |
+
|
47 |
+
|
48 |
+
#### Get Compounds found
|
49 |
+
@st.cache(allow_output_mutation=True)
|
50 |
+
def getdrugs(name,phase):
|
51 |
+
drug_indication = new_client.drug_indication
|
52 |
+
molecules = new_client.molecule
|
53 |
+
obj = drug_indication.filter(efo_term__icontains=name)
|
54 |
+
appdrugs = molecules.filter(molecule_chembl_id__in=[x['molecule_chembl_id'] for x in obj])
|
55 |
+
|
56 |
+
|
57 |
+
if phase!=[]:
|
58 |
+
temp = None
|
59 |
+
for ph in phase:
|
60 |
+
dftemp = pd.DataFrame.from_dict(appdrugs.filter(max_phase=int(ph)))
|
61 |
+
dftemp["phase"] = int(ph)
|
62 |
+
if isinstance(temp,pd.DataFrame):
|
63 |
+
temp= pd.concat([temp,dftemp],axis=0)
|
64 |
+
else:
|
65 |
+
temp = dftemp
|
66 |
+
|
67 |
+
df = temp
|
68 |
+
else:
|
69 |
+
df = pd.DataFrame.from_dict(appdrugs)
|
70 |
+
|
71 |
+
try:
|
72 |
+
df.dropna(subset=["molecule_properties","molecule_structures"],inplace=True)
|
73 |
+
|
74 |
+
df["smiles"] = df.molecule_structures.apply(lambda x:x["canonical_smiles"])
|
75 |
+
df["Acceptors"] = df.molecule_properties.apply(lambda x :x["hba"])
|
76 |
+
df["Donnors"] = df.molecule_properties.apply(lambda x :x["hbd"])
|
77 |
+
df["mol_weight"] = df.molecule_properties.apply(lambda x :x["mw_freebase"])
|
78 |
+
df["Logp"] = df.molecule_properties.apply(lambda x :x["cx_logp"])
|
79 |
+
|
80 |
+
subs = ["pref_name","smiles","Acceptors","Donnors","mol_weight","Logp"]
|
81 |
+
df.dropna(subset=subs,inplace=True)
|
82 |
+
df["Acceptors"] = df["Acceptors"].astype(int)
|
83 |
+
df["Donnors"] = df["Donnors"].astype(int)
|
84 |
+
df["mol_weight"] = df["mol_weight"].astype(float)
|
85 |
+
df["Logp"] = df["Logp"] .astype(float)
|
86 |
+
|
87 |
+
return df.loc[:,subs]
|
88 |
+
except:
|
89 |
+
return None
|
90 |
+
|
91 |
+
### Read Chest X Ray Image
|
92 |
+
def read_image(imgpath):
|
93 |
+
|
94 |
+
if (str(imgpath).find("jpg")!=-1) or (str(imgpath).find("png")!=-1):
|
95 |
+
|
96 |
+
# sample = Image.open("JPG_test/0c4eb1e1-b801903c-bcebe8a4-3da9cd3c-3b94a27c.jpg")
|
97 |
+
sample = Image.open(imgpath)
|
98 |
+
return np.array(sample)
|
99 |
+
if str(imgpath).find("dcm")!=-1:
|
100 |
+
img = dicom.dcmread(imgpath).pixel_array
|
101 |
+
return img
|
102 |
+
|
103 |
+
### Generate torchxrayvision model to find output probabilities
|
104 |
+
def generatemodel(xrvmodel,wts):
|
105 |
+
return xrvmodel(weights=wts)
|
106 |
+
### Transform the image to ouput some illness
|
107 |
+
def transform2(img):
|
108 |
+
input_tensor = torch.from_numpy(img).unsqueeze(0)
|
109 |
+
img = input_tensor.numpy()[0, 0, :]
|
110 |
+
img = (img / 1024.0 / 2.0) + 0.5
|
111 |
+
img = np.clip(img, 0, 1)
|
112 |
+
img = Image.fromarray(np.uint8(img * 255) , 'L')
|
113 |
+
return img
|
114 |
+
### Transform the image to test an output image
|
115 |
+
def transform(img):
|
116 |
+
|
117 |
+
img = ((img-img.min())/(img.max()-img.min())*255)
|
118 |
+
|
119 |
+
|
120 |
+
# img = (img / 1024.0 / 2.0) + 0.5
|
121 |
+
# img = np.clip(img, 0, 1)
|
122 |
+
# img = Image.fromarray(np.uint8(img * 255) , 'L')
|
123 |
+
# print(img.shape)
|
124 |
+
# img = skimage.io.imread("JPG_test/0c4eb1e1-b801903c-bcebe8a4-3da9cd3c-3b94a27c.jpg")
|
125 |
+
# print(img.max())
|
126 |
+
img = xrv.datasets.normalize(np.array(img), 255)
|
127 |
+
|
128 |
+
# Check that images are 2D arrays
|
129 |
+
if len(img.shape) > 2:
|
130 |
+
img = img[:, :, 0]
|
131 |
+
if len(img.shape) < 2:
|
132 |
+
print("error, dimension lower than 2 for image")
|
133 |
+
|
134 |
+
# Add color channel
|
135 |
+
img = img[None, :, :]
|
136 |
+
|
137 |
+
transform = torchvision.transforms.Compose([xrv.datasets.XRayCenterCrop(),
|
138 |
+
xrv.datasets.XRayResizer(224,engine="cv2")])
|
139 |
+
|
140 |
+
img = transform(img)
|
141 |
+
return img
|
142 |
+
### Returns the output probabilities of having certain illnesses anomalies
|
143 |
+
def testimage(model,img):
|
144 |
+
# with torch.no_grad():
|
145 |
+
model.eval()
|
146 |
+
out = model(torch.from_numpy(img).unsqueeze(0)).cpu()
|
147 |
+
# out = model(img).cpu()
|
148 |
+
# out = torch.sigmoid(out)
|
149 |
+
|
150 |
+
return {key:value for (key,value) in zip(model.pathologies, out.detach().numpy()[0]) if len(key)>2}
|
151 |
+
|
152 |
+
### Resize the model
|
153 |
+
def outputprob2(img,pr_model,visimage=True):
|
154 |
+
### Read an image
|
155 |
+
img = resize(img, (img.shape[0] // 2, img.shape[1] // 2),
|
156 |
+
anti_aliasing=True)
|
157 |
+
|
158 |
+
### Preprocessmodel
|
159 |
+
img_t = transform(img)
|
160 |
+
### Test an image
|
161 |
+
return testimage(pr_model,img_t)
|
162 |
+
|
163 |
+
### Pipeline since we read an image until the ouput it is generated
|
164 |
+
def outputprob(imgpath,pr_model,visimage=True):
|
165 |
+
### Read an image
|
166 |
+
img = read_image(imgpath)
|
167 |
+
if visimage:
|
168 |
+
plt.imshow(img,cmap="gray")
|
169 |
+
plt.show()
|
170 |
+
### Preprocessmodel
|
171 |
+
img_t = transform(img)
|
172 |
+
### Test an image
|
173 |
+
return testimage(pr_model,img_t)
|
174 |
+
|
175 |
+
|
176 |
+
### Error in case we do not find compounds
|
177 |
+
def error(option):
|
178 |
+
option = str(option).replace(" ","%20")
|
179 |
+
st.markdown(f"""
|
180 |
+
We have not found compounds for this illness; for more information visit this link:
|
181 |
+
[Chemble](https://www.ebi.ac.uk/chembl/g/#search_results/all/query={option})
|
182 |
+
""", unsafe_allow_html=True)
|
183 |
+
|
184 |
+
### If you insert an image
|
185 |
+
if uploaded_file is not None:
|
186 |
+
## Controller header
|
187 |
+
|
188 |
+
st.sidebar.markdown("<h1 style='text-align: center;'>Compound's filter</h1>",unsafe_allow_html=True)
|
189 |
+
## Write the compound
|
190 |
+
st.sidebar.markdown('''
|
191 |
+
<h4 style='text-align: center;'>This controller sidebar is used to filter the compounds by the following features</h4>
|
192 |
+
|
193 |
+
- Molecular weight : is the weight of a compound in grame per mol
|
194 |
+
- LogP : it measures how hydrophilic or hydrophobic a compound is
|
195 |
+
- NumDonnors : number of chemical components that are able to deliver electrons to other chemical components
|
196 |
+
- NumAcceptors : number of chemical components that are able to accept electrons to other chemical components
|
197 |
+
''',unsafe_allow_html=True)
|
198 |
+
weight_cutoff = st.sidebar.slider(
|
199 |
+
label="Molecular weight",
|
200 |
+
min_value=0,
|
201 |
+
max_value=1000,
|
202 |
+
value=500,
|
203 |
+
step=10,
|
204 |
+
help="Look for compounds that have less or equal molecular weight than the value selected"
|
205 |
+
)
|
206 |
+
logp_cutoff = st.sidebar.slider(
|
207 |
+
label="LogP",
|
208 |
+
min_value=-10,
|
209 |
+
max_value=10,
|
210 |
+
value=5,
|
211 |
+
step=1,
|
212 |
+
help="Look for compounds that have less or equal logp than the value selected"
|
213 |
+
)
|
214 |
+
NumHDonors_cutoff = st.sidebar.slider(
|
215 |
+
label="NumHDonors",
|
216 |
+
min_value=0,
|
217 |
+
max_value=15,
|
218 |
+
value=5,
|
219 |
+
step=1,
|
220 |
+
help="Look for compounds that have less or equal donors weight than the value selected"
|
221 |
+
)
|
222 |
+
NumHAcceptors_cutoff = st.sidebar.slider(
|
223 |
+
label="NumHAcceptors",
|
224 |
+
min_value=0,
|
225 |
+
max_value=20,
|
226 |
+
value=10,
|
227 |
+
step=1,
|
228 |
+
help="Look for compounds that have less or equal acceptors weight than the value selected"
|
229 |
+
)
|
230 |
+
max_phase = st.sidebar.multiselect("Phase of the compound",
|
231 |
+
['1','2', '3', '4'],
|
232 |
+
help="""
|
233 |
+
- Phase 1 : Phase I of the compound in progress
|
234 |
+
- Phase 2 : Phase II of the compound in progress
|
235 |
+
- Phase 3 : Phase III of the compound in progress
|
236 |
+
- Phase 4 : Approved compound
|
237 |
+
"""
|
238 |
+
)
|
239 |
+
|
240 |
+
#### Read an image
|
241 |
+
|
242 |
+
|
243 |
+
imgdef = read_image(uploaded_file)
|
244 |
+
|
245 |
+
### Plot the input image
|
246 |
+
fig, ax = plt.subplots()
|
247 |
+
ax.imshow(imgdef,cmap="gray")
|
248 |
+
st.pyplot(fig=fig)
|
249 |
+
# Printing the possibility of having anomalies
|
250 |
+
st.markdown("<h3 style='text-align: center;'>Possibility of anomalies</h3>",unsafe_allow_html=True)
|
251 |
+
model = generatemodel(xrv.models.DenseNet,"densenet121-res224-mimic_ch") ### MIMIC MODEL+
|
252 |
+
model.eval()
|
253 |
+
pr = outputprob2(imgdef,model)
|
254 |
+
|
255 |
+
# Sort results by the descending probability order
|
256 |
+
pr = dict( sorted(pr.items(), key=operator.itemgetter(1),reverse=True))
|
257 |
+
# Select the treatment
|
258 |
+
option = st.sidebar.selectbox('Select the treatment you believe for these illness',list(pr.keys()))
|
259 |
+
col1,col2,col3 = st.columns((1,1,1))
|
260 |
+
cnt = 1
|
261 |
+
for (key,value) in pr.items():
|
262 |
+
if cnt%3==1:
|
263 |
+
col1.metric(label=key, value=str(cnt), delta=str(value))
|
264 |
+
if cnt%3==2:
|
265 |
+
col2.metric(label=key, value=str(cnt), delta=str(value))
|
266 |
+
if cnt%3==0:
|
267 |
+
col3.metric(label=key, value=str(cnt), delta=str(value))
|
268 |
+
cnt+=1
|
269 |
+
# temp = st.expander("Compunds to take care of {}".format(key))
|
270 |
+
#### Get the compounds for the anomaly selected
|
271 |
+
df = getdrugs(option,max_phase)
|
272 |
+
st.markdown("<h3 style='text-align: center;'>Compounds for {}</h3>".format(option),unsafe_allow_html=True)
|
273 |
+
### If exists the compounds
|
274 |
+
if df is not None:
|
275 |
+
|
276 |
+
#### Filter dataframe by controllers
|
277 |
+
df_result = df[df["mol_weight"] < weight_cutoff]
|
278 |
+
df_result2 = df_result[df_result["Logp"] < logp_cutoff]
|
279 |
+
df_result3 = df_result2[df_result2["Donnors"] < NumHDonors_cutoff]
|
280 |
+
df_result4 = df_result3[df_result3["Acceptors"] < NumHAcceptors_cutoff]
|
281 |
+
|
282 |
+
|
283 |
+
|
284 |
+
if len(df_result4)==0:
|
285 |
+
|
286 |
+
error(option)
|
287 |
+
else:
|
288 |
+
raw_html = mols2grid.display(df_result, mapping={"smiles": "SMILES","pref_name":"Name","Acceptors":"Acceptors","Donnors":"Donnors","Logp":"Logp","mol_weight":"mol_weight"},
|
289 |
+
subset=["img","Name"],tooltip=["Name","Acceptors","Donnors","Logp","mol_weight"],tooltip_placement="top",tooltip_trigger="click hover")._repr_html_()
|
290 |
+
|
291 |
+
components.html(raw_html, width=900, height=900, scrolling=True)
|
292 |
+
#### We do not find compounds for the anomaly
|
293 |
+
else:
|
294 |
+
error(option)
|
295 |
+
|