Update app.py
Browse files
app.py
CHANGED
@@ -277,15 +277,57 @@ def init_agent():
|
|
277 |
return agent
|
278 |
|
279 |
def create_ui(agent):
|
280 |
-
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
281 |
gr.Markdown("<h1 style='text-align: center;'>🩺 Clinical Oversight Assistant</h1>")
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
289 |
|
290 |
prompt_template = """
|
291 |
Analyze the patient record excerpt for missed diagnoses only. Provide a concise, evidence-based summary as a single paragraph without headings or bullet points. Include specific clinical findings (e.g., 'elevated blood pressure (160/95) on page 10'), their potential implications (e.g., 'may indicate untreated hypertension'), and a recommendation for urgent review. Do not include other oversight categories like medication conflicts. If no missed diagnoses are found, state 'No missed diagnoses identified' in a single sentence.
|
@@ -293,9 +335,37 @@ Patient Record Excerpt (Chunk {0} of {1}):
|
|
293 |
{chunk}
|
294 |
"""
|
295 |
|
296 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
297 |
history.append({"role": "user", "content": message})
|
298 |
-
yield
|
|
|
|
|
|
|
|
|
|
|
299 |
|
300 |
extracted = []
|
301 |
file_hash_value = ""
|
@@ -306,11 +376,7 @@ Patient Record Excerpt (Chunk {0} of {1}):
|
|
306 |
futures = []
|
307 |
for f in files:
|
308 |
file_type = f.name.split(".")[-1].lower()
|
309 |
-
futures.append(executor.submit(
|
310 |
-
process_file,
|
311 |
-
f.name,
|
312 |
-
file_type
|
313 |
-
))
|
314 |
|
315 |
for future in as_completed(futures):
|
316 |
try:
|
@@ -321,7 +387,12 @@ Patient Record Excerpt (Chunk {0} of {1}):
|
|
321 |
|
322 |
file_hash_value = file_hash(files[0].name) if files else ""
|
323 |
history.append({"role": "assistant", "content": "✅ File processing complete"})
|
324 |
-
yield
|
|
|
|
|
|
|
|
|
|
|
325 |
|
326 |
# Convert extracted data to JSON text
|
327 |
text_content = "\n".join(json.dumps(item) for item in extracted)
|
@@ -329,56 +400,45 @@ Patient Record Excerpt (Chunk {0} of {1}):
|
|
329 |
# Tokenize and chunk the content properly
|
330 |
chunks = tokenize_and_chunk(text_content)
|
331 |
combined_response = ""
|
332 |
-
batch_size = 2 # Reduced batch size to prevent token overflow
|
333 |
|
334 |
try:
|
335 |
-
for
|
336 |
-
|
337 |
-
batch_prompts = [
|
338 |
-
prompt_template.format(
|
339 |
-
batch_idx + i + 1,
|
340 |
-
len(chunks),
|
341 |
-
chunk=chunk[:1800] # Conservative chunk size
|
342 |
-
)
|
343 |
-
for i, chunk in enumerate(batch_chunks)
|
344 |
-
]
|
345 |
|
346 |
-
|
347 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
348 |
|
349 |
-
# Process
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
357 |
}
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
364 |
-
if isinstance(chunk_output, list):
|
365 |
-
for m in chunk_output:
|
366 |
-
if hasattr(m, 'content') and m.content:
|
367 |
-
cleaned = clean_response(m.content)
|
368 |
-
if cleaned:
|
369 |
-
chunk_response += cleaned + " "
|
370 |
-
elif isinstance(chunk_output, str) and chunk_output.strip():
|
371 |
-
cleaned = clean_response(chunk_output)
|
372 |
-
if cleaned:
|
373 |
-
chunk_response += cleaned + " "
|
374 |
-
|
375 |
-
combined_response += f"--- Analysis for Chunk {batch_idx + 1} ---\n{chunk_response.strip()}\n"
|
376 |
-
history[-1] = {"role": "assistant", "content": combined_response.strip()}
|
377 |
-
yield history, None, ""
|
378 |
-
|
379 |
-
# Clean up memory
|
380 |
-
torch.cuda.empty_cache()
|
381 |
-
gc.collect()
|
382 |
|
383 |
# Generate final summary
|
384 |
summary = summarize_findings(combined_response)
|
@@ -387,15 +447,53 @@ Patient Record Excerpt (Chunk {0} of {1}):
|
|
387 |
with open(report_path, "w", encoding="utf-8") as f:
|
388 |
f.write(combined_response + "\n\n" + summary)
|
389 |
|
390 |
-
yield
|
|
|
|
|
|
|
|
|
|
|
391 |
|
392 |
except Exception as e:
|
393 |
logger.error("Analysis error: %s", e)
|
394 |
history.append({"role": "assistant", "content": f"❌ Error occurred: {str(e)}"})
|
395 |
-
yield
|
396 |
-
|
397 |
-
|
398 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
399 |
return demo
|
400 |
|
401 |
if __name__ == "__main__":
|
@@ -403,13 +501,20 @@ if __name__ == "__main__":
|
|
403 |
logger.info("Launching app...")
|
404 |
agent = init_agent()
|
405 |
demo = create_ui(agent)
|
406 |
-
demo.queue(
|
|
|
|
|
|
|
407 |
server_name="0.0.0.0",
|
408 |
server_port=7860,
|
409 |
show_error=True,
|
410 |
allowed_paths=[report_dir],
|
411 |
-
share=False
|
|
|
412 |
)
|
|
|
|
|
|
|
413 |
finally:
|
414 |
if torch.distributed.is_initialized():
|
415 |
torch.distributed.destroy_process_group()
|
|
|
277 |
return agent
|
278 |
|
279 |
def create_ui(agent):
|
280 |
+
with gr.Blocks(theme=gr.themes.Soft(), title="Clinical Oversight Assistant") as demo:
|
281 |
gr.Markdown("<h1 style='text-align: center;'>🩺 Clinical Oversight Assistant</h1>")
|
282 |
+
|
283 |
+
with gr.Row():
|
284 |
+
with gr.Column(scale=3):
|
285 |
+
chatbot = gr.Chatbot(
|
286 |
+
label="Analysis Conversation",
|
287 |
+
height=600,
|
288 |
+
bubble_full_width=False,
|
289 |
+
show_copy_button=True,
|
290 |
+
avatar_images=(
|
291 |
+
"assets/user.png",
|
292 |
+
"assets/assistant.png"
|
293 |
+
)
|
294 |
+
)
|
295 |
+
with gr.Column(scale=1):
|
296 |
+
final_summary = gr.Markdown(
|
297 |
+
label="Summary of Findings",
|
298 |
+
value="### Summary will appear here\nAfter analysis completes"
|
299 |
+
)
|
300 |
+
download_output = gr.File(
|
301 |
+
label="Download Full Report",
|
302 |
+
visible=False
|
303 |
+
)
|
304 |
+
|
305 |
+
with gr.Row():
|
306 |
+
file_upload = gr.File(
|
307 |
+
file_types=[".pdf", ".csv", ".xls", ".xlsx"],
|
308 |
+
file_count="multiple",
|
309 |
+
label="Upload Patient Records"
|
310 |
+
)
|
311 |
+
|
312 |
+
with gr.Row():
|
313 |
+
msg_input = gr.Textbox(
|
314 |
+
placeholder="Ask about potential oversights...",
|
315 |
+
show_label=False,
|
316 |
+
container=False,
|
317 |
+
scale=7,
|
318 |
+
autofocus=True
|
319 |
+
)
|
320 |
+
send_btn = gr.Button(
|
321 |
+
"Analyze",
|
322 |
+
variant="primary",
|
323 |
+
scale=1,
|
324 |
+
min_width=100
|
325 |
+
)
|
326 |
+
|
327 |
+
progress_bar = gr.Progress(
|
328 |
+
label="Processing Progress",
|
329 |
+
visible=False
|
330 |
+
)
|
331 |
|
332 |
prompt_template = """
|
333 |
Analyze the patient record excerpt for missed diagnoses only. Provide a concise, evidence-based summary as a single paragraph without headings or bullet points. Include specific clinical findings (e.g., 'elevated blood pressure (160/95) on page 10'), their potential implications (e.g., 'may indicate untreated hypertension'), and a recommendation for urgent review. Do not include other oversight categories like medication conflicts. If no missed diagnoses are found, state 'No missed diagnoses identified' in a single sentence.
|
|
|
335 |
{chunk}
|
336 |
"""
|
337 |
|
338 |
+
def process_response_stream(prompt: str, history: List[dict]) -> Generator[dict, None, None]:
|
339 |
+
"""Process a single prompt and stream the response"""
|
340 |
+
full_response = ""
|
341 |
+
for chunk_output in agent.run_gradio_chat(prompt, [], 0.2, 512, 2048, False, []):
|
342 |
+
if chunk_output is None:
|
343 |
+
continue
|
344 |
+
|
345 |
+
if isinstance(chunk_output, list):
|
346 |
+
for m in chunk_output:
|
347 |
+
if hasattr(m, 'content') and m.content:
|
348 |
+
cleaned = clean_response(m.content)
|
349 |
+
if cleaned:
|
350 |
+
full_response += cleaned + " "
|
351 |
+
yield {"role": "assistant", "content": full_response}
|
352 |
+
elif isinstance(chunk_output, str) and chunk_output.strip():
|
353 |
+
cleaned = clean_response(chunk_output)
|
354 |
+
if cleaned:
|
355 |
+
full_response += cleaned + " "
|
356 |
+
yield {"role": "assistant", "content": full_response}
|
357 |
+
|
358 |
+
return full_response
|
359 |
+
|
360 |
+
def analyze(message: str, history: List[dict], files: List) -> Generator[dict, None, None]:
|
361 |
+
# Start with user message
|
362 |
history.append({"role": "user", "content": message})
|
363 |
+
yield {
|
364 |
+
"chatbot": history,
|
365 |
+
"download_output": None,
|
366 |
+
"final_summary": "",
|
367 |
+
"progress_bar": gr.Progress(visible=True)
|
368 |
+
}
|
369 |
|
370 |
extracted = []
|
371 |
file_hash_value = ""
|
|
|
376 |
futures = []
|
377 |
for f in files:
|
378 |
file_type = f.name.split(".")[-1].lower()
|
379 |
+
futures.append(executor.submit(process_file, f.name, file_type))
|
|
|
|
|
|
|
|
|
380 |
|
381 |
for future in as_completed(futures):
|
382 |
try:
|
|
|
387 |
|
388 |
file_hash_value = file_hash(files[0].name) if files else ""
|
389 |
history.append({"role": "assistant", "content": "✅ File processing complete"})
|
390 |
+
yield {
|
391 |
+
"chatbot": history,
|
392 |
+
"download_output": None,
|
393 |
+
"final_summary": "",
|
394 |
+
"progress_bar": gr.Progress(0.2, visible=True, label="Processing files")
|
395 |
+
}
|
396 |
|
397 |
# Convert extracted data to JSON text
|
398 |
text_content = "\n".join(json.dumps(item) for item in extracted)
|
|
|
400 |
# Tokenize and chunk the content properly
|
401 |
chunks = tokenize_and_chunk(text_content)
|
402 |
combined_response = ""
|
|
|
403 |
|
404 |
try:
|
405 |
+
for chunk_idx, chunk in enumerate(chunks, 1):
|
406 |
+
prompt = prompt_template.format(chunk_idx, len(chunks), chunk=chunk[:1800])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
407 |
|
408 |
+
# Create a placeholder message
|
409 |
+
history.append({"role": "assistant", "content": ""})
|
410 |
+
yield {
|
411 |
+
"chatbot": history,
|
412 |
+
"download_output": None,
|
413 |
+
"final_summary": "",
|
414 |
+
"progress_bar": gr.Progress(
|
415 |
+
0.2 + (chunk_idx/len(chunks))*0.7,
|
416 |
+
visible=True,
|
417 |
+
label=f"Analyzing chunk {chunk_idx}/{len(chunks)}"
|
418 |
+
)
|
419 |
+
}
|
420 |
|
421 |
+
# Process and stream the response
|
422 |
+
chunk_response = ""
|
423 |
+
for update in process_response_stream(prompt, history):
|
424 |
+
# Update the last message with streaming content
|
425 |
+
history[-1] = update
|
426 |
+
chunk_response = update["content"]
|
427 |
+
yield {
|
428 |
+
"chatbot": history,
|
429 |
+
"download_output": None,
|
430 |
+
"final_summary": "",
|
431 |
+
"progress_bar": gr.Progress(
|
432 |
+
0.2 + (chunk_idx/len(chunks))*0.7,
|
433 |
+
visible=True
|
434 |
+
)
|
435 |
}
|
436 |
+
|
437 |
+
combined_response += f"--- Analysis for Chunk {chunk_idx} ---\n{chunk_response}\n"
|
438 |
+
|
439 |
+
# Clean up memory
|
440 |
+
torch.cuda.empty_cache()
|
441 |
+
gc.collect()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
442 |
|
443 |
# Generate final summary
|
444 |
summary = summarize_findings(combined_response)
|
|
|
447 |
with open(report_path, "w", encoding="utf-8") as f:
|
448 |
f.write(combined_response + "\n\n" + summary)
|
449 |
|
450 |
+
yield {
|
451 |
+
"chatbot": history,
|
452 |
+
"download_output": gr.File(report_path) if report_path and os.path.exists(report_path) else None,
|
453 |
+
"final_summary": summary,
|
454 |
+
"progress_bar": gr.Progress(1.0, visible=False)
|
455 |
+
}
|
456 |
|
457 |
except Exception as e:
|
458 |
logger.error("Analysis error: %s", e)
|
459 |
history.append({"role": "assistant", "content": f"❌ Error occurred: {str(e)}"})
|
460 |
+
yield {
|
461 |
+
"chatbot": history,
|
462 |
+
"download_output": None,
|
463 |
+
"final_summary": f"Error occurred during analysis: {str(e)}",
|
464 |
+
"progress_bar": gr.Progress(visible=False)
|
465 |
+
}
|
466 |
+
|
467 |
+
def clear_and_start():
|
468 |
+
return {
|
469 |
+
"chatbot": [],
|
470 |
+
"download_output": None,
|
471 |
+
"final_summary": "",
|
472 |
+
"msg_input": "",
|
473 |
+
"file_upload": None
|
474 |
+
}
|
475 |
+
|
476 |
+
# Event handlers
|
477 |
+
send_btn.click(
|
478 |
+
analyze,
|
479 |
+
inputs=[msg_input, chatbot, file_upload],
|
480 |
+
outputs=[chatbot, download_output, final_summary, progress_bar],
|
481 |
+
show_progress="hidden"
|
482 |
+
)
|
483 |
+
|
484 |
+
msg_input.submit(
|
485 |
+
analyze,
|
486 |
+
inputs=[msg_input, chatbot, file_upload],
|
487 |
+
outputs=[chatbot, download_output, final_summary, progress_bar],
|
488 |
+
show_progress="hidden"
|
489 |
+
)
|
490 |
+
|
491 |
+
demo.load(
|
492 |
+
clear_and_start,
|
493 |
+
outputs=[chatbot, download_output, final_summary, msg_input, file_upload],
|
494 |
+
queue=False
|
495 |
+
)
|
496 |
+
|
497 |
return demo
|
498 |
|
499 |
if __name__ == "__main__":
|
|
|
501 |
logger.info("Launching app...")
|
502 |
agent = init_agent()
|
503 |
demo = create_ui(agent)
|
504 |
+
demo.queue(
|
505 |
+
api_open=False,
|
506 |
+
max_size=20
|
507 |
+
).launch(
|
508 |
server_name="0.0.0.0",
|
509 |
server_port=7860,
|
510 |
show_error=True,
|
511 |
allowed_paths=[report_dir],
|
512 |
+
share=False,
|
513 |
+
favicon_path="assets/favicon.ico"
|
514 |
)
|
515 |
+
except Exception as e:
|
516 |
+
logger.error(f"Failed to launch app: {e}")
|
517 |
+
raise
|
518 |
finally:
|
519 |
if torch.distributed.is_initialized():
|
520 |
torch.distributed.destroy_process_group()
|