@nahisaho/satori 0.9.0 → 0.11.0

package/src/.github/skills/scientific-neuroscience-electrophysiology/SKILL.md

@@ -0,0 +1,400 @@
+---
+name: scientific-neuroscience-electrophysiology
+description: |
+  神経科学・電気生理学解析スキル。Neuropixels/マルチ電極アレイの
+  スパイクソーティング (SpikeInterface + Kilosort4)、品質指標 (SNR/ISI 違反率)、
+  EEG マイクロステート・事象関連電位 (MNE-Python)、ECG HRV 解析、
+  EDA 皮膚コンダクタンス応答、EMG 筋活動解析 (NeuroKit2)、
+  脳結合性解析 (機能的/実効的) を統合した神経科学パイプライン。
+---
+
+# Scientific Neuroscience & Electrophysiology
+
+Neuropixels 高密度電極記録・EEG・ECG・EMG・EDA 等の神経生理信号を対象に、
+スパイクソーティング→品質管理→特徴抽出→統計解析の
+標準パイプラインを提供する。SpikeInterface + NeuroKit2 + MNE-Python ベース。
+
+## When to Use
+
+- Neuropixels/MEA のスパイクソーティングパイプラインが必要なとき
+- EEG マイクロステート・事象関連電位 (ERP) 解析を行うとき
+- ECG からの心拍変動 (HRV) 解析・自律神経系評価が必要なとき
+- EDA (皮膚電気活動) の交感神経応答解析を行うとき
+- 脳結合性 (機能的/実効的) の推定が必要なとき
+
+---
+
+## Quick Start
+
+## 1. Neuropixels スパイクソーティング (SpikeInterface)
+
+```python
+import numpy as np
+import spikeinterface.full as si
+
+
+def neuropixels_preprocessing(recording_path, probe_type="NP1",
+                               output_dir="results/neuroscience"):
+    """
+    Neuropixels 記録の前処理パイプライン (SpikeInterface)。
+
+    1. SpikeGLX/OpenEphys ファイル読み込み
+    2. バンドパスフィルタ (300-6000 Hz)
+    3. 位相シフト補正 (ADC multiplexing)
+    4. CMR (Common Median Reference) ノイズ除去
+    5. ドリフト補正 (Motion correction)
+    """
+    import os
+    os.makedirs(output_dir, exist_ok=True)
+
+    # 記録読み込み
+    recording = si.read_spikeglx(recording_path)
+    print(f"  Probe: {probe_type}")
+    print(f"  Channels: {recording.get_num_channels()}")
+    print(f"  Sampling rate: {recording.get_sampling_frequency()} Hz")
+    print(f"  Duration: {recording.get_total_duration():.1f} sec")
+
+    # 前処理チェーン
+    rec_filt = si.bandpass_filter(recording, freq_min=300, freq_max=6000)
+    rec_shift = si.phase_shift(rec_filt)  # ADC 位相補正
+    rec_cmr = si.common_reference(rec_shift, reference="global",
+                                   operator="median")
+
+    # モーション補正 (非剛体)
+    motion_info = si.correct_motion(
+        rec_cmr, preset="nonrigid_accurate",
+        output_motion_info=True
+    )
+
+    print(f"  Preprocessing complete: filter → phase_shift → CMR → motion")
+
+    return rec_cmr
+
+
+def spike_sorting_kilosort4(recording, output_dir="results/spike_sorting"):
+    """
+    Kilosort4 によるスパイクソーティング。
+
+    SpikeInterface wrapper を使用。
+    """
+    import os
+    os.makedirs(output_dir, exist_ok=True)
+
+    sorting = si.run_sorter(
+        sorter_name="kilosort4",
+        recording=recording,
+        output_folder=output_dir,
+        verbose=True,
+    )
+
+    print(f"  Kilosort4 results:")
+    print(f"    Units found: {len(sorting.get_unit_ids())}")
+
+    return sorting
+
+
+def spike_sorting_quality_metrics(sorting, recording,
+                                   output_dir="results/spike_sorting"):
+    """
+    スパイクソーティング品質指標。
+
+    ENCODE/Allen Institute 基準:
+    - SNR ≥ 5: 高品質ユニット
+    - ISI violation rate < 0.5%: 良好な分離
+    - Presence ratio > 0.9: 安定した記録
+    - Amplitude cutoff < 0.1
+    """
+    we = si.extract_waveforms(recording, sorting, folder=f"{output_dir}/waveforms")
+    metrics = si.compute_quality_metrics(we)
+
+    # Allen Institute 基準フィルタ
+    good_units = metrics[
+        (metrics["snr"] >= 5) &
+        (metrics["isi_violations_ratio"] < 0.005) &
+        (metrics["presence_ratio"] > 0.9) &
+        (metrics["amplitude_cutoff"] < 0.1)
+    ]
+
+    print(f"  Quality metrics computed for {len(metrics)} units")
+    print(f"  Good units (Allen criteria): {len(good_units)} / {len(metrics)}")
+    print(f"  Mean SNR: {metrics['snr'].mean():.1f}")
+    print(f"  Mean ISI violation rate: {metrics['isi_violations_ratio'].mean():.4f}")
+    print(f"  Mean firing rate: {metrics['firing_rate'].mean():.1f} Hz")
+
+    return metrics, good_units
+```
+
+## 2. EEG 解析 (MNE-Python)
+
+```python
+import numpy as np
+import mne
+
+
+def eeg_preprocessing(raw_file, montage="standard_1020",
+                       l_freq=1.0, h_freq=40.0):
+    """
+    EEG 前処理パイプライン (MNE-Python)。
+
+    1. ファイル読み込み (BDF/EDF/FIF/BrainVision)
+    2. モンタージュ設定
+    3. バンドパスフィルタ (1-40 Hz)
+    4. ICA によるアーティファクト除去 (眼球運動, 心拍)
+    5. 再参照 (平均参照)
+    """
+    raw = mne.io.read_raw(raw_file, preload=True)
+    raw.set_montage(montage)
+
+    print(f"  Channels: {len(raw.ch_names)}")
+    print(f"  Sampling rate: {raw.info['sfreq']} Hz")
+    print(f"  Duration: {raw.times[-1]:.1f} sec")
+
+    # バンドパスフィルタ
+    raw.filter(l_freq=l_freq, h_freq=h_freq, fir_design="firwin")
+
+    # ICA アーティファクト除去
+    ica = mne.preprocessing.ICA(n_components=20, random_state=42)
+    ica.fit(raw)
+
+    # EOG/ECG コンポーネント自動検出
+    eog_indices, eog_scores = ica.find_bads_eog(raw)
+    ica.exclude = eog_indices
+    raw_clean = ica.apply(raw.copy())
+
+    print(f"  ICA components excluded: {len(eog_indices)} (EOG artifacts)")
+
+    # 平均参照
+    raw_clean.set_eeg_reference("average")
+
+    return raw_clean
+
+
+def eeg_erp_analysis(raw, events, event_id, tmin=-0.2, tmax=0.8,
+                      baseline=(-0.2, 0)):
+    """
+    事象関連電位 (ERP) 解析。
+
+    Parameters:
+        events: MNE events 配列
+        event_id: イベント辞書 (e.g., {"target": 1, "standard": 2})
+    """
+    epochs = mne.Epochs(raw, events, event_id,
+                        tmin=tmin, tmax=tmax,
+                        baseline=baseline,
+                        reject=dict(eeg=100e-6),
+                        preload=True)
+
+    print(f"  Epochs: {len(epochs)} / {len(events)} (after rejection)")
+
+    evokeds = {}
+    for condition in event_id:
+        evoked = epochs[condition].average()
+        evokeds[condition] = evoked
+
+        # ピーク検出
+        ch_name, latency, amplitude = evoked.get_peak(
+            ch_type="eeg", tmin=0.05, tmax=0.5
+        )
+        print(f"  {condition}: peak at {latency*1000:.0f} ms, "
+              f"{amplitude*1e6:.1f} µV ({ch_name})")
+
+    return epochs, evokeds
+
+
+def eeg_microstate_analysis(raw, n_states=4):
+    """
+    EEG マイクロステート解析。
+
+    脳の準安定状態 (4-7 マイクロステート A-G) を GFP ピークから同定。
+    各マイクロステートの出現頻度・持続時間・遷移確率を算出。
+    """
+    from sklearn.cluster import KMeans
+
+    # GFP (Global Field Power) 計算
+    data = raw.get_data(picks="eeg")
+    gfp = np.std(data, axis=0)
+
+    # GFP ピーク抽出
+    from scipy.signal import find_peaks
+    peaks, _ = find_peaks(gfp, height=np.percentile(gfp, 50))
+
+    # K-means クラスタリング
+    peak_maps = data[:, peaks].T
+    kmeans = KMeans(n_clusters=n_states, random_state=42, n_init=50)
+    labels = kmeans.fit_predict(peak_maps)
+
+    # 各マイクロステートの統計
+    print(f"  Microstate analysis ({n_states} states):")
+    for i in range(n_states):
+        ms_label = chr(65 + i)  # A, B, C, D, ...
+        count = np.sum(labels == i)
+        pct = count / len(labels)
+        print(f"    State {ms_label}: {pct:.1%} occurrence")
+
+    return kmeans, labels
+```
+
+## 3. ECG 心拍変動解析 (NeuroKit2)
+
+```python
+import numpy as np
+import pandas as pd
+import neurokit2 as nk
+
+
+def ecg_hrv_analysis(ecg_signal, sampling_rate=1000):
+    """
+    ECG からの心拍変動 (HRV) 解析パイプライン。
+
+    1. R ピーク検出
+    2. RR 間隔算出
+    3. 時間領域 HRV (SDNN, RMSSD, pNN50)
+    4. 周波数領域 HRV (VLF, LF, HF, LF/HF 比)
+    5. 非線形 HRV (SD1, SD2, ApEn, SampEn)
+    """
+    # ECG 前処理 + R ピーク検出
+    ecg_cleaned = nk.ecg_clean(ecg_signal, sampling_rate=sampling_rate)
+    signals, info = nk.ecg_process(ecg_cleaned, sampling_rate=sampling_rate)
+
+    r_peaks = info["ECG_R_Peaks"]
+    print(f"  R-peaks detected: {len(r_peaks)}")
+    print(f"  Mean HR: {signals['ECG_Rate'].mean():.1f} bpm")
+
+    # HRV 解析 (時間 + 周波数 + 非線形)
+    hrv_time = nk.hrv_time(signals, sampling_rate=sampling_rate)
+    hrv_freq = nk.hrv_frequency(signals, sampling_rate=sampling_rate)
+    hrv_nonlinear = nk.hrv_nonlinear(signals, sampling_rate=sampling_rate)
+
+    hrv_results = pd.concat([hrv_time, hrv_freq, hrv_nonlinear], axis=1)
+
+    print(f"  Time-domain HRV:")
+    print(f"    SDNN: {hrv_results['HRV_SDNN'].values[0]:.1f} ms")
+    print(f"    RMSSD: {hrv_results['HRV_RMSSD'].values[0]:.1f} ms")
+    print(f"    pNN50: {hrv_results['HRV_pNN50'].values[0]:.1f}%")
+    print(f"  Frequency-domain HRV:")
+    print(f"    LF/HF ratio: {hrv_results['HRV_LFHF'].values[0]:.2f}")
+
+    return hrv_results, signals
+
+
+def eda_sympathetic_analysis(eda_signal, sampling_rate=1000):
+    """
+    EDA (皮膚電気活動) 解析 — 交感神経系応答。
+
+    1. トニック成分 (Skin Conductance Level, SCL) 分離
+    2. ファジック成分 (Skin Conductance Response, SCR) 検出
+    3. SCR ピーク振幅・潜時・回復時間
+    """
+    eda_cleaned = nk.eda_clean(eda_signal, sampling_rate=sampling_rate)
+    signals, info = nk.eda_process(eda_cleaned, sampling_rate=sampling_rate)
+
+    scr_peaks = info["SCR_Peaks"]
+    print(f"  SCR peaks detected: {len(scr_peaks)}")
+    print(f"  Mean SCL (tonic): {signals['EDA_Tonic'].mean():.3f} µS")
+    if len(scr_peaks) > 0:
+        print(f"  Mean SCR amplitude: {signals['SCR_Amplitude'].dropna().mean():.3f} µS")
+
+    return signals, info
+```
+
+## 4. 脳機能結合性解析
+
+```python
+import numpy as np
+import pandas as pd
+
+
+def functional_connectivity(epochs, method="wpli2_debiased",
+                              fmin=8, fmax=13):
+    """
+    EEG/MEG 脳機能的結合性の推定。
+
+    Methods:
+    - coh: Coherence
+    - imcoh: Imaginary Coherence
+    - plv: Phase Locking Value
+    - wpli: Weighted Phase Lag Index
+    - wpli2_debiased: Debiased Squared WPLI (推奨)
+    """
+    from mne_connectivity import spectral_connectivity_epochs
+
+    con = spectral_connectivity_epochs(
+        epochs, method=method,
+        fmin=fmin, fmax=fmax,
+        faverage=True, n_jobs=-1
+    )
+
+    conn_matrix = con.get_data(output="dense")[:, :, 0]
+
+    print(f"  Connectivity method: {method}")
+    print(f"  Frequency band: {fmin}-{fmax} Hz")
+    print(f"  Matrix shape: {conn_matrix.shape}")
+    print(f"  Mean connectivity: {np.mean(conn_matrix[np.triu_indices_from(conn_matrix, k=1)]):.3f}")
+
+    return conn_matrix, con
+
+
+def graph_theory_brain_network(conn_matrix, ch_names, threshold=0.3):
+    """
+    脳ネットワークのグラフ理論的解析。
+
+    - Global efficiency
+    - Clustering coefficient
+    - Small-world index
+    - Hub 同定 (degree centrality)
+    """
+    import networkx as nx
+
+    n = conn_matrix.shape[0]
+    G = nx.Graph()
+    for i in range(n):
+        G.add_node(ch_names[i])
+        for j in range(i + 1, n):
+            if conn_matrix[i, j] > threshold:
+                G.add_edge(ch_names[i], ch_names[j],
+                           weight=conn_matrix[i, j])
+
+    # グラフ指標
+    eff = nx.global_efficiency(G)
+    cc = nx.average_clustering(G, weight="weight")
+    degree = dict(G.degree(weight="weight"))
+    top_hubs = sorted(degree.items(), key=lambda x: x[1], reverse=True)[:5]
+
+    print(f"  Brain network: {G.number_of_nodes()} nodes, {G.number_of_edges()} edges")
+    print(f"  Global efficiency: {eff:.3f}")
+    print(f"  Clustering coefficient: {cc:.3f}")
+    print(f"  Top hubs: {[h[0] for h in top_hubs]}")
+
+    return G
+```
+
+## References
+
+### Output Files
+
+| ファイル | 形式 |
+|---|---|
+| `results/spike_sorting/sorting_results.npz` | NPZ |
+| `results/spike_sorting/quality_metrics.csv` | CSV |
+| `results/eeg/erp_evokeds.fif` | FIF |
+| `results/eeg/microstates.csv` | CSV |
+| `results/ecg/hrv_results.csv` | CSV |
+| `results/eda/scr_peaks.csv` | CSV |
+| `results/connectivity/conn_matrix.csv` | CSV |
+| `figures/spike_rasters.png` | PNG |
+| `figures/erp_waveforms.png` | PNG |
+| `figures/hrv_poincare.png` | PNG |
+
+### 参照スキル
+
+| スキル | 関連 |
+|---|---|
+| `scientific-biosignal-processing` | 汎用生体信号処理 |
+| `scientific-spectral-signal` | スペクトル・周波数解析 |
+| `scientific-time-series` | 時系列分解・予測 |
+| `scientific-network-analysis` | グラフ理論・ネットワーク指標 |
+| `scientific-deep-learning` | DL ベーススパイク分類 |
+
+### 依存パッケージ
+
+`spikeinterface`, `mne`, `mne-connectivity`, `neurokit2`, `numpy`, `pandas`, `scipy`, `scikit-learn`, `networkx`