"""Streamlit UI for CVOT-MACEAdjudicate-Kor. 참고용 · 연구용 도구입니다. 임상 의사결정용이 아닙니다 (Not for clinical decision). Run: streamlit run app.py """ from __future__ import annotations import io import os from collections import Counter from typing import Any, Dict, List try: import streamlit as st import pandas as pd import matplotlib.pyplot as plt except ImportError as e: print( f"[CVOT-MACEAdjudicate-Kor] Streamlit/pandas/matplotlib not installed: {e}\n" "Install dependencies first:\n" " pip install -r requirements.txt\n" "Then run: streamlit run app.py" ) raise SystemExit(0) from modules import ingest, classify, adjudicate, sensitivity DISCLAIMER = ( "참고용 · 연구용 도구입니다. 임상 의사결정용이 아닙니다 " "(Not for clinical decision)." ) @st.cache_data(show_spinner=False) def _load_packets(packets_path: str, troponin_path: str): return ingest.load_packets_csv(packets_path, troponin_path) @st.cache_data(show_spinner=False) def _load_charter(path: str): return ingest.load_charter(path) @st.cache_data(show_spinner=False) def _load_judgements(path: str): return adjudicate.load_judgements_csv(path) def main() -> None: st.set_page_config(page_title="CVOT-MACEAdjudicate-Kor", layout="wide") st.title("CVOT-MACEAdjudicate-Kor") st.warning(DISCLAIMER) st.caption( "T2DM CVOT MACE 의심 사건의 charter-driven 자동 보조 분류 · CEC 2-reader " "blinded 워크플로우 · sensitivity analysis · DSMB/CSR 리포트 생성." ) here = os.path.dirname(os.path.abspath(__file__)) with st.sidebar: st.header("입력") packets_csv = st.text_input("Event packets CSV", value=os.path.join(here, "data/synthetic_packets.csv")) troponin_csv = st.text_input("Troponin time-series CSV", value=os.path.join(here, "data/synthetic_troponin.csv")) adj_csv = st.text_input("Adjudications CSV", value=os.path.join(here, "data/synthetic_adjudications.csv")) charter_yaml = st.text_input("Charter YAML", value=os.path.join(here, "data/charter_example.yaml")) st.divider() include_type2 = st.checkbox("Type 2 MI 포함 (charter override)", value=False) try: raw_packets = _load_packets(packets_csv, troponin_csv) except FileNotFoundError: st.error(f"Packets CSV not found: {packets_csv}") return charter = _load_charter(charter_yaml) effective_include_type2 = include_type2 or bool(charter.get("include_type2_mi", False)) packets = ingest.deidentify_all(raw_packets) tab1, tab2, tab3, tab4 = st.tabs([ "1. Ingest / De-ID", "2. 자동 분류 보조", "3. CEC 2-reader 워크플로우 / Drift", "4. Sensitivity + 리포트", ]) with tab1: st.subheader(f"Packets: {len(packets)}") rows = [] for p in packets: rows.append({ "event_id": p.event_id, "patient_id (hashed)": p.patient_id, "site (masked)": p.site_id, "event_date (shifted)": p.event_date, "age": p.age, "sex": p.sex, "arm": p.treatment_arm, "suspected": p.suspected_category, "n_troponin_points": len(p.troponin_series), }) st.dataframe(pd.DataFrame(rows), use_container_width=True) st.json(charter, expanded=False) with tab2: st.subheader("자동 보조 분류 (UDM 2018 / AHA / ESC / CV death)") results = classify.classify_all(packets, include_type2_mi=effective_include_type2) rows = [r.to_dict() for r in results] df = pd.DataFrame(rows) st.dataframe(df, use_container_width=True) # Distribution col1, col2 = st.columns(2) with col1: st.markdown("**Domain 분포**") st.bar_chart(df["domain"].value_counts()) with col2: st.markdown("**Label 분포**") st.bar_chart(df["label"].value_counts()) st.session_state["classifications"] = rows with tab3: st.subheader("2-reader paired blinded adjudication") if not os.path.exists(adj_csv): st.warning(f"Adjudications CSV not found: {adj_csv}") return jdgs = _load_judgements(adj_csv) adj_summary = adjudicate.summarize(jdgs) c1, c2, c3, c4 = st.columns(4) c1.metric("Paired events", adj_summary["n_events_paired"]) c2.metric("Discordance rate", adj_summary["discordance_rate"]) c3.metric("Overall kappa", adj_summary["overall_kappa"]) c4.metric("Discordant", adj_summary["n_discordant"]) st.markdown("**Routing on discordance**") st.json(adj_summary.get("routing", {})) st.markdown("**Kappa by quarter (drift)**") kq = adj_summary.get("kappa_by_quarter", {}) if kq: kq_df = pd.DataFrame({"quarter": list(kq.keys()), "kappa": list(kq.values())}) st.line_chart(kq_df.set_index("quarter")) st.markdown("**Label distribution by quarter**") ldist = adj_summary.get("label_distribution_by_quarter", {}) if ldist: labels = sorted({lab for c in ldist.values() for lab in c}) df_ld = pd.DataFrame( [[ldist[q].get(lab, 0) for lab in labels] for q in sorted(ldist)], index=sorted(ldist), columns=labels, ) st.bar_chart(df_ld) st.markdown("**Workload / turnaround**") st.json({ "workload": adj_summary.get("workload_by_adjudicator", {}), "turnaround_h": adj_summary.get("turnaround_h_by_adjudicator", {}), }) st.session_state["adjudication_summary"] = adj_summary with tab4: st.subheader("Sensitivity analysis") cls_dicts = st.session_state.get("classifications") if cls_dicts is None: results = classify.classify_all(packets, include_type2_mi=effective_include_type2) cls_dicts = [r.to_dict() for r in results] packet_arms = {p.event_id: p.treatment_arm for p in raw_packets} grid = sensitivity.run_sensitivity_grid(cls_dicts, packet_arms) rows = [] for s in grid: rows.append({ "endpoint": s.name, "n_total": s.n_total, "components": ", ".join(f"{k}={v}" for k, v in s.by_component.items()), "by_arm": ", ".join(f"{k}={v}" for k, v in s.by_arm.items()), }) st.dataframe(pd.DataFrame(rows), use_container_width=True) st.markdown("**DSMB-ready interim summary**") adj_summary = st.session_state.get("adjudication_summary", {}) dsmb = sensitivity.dsmb_interim_summary(grid, adj_summary) st.json(dsmb, expanded=False) st.markdown("**CSR/manuscript supplementary report**") out_path = st.text_input("Report output path", value=os.path.join(here, "reports/run.docx")) if st.button("Export report"): path = sensitivity.export_csr_docx(out_path, grid, adj_summary, charter) st.success(f"Saved: {path}") if __name__ == "__main__": main()