Spaces:
Sleeping
Sleeping
File size: 11,570 Bytes
2ebe8a4 | 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 | from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import time
import json
import numpy as np
app = FastAPI(title="EdgeMed Clinical BERT API")
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_methods=["*"],
allow_headers=["*"],
)
# ββ Label maps (from your notebook) ββββββββββββββββββββββββββββββββββββββββββ
id2label = {0: "ESI_1", 1: "ESI_2", 2: "ESI_3", 3: "ESI_4", 4: "ESI_5"}
label2id = {v: k for k, v in id2label.items()}
ESI_SLA = {"ESI_1": 2, "ESI_2": 10, "ESI_3": 30, "ESI_4": 60, "ESI_5": 120}
ESI_LABEL = {"ESI_1": "Resuscitation", "ESI_2": "Emergent",
"ESI_3": "Urgent", "ESI_4": "Less Urgent", "ESI_5": "Non-Urgent"}
# ββ CAG keyword lookup (exact from your notebook) ββββββββββββββββββββββββββββ
CAG_RULES = {
"ESI_1": ["cardiac arrest","not breathing","no pulse","unresponsive",
"unconscious","active seizure","anaphylaxis","major trauma",
"respiratory arrest","hemorrhagic shock","arrest","cpr",
"resus","apnea","shock","code"],
"ESI_2": ["chest pain","acute stroke","stroke","altered mental status",
"severe pain","overdose","sepsis","hypertensive emergency",
"myocardial infarction","difficulty breathing",
"shortness of breath","loss of consciousness","syncope",
"fainting","high fever","sob","dyspnea","loc","seizure",
"convulsion","palpitation","hypotension","ams","cp"],
"ESI_3": ["moderate pain","fever","fracture","vomiting","dizziness",
"weakness","wound","laceration","burn","abdominal pain",
"back pain","headache","swelling","infection","urinary",
"bleeding","trauma","injury","pain"],
"ESI_4": ["mild pain","rash","sore throat","ear pain","eye pain",
"minor","sprain","cough","cold","mild","ocular"],
"ESI_5": ["prescription refill","routine","paperwork",
"immunization","administrative","certificate"],
}
PRIORITY = ["ESI_1", "ESI_2", "ESI_3", "ESI_4", "ESI_5"]
# Build flat lookup
CAG_LOOKUP = {}
for esi, keywords in CAG_RULES.items():
for kw in keywords:
CAG_LOOKUP[kw] = esi
def cag_classify(text: str):
t = text.lower()
matched_esi, matched_kw = None, None
for kw, esi in CAG_LOOKUP.items():
if kw in t:
if matched_esi is None or PRIORITY.index(esi) < PRIORITY.index(matched_esi):
matched_esi = esi
matched_kw = kw
return matched_esi, matched_kw
# ββ Keyword β specialty map (from your notebook) βββββββββββββββββββββββββββββ
KEYWORD_SPECIALTY = {
"cardiac": "Cardiology", "chest": "Cardiology",
"heart": "Cardiology", "neuro": "Neurology",
"stroke": "Neurology", "seizure": "Neurology",
"head": "Neurology", "fracture": "Orthopedic",
"bone": "Orthopedic", "joint": "Orthopedic",
"abdom": "General Surgery","bowel": "Gastroenterology",
"liver": "Gastroenterology","breath": "Pulmonology",
"lung": "Pulmonology", "psych": "Psychiatry",
"mental": "Psychiatry", "eye": "Ophthalmology",
"ocular": "Ophthalmology", "ear": "ENT",
"throat": "ENT", "urin": "Urology",
"kidney": "Nephrology", "renal": "Nephrology",
"burn": "General Surgery","wound": "General Surgery",
}
ESI_DEFAULT_SPECIALTY = {
"ESI_1": "Emergency Medicine", "ESI_2": "Emergency Medicine",
"ESI_3": "General Surgery", "ESI_4": "General Surgery",
"ESI_5": "General Surgery",
}
def detect_specialty(symptom_text: str, esi_level: str) -> str:
t = symptom_text.lower()
for kw, spec in KEYWORD_SPECIALTY.items():
if kw in t:
return spec
return ESI_DEFAULT_SPECIALTY.get(esi_level, "Emergency Medicine")
# ββ Load BERT model βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
print("Loading Mahdiya/edgemed-clinical-bert ...")
tokenizer = AutoTokenizer.from_pretrained("Mahdiya/edgemed-clinical-bert")
model = AutoModelForSequenceClassification.from_pretrained(
"Mahdiya/edgemed-clinical-bert")
model.eval()
device = "cpu" # CPU Basic Space β no GPU available
model.to(device)
print(f"β
Model loaded on {device}")
def bert_classify(text: str):
enc = tokenizer(
text[:400], return_tensors="pt",
max_length=128, truncation=True, padding="max_length"
).to(device)
t0 = time.time()
with torch.no_grad():
logits = model(**enc).logits
latency_ms = round((time.time() - t0) * 1000, 1)
probs = torch.softmax(logits, dim=-1)[0].cpu().tolist()
pred_id = int(torch.argmax(logits, dim=-1).item())
pred_esi = id2label[pred_id]
conf = round(probs[pred_id], 4)
all_probs = {id2label[i]: round(p, 4) for i, p in enumerate(probs)}
return pred_esi, conf, latency_ms, all_probs
# ββ Hospital data (200 hospitals, 5 zones β from your notebook seed=42) βββββββ
np.random.seed(42)
SPECIALTIES_ALL = [
"Cardiology","Neurology","Orthopedic","General Surgery",
"Emergency Medicine","Gastroenterology","Pulmonology",
"Nephrology","Psychiatry","Ophthalmology","ENT","Urology",
"Oncology","Dermatology","Pediatrics","Gynecology",
"Radiology","Anesthesiology","Hematology","Rheumatology"
]
ZONES = ["Zone-A","Zone-B","Zone-C","Zone-D","Zone-E"]
HOSPITALS = []
for i in range(200):
zone = ZONES[i // 40]
n_specs = int(np.random.randint(3, 7))
specs = list(np.random.choice(SPECIALTIES_ALL, n_specs, replace=False))
HOSPITALS.append({
"hospital_id": f"H{str(i).zfill(3)}",
"name": f"{zone.replace('Zone-','').strip()} Medical Center {i%40+1}",
"zone": zone,
"specialties": specs,
"response_time": round(float(np.random.uniform(1, 30)), 1),
"quality_score": round(float(np.random.uniform(0.5, 1.0)), 2),
"current_load": round(float(np.random.uniform(0.1, 0.9)), 2),
"availability": bool(np.random.random() > 0.2),
})
def routing_score(h: dict, alpha: float) -> float:
"""Exact formula from your notebook."""
speed = 1.0 - (h["response_time"] / 30.0)
quality = h["quality_score"]
load = h["current_load"] * 0.3
return round((alpha * speed + (1 - alpha) * quality) * (1 - load), 4)
def get_top_hospitals(specialty: str, zone: str, alpha: float,
esi: str, top_n: int = 10) -> list:
is_emergency = esi in ("ESI_1", "ESI_2")
results = []
for h in HOSPITALS:
if not h["availability"]:
continue
if h["current_load"] > 0.85:
continue
spec_match = any(specialty.lower() in s.lower() for s in h["specialties"])
zone_match = h["zone"] == zone
if is_emergency:
# Emergency β any available hospital with any specialty
eff_alpha = 1.0 # pure speed
score = routing_score(h, eff_alpha)
results.append({**h, "score": score,
"zone_match": zone_match,
"spec_match": spec_match,
"cross_zone": not zone_match})
else:
if spec_match:
score = routing_score(h, alpha)
results.append({**h, "score": score,
"zone_match": zone_match,
"spec_match": spec_match,
"cross_zone": not zone_match})
# Sort: zone-local first, then by score
if is_emergency:
results.sort(key=lambda x: x["response_time"])
else:
results.sort(key=lambda x: (-int(x["zone_match"]), -x["score"]))
return results[:top_n]
# ββ Request / Response models βββββββββββββββββββββββββββββββββββββββββββββββββ
class TriageRequest(BaseModel):
symptom_text: str
zone: str
alpha: float = 0.5
class RouteRequest(BaseModel):
symptom_text: str
zone: str
alpha: float
esi_level: str # already determined (from triage step)
specialty: str # already determined
# ββ Endpoints βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@app.get("/")
def root():
return {"status": "EdgeMed API running",
"model": "Mahdiya/edgemed-clinical-bert",
"device": device}
@app.post("/triage")
def triage(req: TriageRequest):
"""
Full triage pipeline:
1. CAG keyword check
2. If no CAG hit β BERT inference (ESI 3-5)
Returns ESI level, confidence, method used, latency.
"""
t_total = time.time()
# Step 1: CAG
cag_esi, cag_kw = cag_classify(req.symptom_text)
if cag_esi in ("ESI_1", "ESI_2"):
# Bypass BERT β critical keyword found
specialty = detect_specialty(req.symptom_text, cag_esi)
return {
"esi_level": cag_esi,
"esi_label": ESI_LABEL[cag_esi],
"sla_minutes": ESI_SLA[cag_esi],
"confidence": 1.0,
"method": "CAG_BYPASS",
"cag_keyword": cag_kw,
"specialty": specialty,
"bert_probs": None,
"latency_ms": round((time.time() - t_total) * 1000, 1),
}
# Step 2: BERT inference
bert_esi, conf, bert_latency, all_probs = bert_classify(req.symptom_text)
# CAG may have a lower-priority hint (ESI 3-5) β use whichever is more urgent
final_esi = bert_esi
method = "BERT"
if cag_esi and PRIORITY.index(cag_esi) < PRIORITY.index(bert_esi):
final_esi = cag_esi
method = "CAG+BERT"
specialty = detect_specialty(req.symptom_text, final_esi)
return {
"esi_level": final_esi,
"esi_label": ESI_LABEL[final_esi],
"sla_minutes": ESI_SLA[final_esi],
"confidence": conf,
"method": method,
"cag_keyword": cag_kw,
"specialty": specialty,
"bert_probs": all_probs,
"latency_ms": round((time.time() - t_total) * 1000, 1),
}
@app.post("/route")
def route(req: RouteRequest):
"""
KAG routing: given ESI + specialty + zone + alpha,
returns top 10 hospitals ranked by routing score.
"""
hospitals = get_top_hospitals(
specialty=req.specialty,
zone=req.zone,
alpha=req.alpha,
esi=req.esi_level,
top_n=10,
)
return {
"zone": req.zone,
"specialty": req.specialty,
"esi_level": req.esi_level,
"alpha": req.alpha,
"hospitals": hospitals,
"total": len(hospitals),
}
@app.get("/zones")
def zones():
counts = {}
for z in ZONES:
avail = sum(1 for h in HOSPITALS if h["zone"] == z and h["availability"])
counts[z] = {"total": 40, "available": avail}
return {"zones": ZONES, "counts": counts}
|