Spaces:
Sleeping
Sleeping
File size: 19,829 Bytes
ceaa691 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 |
#!/usr/bin/env python3
# app.py - Health Reports processing agent (PDF -> cleaned text -> structured JSON)
import os
import json
import logging
import re
from pathlib import Path
from typing import List, Dict, Any
from flask import Flask, request, jsonify
from flask_cors import CORS
from dotenv import load_dotenv
from unstructured.partition.pdf import partition_pdf
# Bloatectomy class (as per the source you provided)
from bloatectomy import bloatectomy
# LLM / agent
from langchain_groq import ChatGroq
from langgraph.prebuilt import create_react_agent
# LangGraph imports
from langgraph.graph import StateGraph, START, END
from typing_extensions import TypedDict, NotRequired
# --- Logging ---------------------------------------------------------------
logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
logger = logging.getLogger("health-agent")
# --- Environment & config -------------------------------------------------
load_dotenv()
from pathlib import Path
REPORTS_ROOT = Path(os.getenv("REPORTS_ROOT", r"app\reports")) # e.g. /app/reports/<patient_id>/<file.pdf>
SSRI_FILE = Path(os.getenv("SSRI_FILE", r"app\medicationCategories\SSRI_list.txt"))
MISC_FILE = Path(os.getenv("MISC_FILE", r"app\medicationCategories\MISC_list.txt"))
GROQ_API_KEY = os.getenv("GROQ_API_KEY", None)
# --- LLM setup -------------------------------------------------------------
llm = ChatGroq(
model=os.getenv("LLM_MODEL", "meta-llama/llama-4-scout-17b-16e-instruct"),
temperature=0.0,
max_tokens=None,
)
# Top-level strict system prompt for report JSON pieces (each node will use a more specific prompt)
NODE_BASE_INSTRUCTIONS = """
You are HealthAI — a clinical assistant producing JSON for downstream processing.
Produce only valid JSON (no extra text). Follow field types exactly. If missing data, return empty strings or empty arrays.
Be conservative: do not assert diagnoses; provide suggestions and ask physician confirmation where needed.
"""
# Build a generic agent and a JSON resolver agent (to fix broken JSON from LLM)
agent = create_react_agent(model=llm, tools=[], prompt=NODE_BASE_INSTRUCTIONS)
agent_json_resolver = create_react_agent(model=llm, tools=[], prompt="""
You are a JSON fixer. Input: a possibly-malformed JSON-like text. Output: valid JSON only (enclosed in triple backticks).
Fix missing quotes, trailing commas, unescaped newlines, stray assistant labels, and ensure schema compliance.
""")
# -------------------- JSON extraction / sanitizer ---------------------------
def extract_json_from_llm_response(raw_response: str) -> dict:
"""
Try extracting a JSON object from raw LLM text. Performs common cleanups seen in LLM outputs.
Raises JSONDecodeError if parsing still fails.
"""
# --- 1) Pull out the JSON code-block if present ---
md = re.search(r"```(?:json)?\s*([\s\S]*?)\s*```", raw_response)
json_string = md.group(1).strip() if md else raw_response
# --- 2) Trim to the outermost { … } so we drop any prefix/suffix junk ---
first, last = json_string.find('{'), json_string.rfind('}')
if 0 <= first < last:
json_string = json_string[first:last+1]
# --- 3) PRE-CLEANUP: remove rogue assistant labels, fix boolean quotes ---
json_string = re.sub(r'\b\w+\s*{', '{', json_string)
json_string = re.sub(r'"assistant"\s*:', '', json_string)
json_string = re.sub(r'\b(false|true)"', r'\1', json_string)
# --- 4) Escape embedded quotes in long string fields (best-effort) ---
def _esc(m):
prefix, body = m.group(1), m.group(2)
return prefix + body.replace('"', r'\"')
json_string = re.sub(
r'("logic"\s*:\s*")([\s\S]+?)(?=",\s*"[A-Za-z_]\w*"\s*:\s*)',
_esc,
json_string
)
# --- 5) Remove trailing commas before } or ] ---
json_string = re.sub(r',\s*(?=[}\],])', '', json_string)
json_string = re.sub(r',\s*,', ',', json_string)
# --- 6) Balance braces if obvious excess ---
ob, cb = json_string.count('{'), json_string.count('}')
if cb > ob:
excess = cb - ob
json_string = json_string.rstrip()[:-excess]
# --- 7) Escape literal newlines inside strings so json.loads can parse ---
def _escape_newlines_in_strings(s: str) -> str:
return re.sub(
r'"((?:[^"\\]|\\.)*?)"',
lambda m: '"' + m.group(1).replace('\n', '\\n').replace('\r', '\\r') + '"',
s,
flags=re.DOTALL
)
json_string = _escape_newlines_in_strings(json_string)
# Final parse
return json.loads(json_string)
# -------------------- Utility: Bloatectomy wrapper ------------------------
def clean_notes_with_bloatectomy(text: str, style: str = "remov") -> str:
"""
Uses the bloatectomy class to remove duplicates.
style: 'highlight'|'bold'|'remov' ; we use 'remov' to delete duplicates.
Returns cleaned text (single string).
"""
try:
b = bloatectomy(text, style=style, output="html")
tokens = getattr(b, "tokens", None)
if not tokens:
return text
return "\n".join(tokens)
except Exception:
logger.exception("Bloatectomy cleaning failed; returning original text")
return text
# --------------- Utility: medication extraction (adapted) -----------------
def readDrugs_from_file(path: Path):
if not path.exists():
return {}, []
txt = path.read_text(encoding="utf-8", errors="ignore")
generics = re.findall(r"^(.*?)\|", txt, re.MULTILINE)
generics = [g.lower() for g in generics if g]
lines = [ln.strip().lower() for ln in txt.splitlines() if ln.strip()]
return dict(zip(generics, lines)), generics
def addToDrugs_line(line: str, drugs_flags: List[int], listing: Dict[str,str], genList: List[str]) -> List[int]:
gen_index = {g:i for i,g in enumerate(genList)}
for generic, pattern_line in listing.items():
try:
if re.search(pattern_line, line, re.I):
idx = gen_index.get(generic)
if idx is not None:
drugs_flags[idx] = 1
except re.error:
continue
return drugs_flags
def extract_medications_from_text(text: str) -> List[str]:
ssri_map, ssri_generics = readDrugs_from_file(SSRI_FILE)
misc_map, misc_generics = readDrugs_from_file(MISC_FILE)
combined_map = {**ssri_map, **misc_map}
combined_generics = []
if ssri_generics:
combined_generics.extend(ssri_generics)
if misc_generics:
combined_generics.extend(misc_generics)
flags = [0]* len(combined_generics)
meds_found = set()
for ln in text.splitlines():
ln = ln.strip()
if not ln:
continue
if combined_map:
flags = addToDrugs_line(ln, flags, combined_map, combined_generics)
m = re.search(r"\b(Rx|Drug|Medication|Prescribed|Tablet)\s*[:\-]?\s*([A-Za-z0-9\-\s/\.]+)", ln, re.I)
if m:
meds_found.add(m.group(2).strip())
m2 = re.findall(r"\b([A-Z][a-z0-9\-]{2,}\s*(?:[0-9]{1,4}\s*(?:mg|mcg|g|IU))?)", ln)
for s in m2:
if re.search(r"\b(mg|mcg|g|IU)\b", s, re.I):
meds_found.add(s.strip())
for i, f in enumerate(flags):
if f == 1:
meds_found.add(combined_generics[i])
return list(meds_found)
# -------------------- Node prompts --------------------------
PATIENT_NODE_PROMPT = """
You will extract patientDetails from the provided document texts.
Return ONLY JSON with this exact shape:
{ "patientDetails": {"name": "", "age": "", "sex": "", "pid": ""} }
Fill fields using text evidence or leave empty strings.
"""
DOCTOR_NODE_PROMPT = """
You will extract doctorDetails found in the documents.
Return ONLY JSON with this exact shape:
{ "doctorDetails": {"referredBy": ""} }
"""
TEST_REPORT_NODE_PROMPT = """
You will extract per-test structured results from the documents.
Return ONLY JSON with this exact shape:
{
"reports": [
{
"testName": "",
"dateReported": "",
"timeReported": "",
"abnormalFindings": [
{"investigation": "", "result": 0, "unit": "", "status": "", "referenceValue": ""}
],
"interpretation": "",
"trends": []
}
]
}
- Include only findings that are outside reference ranges OR explicitly called 'abnormal' in the report.
- For result numeric parsing, prefer numeric values; if not numeric, keep original string.
- Use statuses: Low, High, Borderline, Positive, Negative, Normal.
"""
ANALYSIS_NODE_PROMPT = """
You will create an overallAnalysis based on the extracted reports (the agent will give you the 'reports' JSON).
Return ONLY JSON:
{ "overallAnalysis": { "summary": "", "recommendations": "", "longTermTrends": "",""risk_prediction": "","drug_interaction": "" } }
Be conservative, evidence-based, and suggest follow-up steps for physicians.
"""
CONDITION_LOOP_NODE_PROMPT = """
Validation and condition node:
Input: partial JSON (patientDetails, doctorDetails, reports, overallAnalysis).
Task: Check required keys exist and that each report has at least testName and abnormalFindings list.
Return ONLY JSON:
{ "valid": true, "missing": [] }
If missing fields, list keys in 'missing'. Do NOT modify content.
"""
# -------------------- Node helpers -------------------------
def call_node_agent(node_prompt: str, payload: dict) -> dict:
"""
Call the generic agent with a targeted node prompt and the payload.
Tries to parse JSON. If parsing fails, uses the JSON resolver agent once.
"""
try:
content = {
"prompt": node_prompt,
"payload": payload
}
resp = agent.invoke({"messages": [{"role": "user", "content": json.dumps(content)}]})
# Extract raw text from AIMessage or other response types
raw = None
if isinstance(resp, str):
raw = resp
elif hasattr(resp, "content"): # AIMessage or similar
raw = resp.content
elif isinstance(resp, dict):
msgs = resp.get("messages")
if msgs:
last_msg = msgs[-1]
if isinstance(last_msg, str):
raw = last_msg
elif hasattr(last_msg, "content"):
raw = last_msg.content
elif isinstance(last_msg, dict):
raw = last_msg.get("content", "")
else:
raw = str(last_msg)
else:
raw = json.dumps(resp)
else:
raw = str(resp)
parsed = extract_json_from_llm_response(raw)
return parsed
except Exception as e:
logger.warning("Node agent JSON parse failed: %s. Attempting JSON resolver.", e)
try:
resolver_prompt = f"Fix this JSON. Input:\n```json\n{raw}\n```\nReturn valid JSON only."
r = agent_json_resolver.invoke({"messages": [{"role": "user", "content": resolver_prompt}]})
rtxt = None
if isinstance(r, str):
rtxt = r
elif hasattr(r, "content"):
rtxt = r.content
elif isinstance(r, dict):
msgs = r.get("messages")
if msgs:
last_msg = msgs[-1]
if isinstance(last_msg, str):
rtxt = last_msg
elif hasattr(last_msg, "content"):
rtxt = last_msg.content
elif isinstance(last_msg, dict):
rtxt = last_msg.get("content", "")
else:
rtxt = str(last_msg)
else:
rtxt = json.dumps(r)
else:
rtxt = str(r)
corrected = extract_json_from_llm_response(rtxt)
return corrected
except Exception as e2:
logger.exception("JSON resolver also failed: %s", e2)
return {}
# -------------------- Define LangGraph State schema -------------------------
class State(TypedDict):
patient_meta: NotRequired[Dict[str, Any]]
patient_id: str
documents: List[Dict[str, Any]]
medications: List[str]
patientDetails: NotRequired[Dict[str, Any]]
doctorDetails: NotRequired[Dict[str, Any]]
reports: NotRequired[List[Dict[str, Any]]]
overallAnalysis: NotRequired[Dict[str, Any]]
valid: NotRequired[bool]
missing: NotRequired[List[str]]
# -------------------- Node implementations as LangGraph nodes -------------------------
def patient_details_node(state: State) -> dict:
payload = {
"patient_meta": state.get("patient_meta", {}),
"documents": state.get("documents", []),
"medications": state.get("medications", [])
}
logger.info("Running patient_details_node")
out = call_node_agent(PATIENT_NODE_PROMPT, payload)
return {"patientDetails": out.get("patientDetails", {}) if isinstance(out, dict) else {}}
def doctor_details_node(state: State) -> dict:
payload = {
"documents": state.get("documents", []),
"medications": state.get("medications", [])
}
logger.info("Running doctor_details_node")
out = call_node_agent(DOCTOR_NODE_PROMPT, payload)
return {"doctorDetails": out.get("doctorDetails", {}) if isinstance(out, dict) else {}}
def test_report_node(state: State) -> dict:
payload = {
"documents": state.get("documents", []),
"medications": state.get("medications", [])
}
logger.info("Running test_report_node")
out = call_node_agent(TEST_REPORT_NODE_PROMPT, payload)
return {"reports": out.get("reports", []) if isinstance(out, dict) else []}
def analysis_node(state: State) -> dict:
payload = {
"patientDetails": state.get("patientDetails", {}),
"doctorDetails": state.get("doctorDetails", {}),
"reports": state.get("reports", []),
"medications": state.get("medications", [])
}
logger.info("Running analysis_node")
out = call_node_agent(ANALYSIS_NODE_PROMPT, payload)
return {"overallAnalysis": out.get("overallAnalysis", {}) if isinstance(out, dict) else {}}
def condition_loop_node(state: State) -> dict:
payload = {
"patientDetails": state.get("patientDetails", {}),
"doctorDetails": state.get("doctorDetails", {}),
"reports": state.get("reports", []),
"overallAnalysis": state.get("overallAnalysis", {})
}
logger.info("Running condition_loop_node (validation)")
out = call_node_agent(CONDITION_LOOP_NODE_PROMPT, payload)
if isinstance(out, dict) and "valid" in out:
return {"valid": bool(out.get("valid")), "missing": out.get("missing", [])}
missing = []
if not state.get("patientDetails"):
missing.append("patientDetails")
if not state.get("reports"):
missing.append("reports")
return {"valid": len(missing) == 0, "missing": missing}
# -------------------- Build LangGraph StateGraph -------------------------
graph_builder = StateGraph(State)
graph_builder.add_node("patient_details", patient_details_node)
graph_builder.add_node("doctor_details", doctor_details_node)
graph_builder.add_node("test_report", test_report_node)
graph_builder.add_node("analysis", analysis_node)
graph_builder.add_node("condition_loop", condition_loop_node)
graph_builder.add_edge(START, "patient_details")
graph_builder.add_edge("patient_details", "doctor_details")
graph_builder.add_edge("doctor_details", "test_report")
graph_builder.add_edge("test_report", "analysis")
graph_builder.add_edge("analysis", "condition_loop")
graph_builder.add_edge("condition_loop", END)
graph = graph_builder.compile()
# -------------------- Flask app & endpoints -------------------------------
BASE_DIR = Path(__file__).resolve().parent
static_folder = BASE_DIR / "static"
app = Flask(__name__, static_folder=str(static_folder), static_url_path="/static")
CORS(app) # dev convenience; lock down in production
# serve frontend root
@app.route("/", methods=["GET"])
def serve_frontend():
try:
return app.send_static_file("frontend.html")
except Exception:
return "<h3>frontend.html not found in static/ — drop your frontend.html there.</h3>", 404
@app.route("/process_reports", methods=["POST"])
def process_reports():
data = request.get_json(force=True)
patient_id = data.get("patient_id")
filenames = data.get("filenames", [])
extra_patient_meta = data.get("patientDetails", {})
if not patient_id or not filenames:
return jsonify({"error": "missing patient_id or filenames"}), 400
patient_folder = REPORTS_ROOT / str(patient_id)
if not patient_folder.exists() or not patient_folder.is_dir():
return jsonify({"error": f"patient folder not found: {patient_folder}"}), 404
documents = []
combined_text_parts = []
for fname in filenames:
file_path = patient_folder / fname
if not file_path.exists():
logger.warning("file not found: %s", file_path)
continue
try:
elements = partition_pdf(filename=str(file_path))
page_text = "\n".join([el.text for el in elements if hasattr(el, "text") and el.text])
except Exception:
logger.exception("Failed to parse PDF %s", file_path)
page_text = ""
cleaned = clean_notes_with_bloatectomy(page_text, style="remov")
documents.append({
"filename": fname,
"raw_text": page_text,
"cleaned_text": cleaned
})
combined_text_parts.append(cleaned)
if not documents:
return jsonify({"error": "no valid documents found"}), 400
combined_text = "\n\n".join(combined_text_parts)
meds = extract_medications_from_text(combined_text)
initial_state = {
"patient_meta": extra_patient_meta,
"patient_id": patient_id,
"documents": documents,
"medications": meds
}
try:
result_state = graph.invoke(initial_state)
# Validate and fill placeholders if needed
if not result_state.get("valid", True):
missing = result_state.get("missing", [])
logger.info("Validation failed; missing keys: %s", missing)
if "patientDetails" in missing:
result_state["patientDetails"] = extra_patient_meta or {"name": "", "age": "", "sex": "", "pid": patient_id}
if "reports" in missing:
result_state["reports"] = []
# Re-run analysis node to keep overallAnalysis consistent
result_state.update(analysis_node(result_state))
# Re-validate
cond = condition_loop_node(result_state)
result_state.update(cond)
safe_response = {
"patientDetails": result_state.get("patientDetails", {"name": "", "age": "", "sex": "", "pid": patient_id}),
"doctorDetails": result_state.get("doctorDetails", {"referredBy": ""}),
"reports": result_state.get("reports", []),
"overallAnalysis": result_state.get("overallAnalysis", {"summary": "", "recommendations": "", "longTermTrends": ""}),
"_pre_extracted_medications": result_state.get("medications", []),
"_validation": {
"valid": result_state.get("valid", True),
"missing": result_state.get("missing", [])
}
}
return jsonify(safe_response), 200
except Exception as e:
logger.exception("Node pipeline failed")
return jsonify({"error": "Node pipeline failed", "detail": str(e)}), 500
@app.route("/ping", methods=["GET"])
def ping():
return jsonify({"status": "ok"})
if __name__ == "__main__":
port = int(os.getenv("PORT", 5000))
app.run(host="0.0.0.0", port=port, debug=True)
|