Scientific Image Analysis

2. セグメンテーション

2.1 Otsu + 形態演算

def otsu_segmentation(image, min_size=50, closing_disk=3):
    """
    Otsu 閾値 + 形態学的クリーニングによるバイナリセグメンテーション。
    """
    threshold = filters.threshold_otsu(image)
    binary = image > threshold

    # 形態学的クリーニング
    binary = morphology.binary_closing(binary, morphology.disk(closing_disk))
    binary = morphology.remove_small_objects(binary, min_size=min_size)
    binary = ndimage.binary_fill_holes(binary)

    # ラベリング
    labels = measure.label(binary)

    return binary, labels, threshold

2.2 Watershed セグメンテーション

from skimage.feature import peak_local_max

def watershed_segmentation(image, min_distance=10, footprint_size=3):
    """
    Watershed による接触粒子の分離セグメンテーション。
    """
    threshold = filters.threshold_otsu(image)
    binary = image > threshold
    binary = ndimage.binary_fill_holes(binary)

    # 距離変換
    distance = ndimage.distance_transform_edt(binary)

    # ローカルマキシマ
    coords = peak_local_max(distance, min_distance=min_distance,
                             labels=binary)
    mask = np.zeros(distance.shape, dtype=bool)
    mask[tuple(coords.T)] = True
    markers = measure.label(mask)

    # Watershed
    labels = watershed(-distance, markers, mask=binary)

    return labels, distance

3. 形態計測

def morphometric_analysis(labels, pixel_size_um=1.0):
    """
    ラベル画像から各オブジェクトの形態パラメータを計測する。

    計測項目:
        - area: 面積 (μm²)
        - perimeter: 周囲長 (μm)
        - equivalent_diameter: 等価円直径 (μm)
        - circularity: 真円度 4π·A/P²
        - aspect_ratio: アスペクト比 (major/minor axis)
        - solidity: 充填率 (area/convex_area)
        - eccentricity: 離心率
    """
    import pandas as pd

    props = measure.regionprops(labels)
    records = []

    for p in props:
        area = p.area * (pixel_size_um ** 2)
        perimeter = p.perimeter * pixel_size_um
        eq_diameter = p.equivalent_diameter * pixel_size_um
        circularity = (4 * np.pi * p.area) / (p.perimeter**2 + 1e-10)
        aspect_ratio = p.major_axis_length / (p.minor_axis_length + 1e-10)

        records.append({
            "label": p.label,
            "area_um2": area,
            "perimeter_um": perimeter,
            "equivalent_diameter_um": eq_diameter,
            "circularity": circularity,
            "aspect_ratio": aspect_ratio,
            "solidity": p.solidity,
            "eccentricity": p.eccentricity,
            "centroid_y": p.centroid[0],
            "centroid_x": p.centroid[1],
        })

    return pd.DataFrame(records)

4. 粒径分布

def particle_size_distribution(morpho_df, diameter_col="equivalent_diameter_um",
                                 bins=30, figsize=(10, 6)):
    """
    粒径分布のヒストグラムと統計量を算出する。

    出力:
        - ヒストグラム + カーネル密度
        - D10, D50 (中央値), D90
        - 平均, 標準偏差, 変動係数 CV
    """
    diameters = morpho_df[diameter_col].values

    d10 = np.percentile(diameters, 10)
    d50 = np.percentile(diameters, 50)
    d90 = np.percentile(diameters, 90)
    span = (d90 - d10) / d50

    stats = {
        "count": len(diameters),
        "mean_um": np.mean(diameters),
        "std_um": np.std(diameters),
        "cv_pct": np.std(diameters) / np.mean(diameters) * 100,
        "d10_um": d10,
        "d50_um": d50,
        "d90_um": d90,
        "span": span,
    }

    fig, ax = plt.subplots(figsize=figsize)
    ax.hist(diameters, bins=bins, density=True, alpha=0.7, color="steelblue",
            edgecolor="black", linewidth=0.5)

    # KDE
    from scipy.stats import gaussian_kde
    kde = gaussian_kde(diameters)
    x_range = np.linspace(diameters.min(), diameters.max(), 200)
    ax.plot(x_range, kde(x_range), "r-", linewidth=2, label="KDE")

    # D10/D50/D90 線
    for d, label in [(d10, "D10"), (d50, "D50"), (d90, "D90")]:
        ax.axvline(d, color="gray", linestyle="--", alpha=0.7)
        ax.text(d, ax.get_ylim()[1]*0.95, f" {label}={d:.1f}", fontsize=9)

    ax.set_xlabel("Diameter (μm)")
    ax.set_ylabel("Density")
    ax.set_title("Particle Size Distribution", fontweight="bold")
    ax.legend()
    plt.tight_layout()
    plt.savefig("figures/particle_size_distribution.png", dpi=300,
                bbox_inches="tight")
    plt.close()

    return stats

5. テクスチャ解析 (GLCM)

from skimage.feature import graycomatrix, graycoprops

def glcm_texture_features(image, distances=[1, 3, 5], angles=[0, np.pi/4,
                            np.pi/2, 3*np.pi/4], levels=256):
    """
    GLCM (Gray-Level Co-occurrence Matrix) テクスチャ特徴量を抽出する。

    Features: contrast, dissimilarity, homogeneity, energy, correlation, ASM
    """
    img_uint8 = (image * 255).astype(np.uint8) if image.max() <= 1 else image.astype(np.uint8)

    glcm = graycomatrix(img_uint8, distances=distances, angles=angles,
                         levels=levels, symmetric=True, normed=True)

    features = {}
    for prop in ["contrast", "dissimilarity", "homogeneity", "energy",
                 "correlation", "ASM"]:
        values = graycoprops(glcm, prop)
        features[f"{prop}_mean"] = values.mean()
        features[f"{prop}_std"] = values.std()

    return features

6. 蛍光マルチチャネル合成

def merge_fluorescence_channels(channels, colors=None, figsize=(10, 10)):
    """
    蛍光顕微鏡のマルチチャネル画像を合成する。

    Parameters:
        channels: dict {"DAPI": array, "GFP": array, "RFP": array}
        colors: dict {"DAPI": [0,0,1], "GFP": [0,1,0], "RFP": [1,0,0]}
    """
    if colors is None:
        colors = {"DAPI": [0, 0, 1], "GFP": [0, 1, 0], "RFP": [1, 0, 0],
                  "Cy5": [1, 0, 1]}

    ch_names = list(channels.keys())
    shape = list(channels.values())[0].shape
    merged = np.zeros((*shape, 3))

    n = len(ch_names) + 1
    fig, axes = plt.subplots(1, n, figsize=(5*n, 5))

    for i, (name, img) in enumerate(channels.items()):
        # 正規化
        norm_img = (img - img.min()) / (img.max() - img.min() + 1e-10)

        # 個別チャネル表示
        ch_color = np.array(colors.get(name, [1, 1, 1]))
        colored = norm_img[:, :, np.newaxis] * ch_color
        axes[i].imshow(np.clip(colored, 0, 1))
        axes[i].set_title(name, fontweight="bold")
        axes[i].axis("off")

        # マージ画像に加算
        merged += colored

    # 合成画像
    axes[-1].imshow(np.clip(merged, 0, 1))
    axes[-1].set_title("Merged", fontweight="bold")
    axes[-1].axis("off")

    plt.tight_layout()
    plt.savefig("figures/fluorescence_merged.png", dpi=300, bbox_inches="tight")
    plt.close()

    return np.clip(merged, 0, 1)

ToolUniverse 連携

References

Output Files

依存パッケージ

scikit-image>=0.21
scipy>=1.10