2nd/10_Wiki/Topics/AI_and_ML/Neurodevelopmental Disorders.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-neurodevelopmental-disorders	Neurodevelopmental Disorders	10_Wiki/Topics	verified	self	Neurodevelopmental Disorders ADHD ASD Dyslexia NDD	none	A	0.9	applied	neuroscience psychiatry adhd autism dyslexia dsm5 ai-screening		2026-05-10	pending	language framework python pytorch-scikit-learn

Neurodevelopmental Disorders

매 한 줄

• 신경발달장애(ADHD, ASD, dyslexia)는 DSM-5 기준 발달 초기 발현·지속적 기능 손상이며, AI는 조기 선별·개인화 중재를 보강한다.

매 핵심

• DSM-5-TR 분류: ADHD(주의력결핍/과잉행동), ASD(사회적 의사소통 + 제한 반복 행동), Specific Learning Disorder(dyslexia/dyscalculia/dysgraphia), Intellectual Disability, Communication Disorders, Motor Disorders.
• 신경생물: prefrontal-striatal dopamine 회로(ADHD), cortical underconnectivity & social brain network 이상(ASD), left occipito-temporal "visual word form area" 저활성(dyslexia).
• 유전·환경: heritability ADHD 0.7–0.8, ASD 0.8 이상, dyslexia 0.5–0.7. CNV(16p11.2), SHANK3, FOXP2 등.
• 개입: 행동 중재(ABA, CBT), 약물(methylphenidate, amphetamine, atomoxetine; ASD에 risperidone/aripiprazole 보조), 학습 중재(Orton-Gillingham).
• AI 활용: eye-tracking + ML로 ASD 12–18개월 조기 선별, 음성/언어 분석으로 dyslexia risk score, EEG/fMRI 기반 ADHD 분류.

💻 패턴

# ADHD risk score from CPT (Continuous Performance Task) features
import numpy as np
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import cross_val_score

# features: omission_errors, commission_errors, rt_mean, rt_std, post_error_slowing
X = np.load("cpt_features.npy")  # (n_subjects, 5)
y = np.load("dsm5_adhd_label.npy")  # 0/1, clinician-rated

clf = GradientBoostingClassifier(n_estimators=200, max_depth=3, random_state=0)
scores = cross_val_score(clf, X, y, cv=5, scoring="roc_auc")
print(f"ADHD ROC-AUC: {scores.mean():.3f} ± {scores.std():.3f}")

# ASD early screening: eye-tracking gaze entropy
import numpy as np
from scipy.stats import entropy

def gaze_entropy(fixations_xy, grid=(8, 8)):
    H, _, _ = np.histogram2d(fixations_xy[:, 0], fixations_xy[:, 1], bins=grid)
    p = H.flatten() / H.sum()
    return entropy(p[p > 0])

# ASD 유아: social-scene 응시 시 gaze entropy 낮음(얼굴 회피)
ent = gaze_entropy(fixations_xy)
flag_high_risk = ent < 2.5  # threshold from validated study

# Dyslexia: phonological awareness from speech features
import librosa, numpy as np

def phon_features(wav_path):
    y, sr = librosa.load(wav_path, sr=16000)
    mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13).mean(axis=1)
    rate = len(librosa.effects.split(y, top_db=30)) / (len(y) / sr)
    return np.concatenate([mfcc, [rate]])

# fMRI ROI features for ASD classification (ABIDE-style)
import nilearn.datasets as nd
from nilearn.connectome import ConnectivityMeasure

correlation = ConnectivityMeasure(kind="correlation", vectorize=True)
fc = correlation.fit_transform(timeseries_list)  # (n_subj, n_features)
# downstream: SVM or shallow MLP, AUC ~0.65–0.75 typical

# Methylphenidate dose titration: simple PK model
def mph_concentration(dose_mg, t_hr, ka=1.5, ke=0.3, vd=4.0):
    import numpy as np
    return (dose_mg * ka) / (vd * (ka - ke)) * (np.exp(-ke * t_hr) - np.exp(-ka * t_hr))

# DSM-5 ADHD criterion checker (≥6 of 9 inattention OR hyperactivity-impulsivity)
def adhd_dsm5(symptoms_inatt: list[bool], symptoms_hyper: list[bool], age: int):
    threshold = 6 if age < 17 else 5
    inatt = sum(symptoms_inatt) >= threshold
    hyper = sum(symptoms_hyper) >= threshold
    return inatt or hyper

매 결정 기준

• ASD 조기 선별: 18–24개월 M-CHAT-R + eye-tracking 보조 → 양성 시 ADOS-2.
• ADHD 약물 선택: 자극제 1차(methylphenidate/amphetamine), 비자극제(atomoxetine, guanfacine) 부작용·동반질환 시.
• Dyslexia 진단: IQ 정상이면서 reading achievement < 1.5 SD, phonological deficit 확인.
• AI 도구 사용 한계: screening 보조용. 단독 진단 도구로 사용 금지(FDA/PMDA 승인 외).

🔗 Graph

• 관련: Neuroplasticity, Neurorehabilitation-Post-Stroke

🤖 LLM 활용

• DSM-5 기준 충족 여부 점검(증상 리스트 → 조건 매칭).
• 부모 면담 transcript에서 ADHD 행동 빈도 추출(zero-shot classifier).
• 임상의의 진단 대체 금지: 보조 요약·문헌 검색에 한정.

❌ 안티패턴

• 단일 EEG/fMRI feature로 진단 라벨 출력(false positive 다발).
• 약물 권고를 LLM이 직접 생성(처방권 위반).
• AI screening 양성을 부모에게 "확진"으로 전달.

🧪 검증

• 진단 도구: M-CHAT-R(ASD), Conners-3(ADHD), CTOPP-2(dyslexia)와의 일치도(κ ≥ 0.6).
• 약물 반응: 4–6주 follow-up Conners 점수 ≥ 25% 감소.

🕓 Changelog

• 2026-05-08 Phase 1: 초안 자동 생성.
• 2026-05-10 Manual cleanup: 본문 보강(150–250 lines), 코드 패턴 6, DSM-5-TR 반영, AI 한계 명시.