2nd/10_Wiki/Topics/AI_and_ML/Neurorehabilitation-Post-Stroke.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-neurorehabilitation-post-stroke	Neurorehabilitation Post-Stroke	10_Wiki/Topics	verified	self	Post-Stroke Rehabilitation Stroke Recovery Neurorehabilitation after Stroke	none	A	0.9	applied	stroke rehab motor-recovery cimt robotics bci plasticity		2026-05-10	pending	language framework python pytorch-mne

Neurorehabilitation Post-Stroke

매 한 줄

• 뇌졸중 후 신경재활은 손상 주변 뉴런의 가소성을 활용하여 운동·언어·인지 기능을 복원하는 다학제 개입이며, 로봇·BCI·집중훈련(CIMT)이 표준 도구로 합류했다.

매 핵심

• 회복 곡선: 발병 후 0–3개월 spontaneous recovery(신경가소성 활성기), 3–6개월 plateau, 6개월+ slow gain. 시간 의존성 강함.
• 표준 개입: CIMT(constraint-induced movement therapy, 건측 억제 + 환측 집중 사용), task-specific repetition(>300회/세션), mirror therapy, FES, robotic exoskeleton(Lokomat, Armeo), tDCS/rTMS 보조.
• BCI 재활: motor imagery EEG → robot/FES 폐루프, Hebbian "pair pre+post" 자극으로 cortico-spinal pathway 강화.
• 결과 척도: Fugl-Meyer Assessment(FMA, 0–66 upper limb), Modified Rankin Scale, Barthel Index, NIHSS.
• AI 적용: 보행 분석(IMU + LSTM), 회복 궤적 예측(GBM on FMA + lesion volume), 가정 telerehab adherence 모니터.

💻 패턴

# Fugl-Meyer trajectory prediction (early FMA + lesion → 6mo FMA)
import numpy as np
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.model_selection import KFold

X = np.load("baseline_features.npy")  # FMA_t0, lesion_vol, age, NIHSS
y = np.load("FMA_6mo.npy")
gbr = GradientBoostingRegressor(n_estimators=300, max_depth=3)
mae = []
for tr, te in KFold(5, shuffle=True, random_state=0).split(X):
    gbr.fit(X[tr], y[tr])
    mae.append(np.mean(np.abs(gbr.predict(X[te]) - y[te])))
print(f"FMA MAE: {np.mean(mae):.2f}")

# IMU gait segmentation: heel-strike detection
import numpy as np
from scipy.signal import find_peaks

def heel_strikes(acc_z, fs=100):
    peaks, _ = find_peaks(-acc_z, distance=fs * 0.4, prominence=0.5)
    return peaks  # samples

# Motor imagery EEG → CSP + LDA classifier (BCI rehab)
from mne.decoding import CSP
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.pipeline import Pipeline

clf = Pipeline([("csp", CSP(n_components=6)), ("lda", LinearDiscriminantAnalysis())])
clf.fit(X_eeg, y_class)  # 0=rest, 1=hand-MI

# CIMT compliance from wrist accelerometer (use ratio paretic/non-paretic)
import numpy as np
def use_ratio(acc_paretic, acc_non):
    p = np.sqrt(np.sum(acc_paretic ** 2, axis=1))
    n = np.sqrt(np.sum(acc_non ** 2, axis=1))
    return p.sum() / (n.sum() + 1e-9)

# Robotic assist-as-needed: error-based force scaling
def assist_force(target, actual, k_max=20.0, deadband=0.02):
    err = target - actual
    if abs(err) < deadband:
        return 0.0
    return k_max * (1 - 1 / (1 + abs(err))) * (1 if err > 0 else -1)

# tDCS dose calc (anodal M1, 2 mA, 20 min)
def tdcs_charge(current_mA, duration_min, area_cm2=35):
    Q = current_mA * 1e-3 * duration_min * 60  # Coulombs
    return Q / area_cm2  # safety < 0.04 C/cm^2

# Repetition counter from EMG burst detection
import numpy as np
def count_reps(emg, fs=1000, thresh=2.0):
    env = np.abs(emg)
    above = env > thresh * env.std()
    edges = np.diff(above.astype(int))
    return int((edges == 1).sum())

매 결정 기준

• 발병 후 시점: 0–3개월 → 고강도 task-specific. 6개월+ → CIMT, BCI, robotics 보강.
• 운동 손상 정도: FMA<20(severe) → robotic passive/assist. FMA 20–50 → CIMT + active. FMA>50 → fine motor task.
• BCI 적용: 잔존 motor imagery 신호 검출 가능 + 환자 동기 → 폐루프 BCI(8주 프로그램).
• 퇴원 후 telerehab: adherence 측정 가능한 wearable + 주 1회 화상 follow.

🔗 Graph

• 관련: Neuroplasticity, Neuroprosthetics-Development, Neurodevelopmental Disorders, BCI

🤖 LLM 활용

• 환자별 재활 계획 초안(FMA, Barthel 입력 → goal/timeline 생성, 임상의 검토 필수).
• 진료기록 요약(NIHSS 변화, 합병증 추출).
• 가정 운동 안내문 다국어 번역.

❌ 안티패턴

• 발병 24h 이내 고강도 mobilization(AVERT 시험: 90일 outcome 악화 보고).
• BCI calibration 없이 일반 모델 적용(개인차 큼).
• 환측 사용 강제 없이 CIMT 라벨링.

🧪 검증

• FMA, Action Research Arm Test(ARAT), 10m walk test, Berg Balance.
• BCI: classification accuracy ≥ 70%, 사용자 NASA-TLX.

🕓 Changelog

• 2026-05-08 Phase 1: 초안 자동 생성.
• 2026-05-10 Manual cleanup: 본문 보강, AVERT 결과 반영, 코드 패턴 7개, BCI 폐루프 추가.