Spaces:
Build error
Build error
Formatting and examples
Browse files- app.py +78 -34
- examples/BRATS_486.nii.gz +3 -0
- examples/log.csv +0 -2
app.py
CHANGED
@@ -13,22 +13,26 @@ from monai.transforms import (
|
|
13 |
ScaleIntensityd,
|
14 |
)
|
15 |
|
|
|
16 |
BUNDLE_NAME = 'spleen_ct_segmentation_v0.1.0'
|
17 |
BUNDLE_PATH = os.path.join(torch.hub.get_dir(), 'bundle', BUNDLE_NAME)
|
18 |
|
19 |
-
|
|
|
20 |
description = """
|
21 |
-
## Brain Tumor Segmentation π§
|
22 |
-
A pre-trained model for volumetric (3D) segmentation of brain tumor subregions from multimodal MRIs based on BraTS 2018 data.
|
23 |
-
|
24 |
## To run π
|
25 |
|
26 |
-
Upload a image file in the format: 4 channel MRI (4 aligned MRIs T1c, T1, T2, FLAIR at 1x1x1 mm)
|
|
|
|
|
|
|
27 |
|
28 |
## Disclaimer β οΈ
|
29 |
|
30 |
This is an example, not to be used for diagnostic purposes.
|
|
|
31 |
|
|
|
32 |
## References π
|
33 |
|
34 |
1. Myronenko, Andriy. "3D MRI brain tumor segmentation using autoencoder regularization." International MICCAI Brainlesion Workshop. Springer, Cham, 2018. https://arxiv.org/abs/1810.11654.
|
@@ -36,8 +40,12 @@ This is an example, not to be used for diagnostic purposes.
|
|
36 |
3. Bakas S, et al. "Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features", Nature Scientific Data, 4:170117 (2017) DOI:10.1038/sdata.2017.117
|
37 |
"""
|
38 |
|
39 |
-
|
|
|
|
|
|
|
40 |
|
|
|
41 |
model, _, _ = bundle.load(
|
42 |
name = BUNDLE_NAME,
|
43 |
source = 'huggingface_hub',
|
@@ -45,10 +53,13 @@ model, _, _ = bundle.load(
|
|
45 |
load_ts_module=True,
|
46 |
)
|
47 |
|
|
|
48 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
49 |
|
|
|
50 |
parser = bundle.load_bundle_config(BUNDLE_PATH, 'inference.json')
|
51 |
|
|
|
52 |
preproc_transforms = Compose(
|
53 |
[
|
54 |
LoadImaged(keys=["image"]),
|
@@ -57,7 +68,14 @@ preproc_transforms = Compose(
|
|
57 |
NormalizeIntensityd(keys="image", nonzero=True, channel_wise=True),
|
58 |
]
|
59 |
)
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
post_transforms = Compose(
|
62 |
[
|
63 |
Activationsd(keys='pred', sigmoid=True),
|
@@ -66,10 +84,13 @@ post_transforms = Compose(
|
|
66 |
]
|
67 |
)
|
68 |
|
|
|
69 |
def predict(input_file, z_axis, model=model, device=device):
|
|
|
70 |
data = {'image': [input_file.name]}
|
71 |
data = preproc_transforms(data)
|
72 |
|
|
|
73 |
model.to(device)
|
74 |
model.eval()
|
75 |
with torch.no_grad():
|
@@ -77,34 +98,57 @@ def predict(input_file, z_axis, model=model, device=device):
|
|
77 |
data['pred'] = inferer(inputs=inputs[None,...], network=model)
|
78 |
data = post_transforms(data)
|
79 |
|
80 |
-
|
81 |
-
|
|
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
|
|
87 |
|
88 |
-
|
89 |
-
|
90 |
-
|
|
|
91 |
|
92 |
-
return
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
109 |
|
110 |
-
|
|
|
|
13 |
ScaleIntensityd,
|
14 |
)
|
15 |
|
16 |
+
# Define the bundle name and path for downloading
|
17 |
BUNDLE_NAME = 'spleen_ct_segmentation_v0.1.0'
|
18 |
BUNDLE_PATH = os.path.join(torch.hub.get_dir(), 'bundle', BUNDLE_NAME)
|
19 |
|
20 |
+
# Title and description
|
21 |
+
title = "# Segment Brain Tumors with MONAI! π§ "
|
22 |
description = """
|
|
|
|
|
|
|
23 |
## To run π
|
24 |
|
25 |
+
Upload a image file in the format: 4 channel MRI (4 aligned MRIs T1c, T1, T2, FLAIR at 1x1x1 mm), or try out one of the examples below!
|
26 |
+
If you want to see a different slice, update the slider and click the button.
|
27 |
+
|
28 |
+
More details on the model can be found [here!](https://huggingface.co/katielink/brats_mri_segmentation_v0.1.0)
|
29 |
|
30 |
## Disclaimer β οΈ
|
31 |
|
32 |
This is an example, not to be used for diagnostic purposes.
|
33 |
+
"""
|
34 |
|
35 |
+
references = """
|
36 |
## References π
|
37 |
|
38 |
1. Myronenko, Andriy. "3D MRI brain tumor segmentation using autoencoder regularization." International MICCAI Brainlesion Workshop. Springer, Cham, 2018. https://arxiv.org/abs/1810.11654.
|
|
|
40 |
3. Bakas S, et al. "Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features", Nature Scientific Data, 4:170117 (2017) DOI:10.1038/sdata.2017.117
|
41 |
"""
|
42 |
|
43 |
+
examples = [
|
44 |
+
['examples/BRATS_485.nii.gz', 100],
|
45 |
+
['examples/BRATS_', 100]
|
46 |
+
]
|
47 |
|
48 |
+
# Load the MONAI pretrained model from Hugging Face Hub
|
49 |
model, _, _ = bundle.load(
|
50 |
name = BUNDLE_NAME,
|
51 |
source = 'huggingface_hub',
|
|
|
53 |
load_ts_module=True,
|
54 |
)
|
55 |
|
56 |
+
# Use GPU if available
|
57 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
58 |
|
59 |
+
# Load the parser from the MONAI bundle's inference config
|
60 |
parser = bundle.load_bundle_config(BUNDLE_PATH, 'inference.json')
|
61 |
|
62 |
+
# Compose the preprocessing transforms
|
63 |
preproc_transforms = Compose(
|
64 |
[
|
65 |
LoadImaged(keys=["image"]),
|
|
|
68 |
NormalizeIntensityd(keys="image", nonzero=True, channel_wise=True),
|
69 |
]
|
70 |
)
|
71 |
+
|
72 |
+
# Get the inferer from the bundle's inference config
|
73 |
+
inferer = parser.get_parsed_content(
|
74 |
+
'inferer',
|
75 |
+
lazy=True, eval_expr=True, instantiate=True
|
76 |
+
)
|
77 |
+
|
78 |
+
# Compose the postprocessing transforms
|
79 |
post_transforms = Compose(
|
80 |
[
|
81 |
Activationsd(keys='pred', sigmoid=True),
|
|
|
84 |
]
|
85 |
)
|
86 |
|
87 |
+
# Define the predict function for the demo
|
88 |
def predict(input_file, z_axis, model=model, device=device):
|
89 |
+
# Load and process data in MONAI format
|
90 |
data = {'image': [input_file.name]}
|
91 |
data = preproc_transforms(data)
|
92 |
|
93 |
+
# Run inference and post-process predicted labels
|
94 |
model.to(device)
|
95 |
model.eval()
|
96 |
with torch.no_grad():
|
|
|
98 |
data['pred'] = inferer(inputs=inputs[None,...], network=model)
|
99 |
data = post_transforms(data)
|
100 |
|
101 |
+
# Convert tensors back to numpy arrays
|
102 |
+
data['image'] = data['image'].numpy()
|
103 |
+
data['pred'] = data['pred'].cpu().detach().numpy()
|
104 |
|
105 |
+
# Magnetic resonance imaging sequences
|
106 |
+
t1c = data['image'][0, :, :, z_axis] # T1-weighted, post contrast
|
107 |
+
t1 = data['image'][1, :, :, z_axis] # T1-weighted, pre contrast
|
108 |
+
t2 = data['image'][2, :, :, z_axis] # T2-weighted
|
109 |
+
flair = data['image'][3, :, :, z_axis] # FLAIR
|
110 |
|
111 |
+
# BraTS labels
|
112 |
+
tc = data['pred'][0, 0, :, :, z_axis] # Tumor core
|
113 |
+
wt = data['pred'][0, 1, :, :, z_axis] # Whole tumor
|
114 |
+
et = data['pred'][0, 2, :, :, z_axis] # Enhancing tumor
|
115 |
|
116 |
+
return [t1c, t1, t2, flair], [tc, wt, et]
|
117 |
+
|
118 |
+
# Use blocks to set up a more complex demo
|
119 |
+
with gr.Blocks() as demo:
|
120 |
+
|
121 |
+
# Show title and description
|
122 |
+
gr.Markdown(title)
|
123 |
+
gr.Markdown(description)
|
124 |
+
|
125 |
+
# Get the input file and slice slider as inputs
|
126 |
+
input_file = gr.File(label='input file')
|
127 |
+
z_axis = gr.Slider(0, 200, label='z-axis', value=50)
|
128 |
+
|
129 |
+
# Show the button with custom label
|
130 |
+
button = gr.Button("Segment Tumor!")
|
131 |
+
|
132 |
+
# Show examples for the user to try
|
133 |
+
gr.Markdown("Try some examples from MONAI's Decathlon Dataset:")
|
134 |
+
examples = gr.Examples(
|
135 |
+
examples=examples,
|
136 |
+
inputs=[gr.File(), gr.Slider()]
|
137 |
+
)
|
138 |
+
|
139 |
+
# Show the input image with different MR sequences
|
140 |
+
input_image = gr.Gallery(label='input MRI sequences (T1+, T1, T2, FLAIR)')
|
141 |
+
output_segmentation = gr.Gallery(label='output segmentations (TC, EC, WT)')
|
142 |
+
|
143 |
+
# Run prediction on button click
|
144 |
+
button.click(
|
145 |
+
predict,
|
146 |
+
inputs=[input_file, z_axis],
|
147 |
+
outputs=[input_image, output_segmentation]
|
148 |
+
)
|
149 |
+
|
150 |
+
# Show references at the bottom of the demo
|
151 |
+
gr.Markdown(references)
|
152 |
|
153 |
+
# Launch the demo
|
154 |
+
demo.launch()
|
examples/BRATS_486.nii.gz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e8957d67a50b39afd8210f3ca51a20c77ef1c92642800f91b50f16b27778f2b2
|
3 |
+
size 11111216
|
examples/log.csv
DELETED
@@ -1,2 +0,0 @@
|
|
1 |
-
input_file
|
2 |
-
BRATS_485.nii.gz
|
|
|
|
|
|