Spaces:
Build error
Build error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.environ['TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD'] = '1'
|
3 |
+
|
4 |
+
import numpy as np
|
5 |
+
import torch
|
6 |
+
import gradio as gr
|
7 |
+
from PIL import Image
|
8 |
+
import re
|
9 |
+
import fitz # PyMuPDF
|
10 |
+
from torchvision import transforms
|
11 |
+
|
12 |
+
# ==============================
|
13 |
+
# CONFIG
|
14 |
+
# ==============================
|
15 |
+
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
16 |
+
|
17 |
+
# Your 23 custom labels
|
18 |
+
LABELS = [
|
19 |
+
"Pneumonia", "Tuberculosis", "Lung Cancer", "Pulmonary Fibrosis",
|
20 |
+
"COPD", "COVID-19 lung infection", "Pleural Effusion", "Atelectasis",
|
21 |
+
"Cardiomegaly", "Rib Fracture", "Spinal Fracture", "Osteoporosis",
|
22 |
+
"Arthritis", "Bone Tumor", "Scoliosis", "Dental Caries",
|
23 |
+
"Impacted Tooth", "Jaw Fracture", "Intestinal Obstruction",
|
24 |
+
"Kidney Stone", "Gallstone", "Foreign Object"
|
25 |
+
]
|
26 |
+
|
27 |
+
# Disease info (examples, can be extended)
|
28 |
+
DISEASE_INFO = {
|
29 |
+
"Pneumonia": {
|
30 |
+
"description": "Infection of the lung causing inflammation.",
|
31 |
+
"cause": "Bacterial, viral, or fungal pathogens.",
|
32 |
+
"recommendation": "Consult a doctor, antibiotics/antivirals if confirmed."
|
33 |
+
},
|
34 |
+
"Tuberculosis": {
|
35 |
+
"description": "Bacterial infection by Mycobacterium tuberculosis.",
|
36 |
+
"cause": "Airborne spread from infected person.",
|
37 |
+
"recommendation": "Seek TB specialist, long-term antibiotics."
|
38 |
+
},
|
39 |
+
# ... add info for all 23 labels ...
|
40 |
+
}
|
41 |
+
|
42 |
+
def get_disease_info(label):
|
43 |
+
d = DISEASE_INFO.get(label)
|
44 |
+
if d:
|
45 |
+
return (
|
46 |
+
f"<b>{label}</b>: {d['description']}<br>"
|
47 |
+
f"<b>Possible Causes:</b> {d['cause']}<br>"
|
48 |
+
f"<b>Recommendation:</b> {d['recommendation']}"
|
49 |
+
)
|
50 |
+
return f"<b>{label}</b>: No extra info available. Please consult a radiologist."
|
51 |
+
|
52 |
+
# ==============================
|
53 |
+
# MODEL LOADING
|
54 |
+
# ==============================
|
55 |
+
# Replace this with your own Hugging Face model ID
|
56 |
+
MODEL_ID = "your-username/your-xray-multilabel-model"
|
57 |
+
|
58 |
+
# Example: assume model is a torch.nn.Module with sigmoid output
|
59 |
+
# The model should accept [1, 3, 224, 224] tensor and output [1, len(LABELS)]
|
60 |
+
model = torch.hub.load("pytorch/vision", "resnet18", pretrained=False)
|
61 |
+
model.fc = torch.nn.Linear(model.fc.in_features, len(LABELS))
|
62 |
+
model.load_state_dict(torch.load("model_weights.pth", map_location=DEVICE))
|
63 |
+
model.to(DEVICE).eval()
|
64 |
+
|
65 |
+
# ==============================
|
66 |
+
# IMAGE PREPROCESSING
|
67 |
+
# ==============================
|
68 |
+
transform = transforms.Compose([
|
69 |
+
transforms.Resize((224, 224)),
|
70 |
+
transforms.Grayscale(num_output_channels=3),
|
71 |
+
transforms.ToTensor(),
|
72 |
+
transforms.Normalize(mean=[0.5]*3, std=[0.5]*3)
|
73 |
+
])
|
74 |
+
|
75 |
+
def preprocess_image(img: Image.Image) -> torch.Tensor:
|
76 |
+
return transform(img).unsqueeze(0).to(DEVICE)
|
77 |
+
|
78 |
+
# ==============================
|
79 |
+
# XRAY ANALYSIS
|
80 |
+
# ==============================
|
81 |
+
def analyse_xray(img: Image.Image):
|
82 |
+
if img is None:
|
83 |
+
return "Please upload an X-ray image.", None
|
84 |
+
try:
|
85 |
+
x = preprocess_image(img)
|
86 |
+
with torch.no_grad():
|
87 |
+
outputs = model(x)
|
88 |
+
probs = torch.sigmoid(outputs)[0] * 100
|
89 |
+
topk = torch.topk(probs, 5)
|
90 |
+
|
91 |
+
html = "<h3>π©Ί Top 5 Predictions</h3><table border='1'><tr><th>Condition</th><th>Confidence</th><th>Details</th></tr>"
|
92 |
+
for idx in topk.indices:
|
93 |
+
label = LABELS[idx]
|
94 |
+
html += (
|
95 |
+
f"<tr><td>{label}</td>"
|
96 |
+
f"<td>{probs[idx]:.1f}%</td>"
|
97 |
+
f"<td>{get_disease_info(label)}</td></tr>"
|
98 |
+
)
|
99 |
+
html += "</table>"
|
100 |
+
return html, img.resize((224, 224))
|
101 |
+
except Exception as e:
|
102 |
+
return f"Error processing image: {str(e)}", None
|
103 |
+
|
104 |
+
# ==============================
|
105 |
+
# PDF REPORT ANALYSIS
|
106 |
+
# ==============================
|
107 |
+
def analyse_report(file):
|
108 |
+
if file is None:
|
109 |
+
return "Please upload a PDF report."
|
110 |
+
try:
|
111 |
+
doc = fitz.open(file.name)
|
112 |
+
text = "\n".join(page.get_text() for page in doc)
|
113 |
+
doc.close()
|
114 |
+
found = []
|
115 |
+
for label in LABELS:
|
116 |
+
if re.search(rf"\b{label.lower()}\b", text.lower()):
|
117 |
+
found.append(label)
|
118 |
+
if found:
|
119 |
+
html = "<h3>π Findings Detected in Report:</h3><ul>"
|
120 |
+
for label in found:
|
121 |
+
html += f"<li>{get_disease_info(label)}</li>"
|
122 |
+
html += "</ul>"
|
123 |
+
else:
|
124 |
+
html = "<p>No known conditions detected from report text.</p>"
|
125 |
+
return html
|
126 |
+
except Exception as e:
|
127 |
+
return f"Error processing PDF: {str(e)}"
|
128 |
+
|
129 |
+
# ==============================
|
130 |
+
# GRADIO UI
|
131 |
+
# ==============================
|
132 |
+
with gr.Blocks(title="π©» Multi-Xray AI") as demo:
|
133 |
+
gr.Markdown(
|
134 |
+
"## π©» Multi-Xray AI\n"
|
135 |
+
"Detect and classify 23 different medical conditions from various X-ray types."
|
136 |
+
)
|
137 |
+
with gr.Tabs():
|
138 |
+
with gr.Tab("π X-ray Analysis"):
|
139 |
+
x_input = gr.Image(label="Upload X-ray", type="pil")
|
140 |
+
x_out_html = gr.HTML()
|
141 |
+
x_out_image = gr.Image(label="Resized (224x224)")
|
142 |
+
analyze_btn = gr.Button("Analyze X-ray")
|
143 |
+
clear_btn = gr.Button("Clear")
|
144 |
+
analyze_btn.click(analyse_xray, inputs=x_input, outputs=[x_out_html, x_out_image])
|
145 |
+
clear_btn.click(lambda: (None, "", None), None, [x_input, x_out_html, x_out_image])
|
146 |
+
|
147 |
+
with gr.Tab("π PDF Report Analysis"):
|
148 |
+
pdf_input = gr.File(file_types=[".pdf"], label="Upload PDF Medical Report")
|
149 |
+
pdf_output = gr.HTML()
|
150 |
+
analyze_pdf_btn = gr.Button("Analyze Report")
|
151 |
+
clear_pdf_btn = gr.Button("Clear")
|
152 |
+
analyze_pdf_btn.click(analyse_report, inputs=pdf_input, outputs=pdf_output)
|
153 |
+
clear_pdf_btn.click(lambda: (None, ""), None, [pdf_input, pdf_output])
|
154 |
+
|
155 |
+
if __name__ == "__main__":
|
156 |
+
demo.launch(server_port=int(os.getenv("PORT", 7860)), show_error=True)
|