Update app.py
Browse files
app.py
CHANGED
@@ -60,7 +60,7 @@ def extract_priority_pages(file_path: str, max_pages: int = 20) -> str:
|
|
60 |
text_chunks.append(f"=== Page {i+1} ===\n{(page.extract_text() or '').strip()}")
|
61 |
for i, page in enumerate(pdf.pages[3:max_pages], start=4):
|
62 |
page_text = page.extract_text() or ""
|
63 |
-
if any(re.search(rf'
|
64 |
text_chunks.append(f"=== Page {i} ===\n{page_text.strip()}")
|
65 |
return "\n\n".join(text_chunks)
|
66 |
except Exception as e:
|
@@ -130,7 +130,7 @@ def init_agent():
|
|
130 |
enable_checker=True,
|
131 |
step_rag_num=8,
|
132 |
seed=100,
|
133 |
-
additional_default_tools=[]
|
134 |
)
|
135 |
agent.init_model()
|
136 |
return agent
|
@@ -141,26 +141,30 @@ def create_ui(agent: TxAgent):
|
|
141 |
gr.Markdown("<h3 style='text-align: center;'>Identify potential oversights in patient care</h3>")
|
142 |
|
143 |
chatbot = gr.Chatbot(label="Analysis", height=600, type="messages")
|
144 |
-
file_upload = gr.File(
|
|
|
|
|
|
|
|
|
145 |
msg_input = gr.Textbox(placeholder="Ask about potential oversights...", show_label=False)
|
146 |
send_btn = gr.Button("Analyze", variant="primary")
|
147 |
conversation_state = gr.State([])
|
148 |
download_output = gr.File(label="Download Full Report")
|
149 |
|
150 |
def analyze_potential_oversights(message: str, history: list, conversation: list, files: list):
|
|
|
151 |
try:
|
152 |
history.append({"role": "user", "content": message})
|
153 |
-
history.append({"role": "assistant", "content": "Analyzing records for potential oversights..."})
|
154 |
yield history, None
|
155 |
|
156 |
extracted_data = ""
|
157 |
file_hash_value = ""
|
158 |
-
if files:
|
159 |
with ThreadPoolExecutor(max_workers=4) as executor:
|
160 |
futures = [executor.submit(convert_file_to_json, f.name, f.name.split(".")[-1].lower()) for f in files if hasattr(f, 'name')]
|
161 |
-
|
162 |
-
|
163 |
-
file_hash_value = file_hash(files[0].name)
|
164 |
|
165 |
analysis_prompt = f"""Review these medical records and identify EXACTLY what might have been missed:
|
166 |
1. List potential missed diagnoses
|
@@ -172,7 +176,7 @@ Medical Records:\n{extracted_data[:15000]}
|
|
172 |
|
173 |
### Potential Oversights:\n"""
|
174 |
|
175 |
-
response =
|
176 |
for chunk in agent.run_gradio_chat(
|
177 |
message=analysis_prompt,
|
178 |
history=[],
|
@@ -183,23 +187,29 @@ Medical Records:\n{extracted_data[:15000]}
|
|
183 |
conversation=conversation
|
184 |
):
|
185 |
if isinstance(chunk, str):
|
186 |
-
response
|
187 |
elif isinstance(chunk, list):
|
188 |
-
response.
|
189 |
-
|
|
|
190 |
yield history, None
|
191 |
|
192 |
-
final_output = ""
|
193 |
if not final_output:
|
194 |
final_output = "No clear oversights identified. Recommend comprehensive review."
|
195 |
-
history[-1] = {"role": "assistant", "content": final_output}
|
196 |
|
197 |
-
report_path =
|
198 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
|
200 |
except Exception as e:
|
201 |
history.append({"role": "assistant", "content": f"❌ Analysis failed: {str(e)}"})
|
202 |
-
|
203 |
|
204 |
inputs = [msg_input, chatbot, conversation_state, file_upload]
|
205 |
outputs = [chatbot, download_output]
|
@@ -218,22 +228,6 @@ if __name__ == "__main__":
|
|
218 |
print("Initializing medical analysis agent...")
|
219 |
agent = init_agent()
|
220 |
|
221 |
-
print("Performing warm-up call...")
|
222 |
-
try:
|
223 |
-
warm_up = agent.run_gradio_chat(
|
224 |
-
message="Warm up",
|
225 |
-
history=[],
|
226 |
-
temperature=0.1,
|
227 |
-
max_new_tokens=10,
|
228 |
-
max_token=100,
|
229 |
-
call_agent=False,
|
230 |
-
conversation=[]
|
231 |
-
)
|
232 |
-
for _ in warm_up:
|
233 |
-
pass
|
234 |
-
except Exception as e:
|
235 |
-
print(f"Warm-up error: {str(e)}")
|
236 |
-
|
237 |
print("Launching interface...")
|
238 |
demo = create_ui(agent)
|
239 |
demo.queue().launch(
|
@@ -241,4 +235,4 @@ if __name__ == "__main__":
|
|
241 |
server_port=7860,
|
242 |
show_error=True,
|
243 |
share=True
|
244 |
-
)
|
|
|
60 |
text_chunks.append(f"=== Page {i+1} ===\n{(page.extract_text() or '').strip()}")
|
61 |
for i, page in enumerate(pdf.pages[3:max_pages], start=4):
|
62 |
page_text = page.extract_text() or ""
|
63 |
+
if any(re.search(rf'\\b{kw}\\b', page_text.lower()) for kw in MEDICAL_KEYWORDS):
|
64 |
text_chunks.append(f"=== Page {i} ===\n{page_text.strip()}")
|
65 |
return "\n\n".join(text_chunks)
|
66 |
except Exception as e:
|
|
|
130 |
enable_checker=True,
|
131 |
step_rag_num=8,
|
132 |
seed=100,
|
133 |
+
additional_default_tools=[],
|
134 |
)
|
135 |
agent.init_model()
|
136 |
return agent
|
|
|
141 |
gr.Markdown("<h3 style='text-align: center;'>Identify potential oversights in patient care</h3>")
|
142 |
|
143 |
chatbot = gr.Chatbot(label="Analysis", height=600, type="messages")
|
144 |
+
file_upload = gr.File(
|
145 |
+
label="Upload Medical Records",
|
146 |
+
file_types=[".pdf", ".csv", ".xls", ".xlsx"],
|
147 |
+
file_count="multiple"
|
148 |
+
)
|
149 |
msg_input = gr.Textbox(placeholder="Ask about potential oversights...", show_label=False)
|
150 |
send_btn = gr.Button("Analyze", variant="primary")
|
151 |
conversation_state = gr.State([])
|
152 |
download_output = gr.File(label="Download Full Report")
|
153 |
|
154 |
def analyze_potential_oversights(message: str, history: list, conversation: list, files: list):
|
155 |
+
start_time = time.time()
|
156 |
try:
|
157 |
history.append({"role": "user", "content": message})
|
158 |
+
history.append({"role": "assistant", "content": "⏳ Analyzing records for potential oversights..."})
|
159 |
yield history, None
|
160 |
|
161 |
extracted_data = ""
|
162 |
file_hash_value = ""
|
163 |
+
if files and isinstance(files, list):
|
164 |
with ThreadPoolExecutor(max_workers=4) as executor:
|
165 |
futures = [executor.submit(convert_file_to_json, f.name, f.name.split(".")[-1].lower()) for f in files if hasattr(f, 'name')]
|
166 |
+
extracted_data = "\n".join([sanitize_utf8(f.result()) for f in as_completed(futures)])
|
167 |
+
file_hash_value = file_hash(files[0].name) if files else ""
|
|
|
168 |
|
169 |
analysis_prompt = f"""Review these medical records and identify EXACTLY what might have been missed:
|
170 |
1. List potential missed diagnoses
|
|
|
176 |
|
177 |
### Potential Oversights:\n"""
|
178 |
|
179 |
+
response = ""
|
180 |
for chunk in agent.run_gradio_chat(
|
181 |
message=analysis_prompt,
|
182 |
history=[],
|
|
|
187 |
conversation=conversation
|
188 |
):
|
189 |
if isinstance(chunk, str):
|
190 |
+
response += chunk
|
191 |
elif isinstance(chunk, list):
|
192 |
+
response += "".join([c.content for c in chunk if hasattr(c, 'content')])
|
193 |
+
|
194 |
+
history[-1]["content"] = response.replace("[TOOL_CALLS]", "").strip()
|
195 |
yield history, None
|
196 |
|
197 |
+
final_output = response.replace("[TOOL_CALLS]", "").strip()
|
198 |
if not final_output:
|
199 |
final_output = "No clear oversights identified. Recommend comprehensive review."
|
|
|
200 |
|
201 |
+
report_path = None
|
202 |
+
if file_hash_value:
|
203 |
+
possible_report = os.path.join(report_dir, f"{file_hash_value}_report.txt")
|
204 |
+
if os.path.exists(possible_report):
|
205 |
+
report_path = possible_report
|
206 |
+
|
207 |
+
history[-1] = {"role": "assistant", "content": final_output}
|
208 |
+
yield history, report_path
|
209 |
|
210 |
except Exception as e:
|
211 |
history.append({"role": "assistant", "content": f"❌ Analysis failed: {str(e)}"})
|
212 |
+
yield history, None
|
213 |
|
214 |
inputs = [msg_input, chatbot, conversation_state, file_upload]
|
215 |
outputs = [chatbot, download_output]
|
|
|
228 |
print("Initializing medical analysis agent...")
|
229 |
agent = init_agent()
|
230 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
231 |
print("Launching interface...")
|
232 |
demo = create_ui(agent)
|
233 |
demo.queue().launch(
|
|
|
235 |
server_port=7860,
|
236 |
show_error=True,
|
237 |
share=True
|
238 |
+
)
|