feat(ceus,segment): 任务书 02 第二阶段——分割和后半段算法

- ceus 迁入 polygons、dark_region、reference_region、tic_fitting、tic、anchor_frames、
  perfusion_metrics、trajectory、tic_curves、tic_plot、overlay，旧测试照搬（改 import 与配置读取）。
- 对外两段接口：prepare_segmentation（第 1–2 步）与 quantify_case（第 4–9 步）。
- 第 3 条：删掉 D1 取医生标注时刻的模式，D1 只从拟合曲线算。
- 第 4 条：F1/F2 取视频第 20−X、60−X 秒（X 为视频第 0 秒时已推注的秒数），落在视频外记未定位。
- 绿框融合权重、紫圈颜色阈值照搬；关键帧容差窗口的出处写进 ceus.yaml 注释。
- segment：同一进程调用 MedSAM2，设置写进 ceus.yaml 的 segment 段，全部必填；gpu 测试真跑通过。
- 依赖加 scipy、matplotlib、medsam2（可编辑安装，不编译 CUDA 扩展）；uv 只解析 linux x86_64。
- 导入契约：只有 segment 引用 torch、sam2；segment 不引用其他 cadx 模块。

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tests/test_ceus_perfusion_metrics.py

@@ -0,0 +1,161 @@
+"""Lognormal 拟合派生灌注指标的纯算法单元测试（照搬旧项目 test_perfusion_metrics.py）。"""
+
+from dataclasses import replace
+
+import numpy as np
+import pytest
+
+from cadx.ceus import (
+    CeusConfig,
+    LognormalFitResult,
+    LognormalParameters,
+    compute_lognormal_perfusion_metrics,
+    evaluate_lognormal,
+    fit_lognormal_curve,
+)
+from conftest import load_ceus_section
+
+# 配置不设默认值：旧测试靠默认值构造的配置，改为读 config/ceus.yaml 的取值。
+_CONFIG = CeusConfig.model_validate(load_ceus_section())
+
+
+@pytest.fixture(scope="module")
+def complete_fit() -> LognormalFitResult:
+    times = np.linspace(0.0, 60.0, 601)
+    parameters = LognormalParameters(400.0, 2.2, 0.5, 8.0, 30.0)
+    fit = fit_lognormal_curve(
+        times,
+        evaluate_lognormal(times, parameters),
+        baseline=parameters.baseline,
+        baseline_noise=0.2,
+        baseline_end_index=20,
+        duplicate_mask=np.zeros(times.size, dtype=bool),
+        sample_mask=None,
+        cfg=_CONFIG.tic_analysis.fitting,
+        anchor_cfg=_CONFIG.tic_analysis.anchor_frames,
+    )
+    assert fit.success is True
+    assert fit.d4_time_s is not None
+    return fit
+
+
+def test_complete_bolus_metrics_match_curve_integrals(
+    complete_fit: LognormalFitResult,
+) -> None:
+    metrics = compute_lognormal_perfusion_metrics(complete_fit)
+
+    assert metrics is not None
+    assert complete_fit.parameters is not None
+    assert complete_fit.d2_time_s is not None
+    assert complete_fit.d3_time_s is not None
+    assert complete_fit.d4_time_s is not None
+    assert metrics.ti_s is not None
+    assert metrics.to_s is not None
+    assert metrics.mtt_s is not None
+    assert metrics.wi_auc_au_s is not None
+    assert metrics.wo_auc_au_s is not None
+    assert metrics.wiwo_auc_au_s is not None
+    assert metrics.rt_s is not None
+    assert metrics.ft_s is not None
+    assert metrics.mttl_s is not None
+    assert metrics.wipi_au is not None
+    assert metrics.wor_au_per_s is not None
+
+    assert (
+        complete_fit.fit_start_s
+        <= metrics.ti_s
+        < complete_fit.d2_time_s
+        < metrics.ttp_s
+        < complete_fit.d4_time_s
+        < metrics.to_s
+        <= complete_fit.fit_end_s
+    )
+    assert metrics.pe_au == pytest.approx(
+        float(
+            evaluate_lognormal(np.asarray([metrics.ttp_s]), complete_fit.parameters)[0]
+            - complete_fit.parameters.baseline
+        )
+    )
+    assert metrics.ttp_s == pytest.approx(complete_fit.d3_time_s)
+    assert metrics.wir_au_per_s > 0.0
+    assert metrics.wor_au_per_s < 0.0
+    assert metrics.rt_s == pytest.approx(metrics.ttp_s - metrics.ti_s)
+    assert metrics.ft_s == pytest.approx(metrics.to_s - metrics.ttp_s)
+    assert metrics.wiwo_auc_au_s == pytest.approx(metrics.wi_auc_au_s + metrics.wo_auc_au_s)
+    assert metrics.wipi_au == pytest.approx(metrics.wi_auc_au_s / metrics.rt_s)
+    assert metrics.mttl_s == pytest.approx(metrics.mtt_s - metrics.ti_s)
+    assert metrics.ti_s < metrics.mtt_s < metrics.to_s
+
+    grid = np.linspace(metrics.ti_s, metrics.to_s, 100_001)
+    pulse = evaluate_lognormal(grid, complete_fit.parameters) - (complete_fit.parameters.baseline)
+    split = int(np.searchsorted(grid, metrics.ttp_s))
+    wi_grid = np.concatenate((grid[:split], np.asarray([metrics.ttp_s])))
+    wo_grid = np.concatenate((np.asarray([metrics.ttp_s]), grid[split:]))
+    wi_pulse = evaluate_lognormal(wi_grid, complete_fit.parameters) - (
+        complete_fit.parameters.baseline
+    )
+    wo_pulse = evaluate_lognormal(wo_grid, complete_fit.parameters) - (
+        complete_fit.parameters.baseline
+    )
+    assert metrics.wi_auc_au_s == pytest.approx(np.trapezoid(wi_pulse, wi_grid), rel=1e-7)
+    assert metrics.wo_auc_au_s == pytest.approx(np.trapezoid(wo_pulse, wo_grid), rel=1e-7)
+    assert metrics.mtt_s == pytest.approx(
+        np.trapezoid(grid * pulse, grid) / np.trapezoid(pulse, grid), rel=1e-7
+    )
+
+
+def test_missing_washout_keeps_only_wash_in_metrics(
+    complete_fit: LognormalFitResult,
+) -> None:
+    metrics = compute_lognormal_perfusion_metrics(replace(complete_fit, d4_time_s=None))
+
+    assert metrics is not None
+    assert metrics.pe_au > 0.0
+    assert metrics.ttp_s > 0.0
+    assert metrics.wir_au_per_s > 0.0
+    assert metrics.wi_auc_au_s is not None
+    assert metrics.rt_s is not None
+    assert metrics.wipi_au is not None
+    assert metrics.wor_au_per_s is None
+    assert metrics.wo_auc_au_s is None
+    assert metrics.wiwo_auc_au_s is None
+    assert metrics.ft_s is None
+    assert metrics.mttl_s is None
+    assert metrics.to_s is None
+    assert metrics.mtt_s is None
+
+
+def test_failed_fit_has_no_metrics(complete_fit: LognormalFitResult) -> None:
+    assert compute_lognormal_perfusion_metrics(replace(complete_fit, success=False)) is None
+
+
+def test_tangent_intersections_outside_observation_gate_dependents(
+    complete_fit: LognormalFitResult,
+) -> None:
+    complete = compute_lognormal_perfusion_metrics(complete_fit)
+    assert complete is not None
+    assert complete.ti_s is not None
+    assert complete.to_s is not None
+
+    missing_wash_in = compute_lognormal_perfusion_metrics(
+        replace(complete_fit, fit_start_s=complete.ti_s + 0.01)
+    )
+    assert missing_wash_in is not None
+    assert missing_wash_in.ti_s is None
+    assert missing_wash_in.rt_s is None
+    assert missing_wash_in.wi_auc_au_s is None
+    assert missing_wash_in.wipi_au is None
+    assert missing_wash_in.wo_auc_au_s is not None
+    assert missing_wash_in.wiwo_auc_au_s is None
+    assert missing_wash_in.mttl_s is None
+
+    missing_wash_out = compute_lognormal_perfusion_metrics(
+        replace(complete_fit, fit_end_s=complete.to_s - 0.01)
+    )
+    assert missing_wash_out is not None
+    assert missing_wash_out.wor_au_per_s is not None
+    assert missing_wash_out.to_s is None
+    assert missing_wash_out.ft_s is None
+    assert missing_wash_out.wo_auc_au_s is None
+    assert missing_wash_out.wiwo_auc_au_s is None
+    assert missing_wash_out.mttl_s is None