|
|
|
|
|
from concurrent.futures import ThreadPoolExecutor, as_completed |
|
|
from threading import Thread |
|
|
|
|
|
|
|
|
current_dir = os.path.dirname(os.path.abspath(__file__)) |
|
|
src_path = os.path.join(current_dir, "src") |
|
|
sys.path.insert(0, src_path) |
|
|
|
|
|
base_dir = "/data" |
|
|
model_cache_dir = os.path.join(base_dir, "txagent_models") |
|
|
tool_cache_dir = os.path.join(base_dir, "tool_cache") |
|
|
file_cache_dir = os.path.join(base_dir, "cache") |
|
|
report_dir = os.path.join(base_dir, "reports") |
|
|
vllm_cache_dir = os.path.join(base_dir, "vllm_cache") |
|
|
|
|
|
for d in [model_cache_dir, tool_cache_dir, file_cache_dir, report_dir, vllm_cache_dir]: |
|
|
os.makedirs(d, exist_ok=True) |
|
|
|
|
|
|
|
|
os.environ.update({ |
|
|
"HF_HOME": model_cache_dir, |
|
|
"TRANSFORMERS_CACHE": model_cache_dir, |
|
|
"VLLM_CACHE_DIR": vllm_cache_dir, |
|
|
"TOKENIZERS_PARALLELISM": "false", |
|
|
"CUDA_LAUNCH_BLOCKING": "1" |
|
|
}) |
|
|
|
|
|
from txagent.txagent import TxAgent |
|
|
|
|
|
MEDICAL_KEYWORDS = {'diagnosis', 'assessment', 'plan', 'results', 'medications', |
|
|
'allergies', 'summary', 'impression', 'findings', 'recommendations'} |
|
|
|
|
|
def sanitize_utf8(text): return text.encode("utf-8", "ignore").decode("utf-8") |
|
|
def file_hash(path): return hashlib.md5(open(path, "rb").read()).hexdigest() |
|
|
|
|
|
def extract_priority_pages(file_path, max_pages=20): |
|
|
try: |
|
|
with pdfplumber.open(file_path) as pdf: |
|
|
pages = [] |
|
|
for i, page in enumerate(pdf.pages[:3]): |
|
|
pages.append(f"=== Page {i+1} ===\n{(page.extract_text() or '').strip()}") |
|
|
for i, page in enumerate(pdf.pages[3:max_pages], start=4): |
|
|
text = page.extract_text() or "" |
|
|
if any(re.search(rf'\b{kw}\b', text.lower()) for kw in MEDICAL_KEYWORDS): |
|
|
pages.append(f"=== Page {i} ===\n{text.strip()}") |
|
|
return "\n\n".join(pages) |
|
|
except Exception as e: |
|
|
return f"PDF processing error: {str(e)}" |
|
|
|
|
|
def convert_file_to_json(file_path, file_type): |
|
|
try: |
|
|
h = file_hash(file_path) |
|
|
cache_path = os.path.join(file_cache_dir, f"{h}.json") |
|
|
if os.path.exists(cache_path): return open(cache_path, "r", encoding="utf-8").read() |
|
|
|
|
|
if file_type == "pdf": |
|
|
text = extract_priority_pages(file_path) |
|
|
result = json.dumps({"filename": os.path.basename(file_path), "content": text, "status": "initial"}) |
|
|
Thread(target=full_pdf_processing, args=(file_path, h)).start() |
|
|
elif file_type == "csv": |
|
|
df = pd.read_csv(file_path, encoding_errors="replace", header=None, dtype=str, skip_blank_lines=False, on_bad_lines="skip") |
|
|
result = json.dumps({"filename": os.path.basename(file_path), "rows": df.fillna("").astype(str).values.tolist()}) |
|
|
elif file_type in ["xls", "xlsx"]: |
|
|
try: |
|
|
df = pd.read_excel(file_path, engine="openpyxl", header=None, dtype=str) |
|
|
except: |
|
|
df = pd.read_excel(file_path, engine="xlrd", header=None, dtype=str) |
|
|
result = json.dumps({"filename": os.path.basename(file_path), "rows": df.fillna("").astype(str).values.tolist()}) |
|
|
else: |
|
|
return json.dumps({"error": f"Unsupported file type: {file_type}"}) |
|
|
|
|
|
with open(cache_path, "w", encoding="utf-8") as f: f.write(result) |
|
|
return result |
|
|
except Exception as e: |
|
|
return json.dumps({"error": f"Error processing {os.path.basename(file_path)}: {str(e)}"}) |
|
|
|
|
|
def full_pdf_processing(file_path, file_hash_value): |
|
|
try: |
|
|
cache_path = os.path.join(file_cache_dir, f"{file_hash_value}_full.json") |
|
|
if os.path.exists(cache_path): return |
|
|
with pdfplumber.open(file_path) as pdf: |
|
|
full_text = "\n".join([f"=== Page {i+1} ===\n{(page.extract_text() or '').strip()}" for i, page in enumerate(pdf.pages)]) |
|
|
result = json.dumps({"filename": os.path.basename(file_path), "content": full_text, "status": "complete"}) |
|
|
with open(cache_path, "w", encoding="utf-8") as f: f.write(result) |
|
|
with open(os.path.join(report_dir, f"{file_hash_value}_report.txt"), "w", encoding="utf-8") as out: out.write(full_text) |
|
|
except Exception as e: |
|
|
print("PDF processing error:", e) |
|
|
|
|
|
def init_agent(): |
|
|
default_tool_path = os.path.abspath("data/new_tool.json") |
|
|
target_tool_path = os.path.join(tool_cache_dir, "new_tool.json") |
|
|
if not os.path.exists(target_tool_path): |
|
|
shutil.copy(default_tool_path, target_tool_path) |
|
|
|
|
|
agent = TxAgent( |
|
|
model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B", |
|
|
rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B", |
|
|
tool_files_dict={"new_tool": target_tool_path}, |
|
|
force_finish=True, |
|
|
enable_checker=True, |
|
|
step_rag_num=8, |
|
|
seed=100 |
|
|
) |
|
|
agent.init_model() |
|
|
return agent |
|
|
|
|
|
|
|
|
agent_container = {"agent": None} |
|
|
def get_agent(): |
|
|
if agent_container["agent"] is None: |
|
|
agent_container["agent"] = init_agent() |
|
|
return agent_container["agent"] |
|
|
|
|
|
def create_ui(get_agent_func): |
|
|
with gr.Blocks(theme=gr.themes.Soft()) as demo: |
|
|
gr.Markdown("<h1 style='text-align: center;'>🩺 Clinical Oversight Assistant</h1><h3 style='text-align: center;'>Identify potential oversights in patient care</h3>") |
|
|
|
|
|
chatbot = gr.Chatbot(label="Analysis", height=600, type="messages") |
|
|
file_upload = gr.File(file_types=[".pdf", ".csv", ".xls", ".xlsx"], file_count="multiple") |
|
|
msg_input = gr.Textbox(placeholder="Ask about potential oversights...") |
|
|
send_btn = gr.Button("Analyze", variant="primary") |
|
|
state = gr.State([]) |
|
|
download_output = gr.File(label="Download Report") |
|
|
|
|
|
def analyze(message, history, conversation, files): |
|
|
try: |
|
|
extracted_data, file_hash_value = "", "" |
|
|
if files: |
|
|
with ThreadPoolExecutor(max_workers=4) as pool: |
|
|
futures = [pool.submit(convert_file_to_json, f.name, f.name.split(".")[-1].lower()) for f in files] |
|
|
extracted_data = "\n".join([sanitize_utf8(f.result()) for f in as_completed(futures)]) |
|
|
file_hash_value = file_hash(files[0].name) |
|
|
|
|
|
prompt = f"""Review these medical records and identify EXACTLY what might have been missed: |
|
|
1. List potential missed diagnoses |
|
|
2. Flag any medication conflicts |
|
|
3. Note incomplete assessments |
|
|
4. Highlight abnormal results needing follow-up |
|
|
|
|
|
Medical Records:\n{extracted_data[:15000]} |
|
|
|
|
|
### Potential Oversights:\n""" |
|
|
|
|
|
final_response = "" |
|
|
for chunk in get_agent_func().run_gradio_chat( |
|
|
message=prompt, |
|
|
history=[], |
|
|
temperature=0.2, |
|
|
max_new_tokens=1024, |
|
|
max_token=4096, |
|
|
call_agent=False, |
|
|
conversation=conversation |
|
|
): |
|
|
if isinstance(chunk, str): |
|
|
final_response += chunk |
|
|
elif isinstance(chunk, list): |
|
|
final_response += "".join([c.content for c in chunk if hasattr(c, "content")]) |
|
|
|
|
|
cleaned = final_response.replace("[TOOL_CALLS]", "").strip() |
|
|
if not cleaned: |
|
|
cleaned = "No oversights found. Consider further review." |
|
|
|
|
|
updated_history = history + [{"role": "user", "content": message}, {"role": "assistant", "content": cleaned}] |
|
|
|
|
|
report_path = os.path.join(report_dir, f"{file_hash_value}_report.txt") if file_hash_value and os.path.exists(os.path.join(report_dir, f"{file_hash_value}_report.txt")) else None |
|
|
yield updated_history, report_path |
|
|
except Exception as e: |
|
|
updated_history = history + [{"role": "user", "content": message}, {"role": "assistant", "content": f"❌ Error: {str(e)}"}] |
|
|
yield updated_history, None |
|
|
|
|
|
send_btn.click(analyze, inputs=[msg_input, chatbot, state, file_upload], outputs=[chatbot, download_output]) |
|
|
msg_input.submit(analyze, inputs=[msg_input, chatbot, state, file_upload], outputs=[chatbot, download_output]) |
|
|
|
|
|
return demo |
|
|
|
|
|
if __name__ == "__main__": |
|
|
print("Launching interface...") |
|
|
ui = create_ui(get_agent) |
|
|
ui.queue(api_open=False).launch( |
|
|
server_name="0.0.0.0", |
|
|
server_port=7860, |
|
|
show_error=True, |
|
|
allowed_paths=["/data/reports"], |
|
|
share=False |
|
|
) |
|
|
|
