"""합성 지방조직 multi-channel WSI 생성 모듈. 채널 구성: (perilipin, F4/80, CD11c, CD206, DAPI) 그룹별 dead adipocyte 비율과 M1/M2 분극 비율을 다르게 설정해 heuristic 기반 CLS detection / polarization 분석을 시연한다. Reproducible: numpy seed = 42. """ from __future__ import annotations import math from dataclasses import dataclass, field from typing import Dict, List, Tuple import numpy as np CHANNELS: Tuple[str, ...] = ("perilipin", "F480", "CD11c", "CD206", "DAPI") # 그룹별 시뮬레이션 파라미터 GROUP_PARAMS: Dict[str, Dict[str, float]] = { "control": {"dead_frac": 0.02, "m1_ratio": 0.30, "adipo_radius_mean": 55.0}, "HFD": {"dead_frac": 0.12, "m1_ratio": 0.70, "adipo_radius_mean": 75.0}, "HFD+drug": {"dead_frac": 0.05, "m1_ratio": 0.50, "adipo_radius_mean": 60.0}, } DEPOTS: Tuple[str, ...] = ("eWAT", "iWAT", "BAT") @dataclass class SyntheticWSI: """합성 지방조직 mini WSI 컨테이너.""" image: np.ndarray # shape = (H, W, len(CHANNELS)), float32 [0,1] adipocytes: List[Dict] # cx, cy, r, dead(bool) cls_truth: List[Dict] # cx, cy, r (dead adipocyte 주변) nuclei: List[Dict] # cx, cy, m1(bool) group: str depot: str mouse_id: str pixel_um: float = 0.5 # 1 px = 0.5 µm 가정 meta: Dict = field(default_factory=dict) def _seeded_rng(seed: int) -> np.random.Generator: return np.random.default_rng(seed) def _draw_disk(canvas: np.ndarray, cx: float, cy: float, r: float, value: float) -> None: h, w = canvas.shape[:2] y0 = max(int(cy - r) - 1, 0) y1 = min(int(cy + r) + 2, h) x0 = max(int(cx - r) - 1, 0) x1 = min(int(cx + r) + 2, w) if y1 <= y0 or x1 <= x0: return yy, xx = np.ogrid[y0:y1, x0:x1] mask = (yy - cy) ** 2 + (xx - cx) ** 2 <= r * r canvas[y0:y1, x0:x1][mask] = value def _draw_ring(canvas: np.ndarray, cx: float, cy: float, r_inner: float, r_outer: float, value: float) -> None: h, w = canvas.shape[:2] y0 = max(int(cy - r_outer) - 1, 0) y1 = min(int(cy + r_outer) + 2, h) x0 = max(int(cx - r_outer) - 1, 0) x1 = min(int(cx + r_outer) + 2, w) if y1 <= y0 or x1 <= x0: return yy, xx = np.ogrid[y0:y1, x0:x1] d2 = (yy - cy) ** 2 + (xx - cx) ** 2 mask = (d2 >= r_inner * r_inner) & (d2 <= r_outer * r_outer) canvas[y0:y1, x0:x1][mask] = np.maximum(canvas[y0:y1, x0:x1][mask], value) def _pack_circles(rng: np.random.Generator, h: int, w: int, radius_mean: float, radius_std: float = 8.0, jitter: float = 4.0, padding: int = 2) -> List[Tuple[float, float, float]]: """격자 + 흔들림으로 비겹침 원형 adipocyte 배치를 생성.""" spacing = radius_mean * 2 + padding out: List[Tuple[float, float, float]] = [] y = radius_mean row = 0 while y <= h + radius_mean * 0.3: offset = (row % 2) * (spacing / 2) x = radius_mean + offset while x <= w + radius_mean * 0.3: r = max(15.0, float(rng.normal(radius_mean, radius_std))) cx = x + float(rng.uniform(-jitter, jitter)) cy = y + float(rng.uniform(-jitter, jitter)) out.append((cx, cy, r)) x += spacing y += spacing * math.sqrt(3) / 2 row += 1 return out def synthesize_wsi(group: str, depot: str, mouse_id: str, size: int = 512, seed: int = 42) -> SyntheticWSI: """단일 합성 지방조직 mini WSI 생성.""" if group not in GROUP_PARAMS: raise ValueError(f"unknown group: {group}") if depot not in DEPOTS: raise ValueError(f"unknown depot: {depot}") params = GROUP_PARAMS[group] rng = _seeded_rng(seed) # depot에 따른 adipocyte 평균 크기 보정 radius_mean = params["adipo_radius_mean"] if depot == "iWAT": radius_mean *= 0.9 elif depot == "BAT": radius_mean *= 0.55 # BAT는 multilocular: 의도적으로 작게 표현 h = w = size img = np.zeros((h, w, len(CHANNELS)), dtype=np.float32) # 약한 노이즈 배경 img += rng.normal(0.05, 0.01, img.shape).astype(np.float32) img = np.clip(img, 0.0, 1.0) circles = _pack_circles(rng, h, w, radius_mean=radius_mean, radius_std=8.0) adipocytes: List[Dict] = [] cls_truth: List[Dict] = [] nuclei: List[Dict] = [] dead_frac = params["dead_frac"] m1_ratio = params["m1_ratio"] for cx, cy, r in circles: is_dead = bool(rng.uniform() < dead_frac) adipocytes.append({"cx": cx, "cy": cy, "r": r, "dead": is_dead}) if is_dead: # dead adipocyte: perilipin- (drop), F4/80+ shell, nuclei ring # perilipin 채널은 0 # F4/80 shell ring _draw_ring(img[..., CHANNELS.index("F480")], cx, cy, r * 1.0, r * 1.25, value=0.85) # nuclei (DAPI) 환형 5-10개 n_nuc = int(rng.integers(5, 11)) cls_truth.append({"cx": cx, "cy": cy, "r": r, "n_nuclei": n_nuc}) for k in range(n_nuc): theta = 2 * math.pi * k / n_nuc + float(rng.uniform(-0.2, 0.2)) nr = r * 1.12 + float(rng.uniform(-3, 3)) nx = cx + nr * math.cos(theta) ny = cy + nr * math.sin(theta) _draw_disk(img[..., CHANNELS.index("DAPI")], nx, ny, 4.5, 0.95) # M1/M2 마커: 같은 위치에 CD11c 또는 CD206 is_m1 = bool(rng.uniform() < m1_ratio) if is_m1: _draw_disk(img[..., CHANNELS.index("CD11c")], nx, ny, 5.0, 0.85) else: _draw_disk(img[..., CHANNELS.index("CD206")], nx, ny, 5.0, 0.85) nuclei.append({"cx": nx, "cy": ny, "m1": is_m1, "near_dead": True}) else: # 정상 adipocyte: perilipin+ ring (membrane) _draw_ring(img[..., CHANNELS.index("perilipin")], cx, cy, r * 0.92, r * 1.0, value=0.9) # 가끔 단일 nucleus (stromal) 흩뿌리기 if rng.uniform() < 0.25: theta = float(rng.uniform(0, 2 * math.pi)) nr = r * 1.08 nx = cx + nr * math.cos(theta) ny = cy + nr * math.sin(theta) _draw_disk(img[..., CHANNELS.index("DAPI")], nx, ny, 4.0, 0.7) nuclei.append({"cx": nx, "cy": ny, "m1": False, "near_dead": False}) img = np.clip(img, 0.0, 1.0) return SyntheticWSI( image=img, adipocytes=adipocytes, cls_truth=cls_truth, nuclei=nuclei, group=group, depot=depot, mouse_id=mouse_id, meta={"size": size, "seed": seed, "radius_mean": radius_mean}, ) def build_demo_cohort(size: int = 512, base_seed: int = 42) -> List[SyntheticWSI]: """3 groups × 3 mice × 3 depots = 27 mini WSIs.""" cohort: List[SyntheticWSI] = [] seed = base_seed for group in GROUP_PARAMS.keys(): for mouse_idx in range(3): mouse_id = f"{group}_m{mouse_idx + 1}" for depot in DEPOTS: wsi = synthesize_wsi(group=group, depot=depot, mouse_id=mouse_id, size=size, seed=seed) cohort.append(wsi) seed += 1 return cohort