"""CholangioDuctReact v0 — Streamlit 진입점. MASLD/MASH/PBC/PSC/BDL/CDAA/CDAHFD 마우스 모델 간 CK19/SOX9/EpCAM IHC + paired H&E/Sirius Red WSI → portal tract 기반 ductular reaction 정량 + type 분류 standalone 도구. 본 도구는 연구·참고용. 임상 진단 사용 금지. """ from __future__ import annotations import io import json from pathlib import Path from typing import List import numpy as np try: import streamlit as st except Exception as e: # pragma: no cover st = None # type: ignore from lib import synth_liver, registration, portal_detect, dr_classify, fibrosis_overlay, stats PROJECT_DIR = Path(__file__).parent DEMO_PATH = PROJECT_DIR / "data" / "demo_cohort.json" def _slide_to_rgb_he(slide: synth_liver.SynthSlide) -> np.ndarray: return (slide.he * 255).clip(0, 255).astype(np.uint8) def _gray_to_rgb(gray: np.ndarray, color=(0.0, 0.6, 0.0)) -> np.ndarray: """단일 채널을 가짜 IHC 색조로 변환 (CK19=brown, SOX9=blue 등).""" rgb = np.zeros((*gray.shape, 3), dtype=np.float32) for c in range(3): rgb[..., c] = (1 - gray) + gray * color[c] return (np.clip(rgb, 0, 1) * 255).astype(np.uint8) def _ck19_colored(ck19: np.ndarray) -> np.ndarray: # brown DAB rgb = np.zeros((*ck19.shape, 3), dtype=np.float32) rgb[..., 0] = 1 - 0.4 * ck19 rgb[..., 1] = 1 - 0.7 * ck19 rgb[..., 2] = 1 - 0.9 * ck19 return (np.clip(rgb, 0, 1) * 255).astype(np.uint8) def _sirius_colored(sirius: np.ndarray) -> np.ndarray: rgb = np.ones((*sirius.shape, 3), dtype=np.float32) rgb[..., 0] = 1 - 0.05 * sirius rgb[..., 1] = 1 - 0.7 * sirius rgb[..., 2] = 1 - 0.7 * sirius return (np.clip(rgb, 0, 1) * 255).astype(np.uint8) def _draw_overlay(rgb: np.ndarray, dets, drs) -> np.ndarray: """portal tract + DR structure 마커 그리기 (in-place 사본).""" out = rgb.copy() h, w = out.shape[:2] def circle(cx, cy, r, color, thick=2): for theta_deg in range(0, 360, 4): t = np.deg2rad(theta_deg) for rr in range(max(r - thick, 0), r + 1): x = int(cx + rr * np.cos(t)) y = int(cy + rr * np.sin(t)) if 0 <= x < w and 0 <= y < h: out[y, x] = color for d in dets: circle(d.cx, d.cy, max(d.pv_radius + 4, 12), (255, 200, 0), thick=2) # PT outline color_map = {"T1": (0, 200, 0), "T2": (0, 100, 220), "T3": (220, 30, 30)} for s in drs: circle(s.cx, s.cy, max(s.size + 1, 4), color_map.get(s.dr_type, (180, 180, 180)), thick=1) return out def _build_cohort_rows(slides: List[synth_liver.SynthSlide]) -> List[stats.CohortRow]: rows: List[stats.CohortRow] = [] for sl in slides: dets = portal_detect.detect_portal_tracts(sl.he, sl.ck19) drs = dr_classify.classify_dr_structures(sl.ck19, dets) if dets else [] sirius_pct = fibrosis_overlay.collagen_area_fraction(sl.sirius) bridges = fibrosis_overlay.detect_bridges(dets, sl.sirius, sl.ck19) if dets else [] rows.append( stats.per_slide_summary( slide_id=sl.slide_id, mouse_id=sl.mouse_id, group=sl.group, portal_tracts=dets, dr_structures=drs, sirius_area_pct=sirius_pct, bridges=bridges, ) ) return rows def run_streamlit() -> None: if st is None: print("Streamlit이 설치되어 있지 않습니다. requirements.txt 참조.") return st.set_page_config(page_title="CholangioDuctReact v0", layout="wide") st.title("CholangioDuctReact v0") st.caption("MASLD 동물실험 — Ductular Reaction 정량/분류 도구 (연구·참고용. 임상 진단 사용 금지)") with st.sidebar: st.header("코호트") st.write("합성 데모 코호트: 3그룹 × 3마우스") if st.button("데모 코호트 로드/재생성"): slides = synth_liver.make_demo_cohort(seed=42) st.session_state["slides"] = slides if "slides" not in st.session_state: st.session_state["slides"] = synth_liver.make_demo_cohort(seed=42) slides: List[synth_liver.SynthSlide] = st.session_state["slides"] slide_labels = [f"{s.group} | {s.mouse_id} | {s.slide_id}" for s in slides] idx = st.selectbox("슬라이드 선택", range(len(slides)), format_func=lambda i: slide_labels[i]) st.markdown("---") st.subheader("디스클레이머") st.warning("본 도구는 연구·참고용. 임상 진단 사용 금지.") sl = slides[idx] tabs = st.tabs(["Paired WSI", "Registration", "Portal & DR", "Fibrosis Overlay", "Cohort 통계"]) # Tab 1: Paired WSI with tabs[0]: st.subheader(f"{sl.group} / {sl.mouse_id} / {sl.slide_id}") c1, c2, c3 = st.columns(3) with c1: st.image(_slide_to_rgb_he(sl), caption="H&E (pseudo)", use_container_width=True) with c2: st.image(_ck19_colored(sl.ck19), caption="CK19 IHC (DAB)", use_container_width=True) with c3: st.image(_sirius_colored(sl.sirius), caption="Sirius Red", use_container_width=True) st.caption("SOX9 채널은 CK19와 거의 겹쳐 보이지만 별도 채널로 보존됨.") # Tab 2: Registration demo with tabs[1]: st.subheader("Rigid translation registration (합성 offset 시연)") offset_x = st.slider("CK19 offset X", -15, 15, 6) offset_y = st.slider("CK19 offset Y", -15, 15, -4) moved = registration.apply_translation(sl.ck19, offset_x, offset_y) aligned, (dx, dy) = registration.register_pair(sl.ck19, moved) c1, c2, c3 = st.columns(3) c1.image(_ck19_colored(sl.ck19), caption="Reference CK19", use_container_width=True) c2.image(_ck19_colored(moved), caption=f"Misaligned (dx={offset_x}, dy={offset_y})", use_container_width=True) c3.image(_ck19_colored(aligned), caption=f"Aligned (estimated dx={dx}, dy={dy})", use_container_width=True) # Tab 3: Portal tract & DR with tabs[2]: st.subheader("Portal tract auto-detection + DR Type 분류") dets = portal_detect.detect_portal_tracts(sl.he, sl.ck19) drs = dr_classify.classify_dr_structures(sl.ck19, dets) if dets else [] overlay = _draw_overlay(_slide_to_rgb_he(sl), dets, drs) c1, c2 = st.columns([2, 1]) with c1: st.image(overlay, caption=f"검출 portal tract: {len(dets)} / DR structure: {len(drs)}", use_container_width=True) with c2: t1 = sum(1 for d in drs if d.dr_type == "T1") t2 = sum(1 for d in drs if d.dr_type == "T2") t3 = sum(1 for d in drs if d.dr_type == "T3") st.metric("Type1 (reactive ductule)", t1) st.metric("Type2 (intermediate hepatocyte)", t2) st.metric("Type3 (mass-like metaplasia)", t3) st.bar_chart({"count": [t1, t2, t3]}) st.caption("색상: 노랑=portal tract, 초록=T1, 파랑=T2, 빨강=T3") # Tab 4: Fibrosis overlay with tabs[3]: st.subheader("Sirius Red Collagen + Bridging 분석") sir_pct = fibrosis_overlay.collagen_area_fraction(sl.sirius) st.metric("Collagen area (%)", f"{sir_pct * 100:.2f}") overlay_rgb = fibrosis_overlay.make_overlay_rgb(sl.he, sl.sirius) st.image((overlay_rgb * 255).astype(np.uint8), caption="H&E + Sirius Red overlay", use_container_width=True) dets = portal_detect.detect_portal_tracts(sl.he, sl.ck19) bridges = fibrosis_overlay.detect_bridges(dets, sl.sirius, sl.ck19) if dets else [] if bridges: import pandas as pd df = pd.DataFrame( [ { "PT_a": b.pt_a, "PT_b": b.pt_b, "거리(px)": round(b.distance, 1), "Sirius_score": round(b.sirius_along_score, 3), "CK19_score": round(b.ck19_along_score, 3), "타입": b.bridge_type, } for b in bridges ] ) st.dataframe(df, use_container_width=True) else: st.info("검출된 portal tract 쌍이 충분하지 않아 bridge 분석 결과 없음.") # Tab 5: Cohort 통계 with tabs[4]: st.subheader("Cohort 통계 + ANOVA") rows = _build_cohort_rows(slides) import pandas as pd df = pd.DataFrame([r.__dict__ for r in rows]) st.dataframe(df, use_container_width=True) groups: dict = {} for r in rows: groups.setdefault(r.group, []).append(r.dr_per_portal) anova = stats.one_way_anova(groups) st.write(f"DR per portal tract one-way ANOVA: F={anova.get('F', float('nan')):.3f}, p={anova.get('p', float('nan'))}") try: import plotly.express as px fig = px.violin(df, x="group", y="dr_per_portal", box=True, points="all", title="그룹별 DR / Portal Tract") st.plotly_chart(fig, use_container_width=True) fig2 = px.bar( df.melt(id_vars=["group", "mouse_id"], value_vars=["t1_count", "t2_count", "t3_count"]), x="mouse_id", y="value", color="variable", facet_col="group", title="DR Type 분포 (마우스별)", ) st.plotly_chart(fig2, use_container_width=True) except Exception as e: st.info(f"plotly 미설치 또는 오류: {e}") # Export out_csv = df.to_csv(index=False).encode("utf-8") st.download_button("CSV 다운로드", out_csv, file_name="cholangio_cohort.csv", mime="text/csv") st.download_button( "JSON 다운로드", json.dumps(stats.cohort_to_dict(rows), ensure_ascii=False, indent=2).encode("utf-8"), file_name="cholangio_cohort.json", mime="application/json", ) if __name__ == "__main__": run_streamlit()