Docker_MEDSCAN.AI / feature_engineering.py
direkKakkar's picture
Upload folder using huggingface_hub
2a3c2f0 verified
Raw History Blame Contribute Delete
5.57 kB
import logging
from typing import Any, Dict, List, Tuple
"""
Logger Set-up
"""
logger = logging.getLogger(__name__)
"""
Unit Normalization Data
"""
UNIT_NORMALIZATION = {
"glucose": ("mg/dL", {"mmol/L": 18.0182}),
"bun": ("mg/dL", {"mmol/L": 2.8}),
"creatinine": ("mg/dL", {"µmol/L": 0.0113}),
"vitamin_d": ("ng/mL", {"nmol/L": 2.496}),
"testosterone": ("ng/dL", {"nmol/L": 28.8}),
"total_cholesterol": ("mg/dL", {"mmol/L": 38.67}),
"ldl": ("mg/dL", {"mmol/L": 38.67}),
"hdl": ("mg/dL", {"mmol/L": 38.67}),
"triglycerides": ("mg/dL", {"mmol/L": 88.57}),
}
"""
Feature definitions per report
"""
BLOOD_FEATURES: List[Tuple[str, str, float]] = [
("wbc", "wbc", 7.0),
("rbc", "rbc", 4.9),
("hemoglobin", "hemoglobin", 14.0),
("hematocrit", "hematocrit", 43.0),
("platelets", "platelets", 250.0),
("glucose", "glucose", 90.0),
("creatinine", "creatinine", 1.0),
("bun", "bun", 15.0)
]
LIPID_FEATURES: List[Tuple[str, str, float]] = [
("total_cholesterol", "total_cholesterol", 180.0),
("hdl", "hdl", 55.0),
("ldl", "ldl", 100.0),
("triglycerides", "triglycerides", 130.0),
("vldl", "vldl", 26.0)
]
VITAMIND_FEATURES: List[Tuple[str, str, float]] = [
("vitamin_d", "vitamin_d", 30.0)
]
HORMONE_FEATURES: List[Tuple[str, str, float]] = [
("tsh", "tsh", 2.5),
("t3", "t3", 1.2),
("t4", "t4", 8.0),
("testosterone", "testosterone", 500.0),
("estradiol", "estradiol", 25.0),
("progesterone", "progesterone", 1.0),
("prolactin", "prolactin", 10.0),
("lh", "lh", 5.0),
("fsh", "fsh", 5.0),
("cortisol", "cortisol", 10.0)
]
KIDNEY_FEATURES: List[Tuple[str, str, float]] = [
("creatinine", "creatinine", 1.0),
("bun", "bun", 15.0),
("urea", "urea", 25.0),
("uric_acid", "uric_acid", 5.0),
("egfr", "egfr", 90.0)
]
LIVER_FEATURES: List[Tuple[str, str, float]] = [
("alt", "alt", 25.0),
("ast", "ast", 25.0),
("alp", "alp", 100.0),
("bilirubin_total", "bilirubin_total", 1.0),
("bilirubin_direct", "bilirubin_direct", 0.3),
("albumin", "albumin", 4.5),
("total_protein", "total_protein", 7.0)
]
GENERAL_FEATURES: List[Tuple[str, str, float]] = []
FEATURE_MAP: Dict[str, List[Tuple[str, str, float]]] = {
"blood": BLOOD_FEATURES,
"lipid": LIPID_FEATURES,
"vitamin_d": VITAMIND_FEATURES,
"hormone": HORMONE_FEATURES,
"kidney": KIDNEY_FEATURES,
"liver": LIVER_FEATURES,
"general": GENERAL_FEATURES
}
"""
Helper Functions
"""
def infer_unit(key: str, value: float) -> str:
"""
Automatically detects unit when missing based on value ranges.
"""
if key == "glucose":
return "mmol/L" if value < 20 else "mg/dL"
if key == "vitamin_d":
return "nmol/L" if value > 200 else "ng/mL"
return "unknown"
def validate_value(key: str, value: float) -> float:
if key == "glucose" and not (20 <= value <= 600):
return 90.0
if key == "hemoglobin" and not (3 <= value <= 25):
return 14.0
if key == "creatinine" and not (0.1 <= value <= 15):
return 1.0
return value
"""
Public Functions
"""
def normalize_units(metrics: Dict[str, Any]) -> Dict[str, Any]:
normalized = {}
for key, entry in metrics.items():
if isinstance(entry, dict):
value = entry.get("value")
unit = entry.get("unit")
if value is None:
continue
if key in UNIT_NORMALIZATION:
std_unit, conversions = UNIT_NORMALIZATION[key]
# infer unit if missing
if unit is None:
unit = infer_unit(key, float(value))
if unit == std_unit:
normalized[key] = {"value": float(value), "unit": std_unit}
elif unit in conversions:
normalized[key] = {
"value": float(value) * conversions[unit],
"unit": std_unit
}
else:
# fallback (important)
normalized[key] = {
"value": float(value),
"unit": std_unit
}
else:
normalized[key] = {"value": float(value), "unit": unit}
else:
normalized[key] = {"value": float(entry), "unit": "unknown"}
return normalized
def build_feature_vector(
metrics: Dict[str, Any],
report_type: str
) -> Tuple[List[float], List[str]]:
feature_defs = FEATURE_MAP.get(report_type, GENERAL_FEATURES)
normalized_metrics = normalize_units(metrics)
feature_vector = []
feature_names = []
for feat_name, metrics_key, default in feature_defs:
entry = normalized_metrics.get(metrics_key)
if isinstance(entry, dict):
value = float(entry.get("value", default))
elif entry is not None:
value = float(entry)
else:
value = default
value = validate_value(metrics_key, value)
feature_vector.append(value)
feature_names.append(feat_name)
return feature_vector, feature_names