{
  "protocol": "leave_one_design_out",
  "threshold": 0.5,
  "macro_iou": 0.08888888888888889,
  "positive_nodes": 49,
  "detected_positive_nodes": 2,
  "confusion": {
    "tp": 2,
    "fp": 5,
    "fn": 47,
    "tn": 1173
  },
  "note": "Small improvement against weak labels, concentrated in two AC28 scans; 47 of 49 weak positive nodes are still missed. No independently reviewed accuracy claim.",
  "datasets": [
    {
      "dataset_id": "AC28_DT_RR_UPR_STD_20260827_16__NG_008adde420f0",
      "bce": 0.12767126605224433,
      "brier": 0.02411260015876402,
      "iou": 0.3333333333333333,
      "dice": 0.5,
      "precision": 1.0,
      "recall": 0.3333333333333333,
      "nodes": 213,
      "positive_nodes": 3,
      "negative_nodes": 210,
      "confusion": {
        "tp": 1,
        "fp": 0,
        "fn": 2,
        "tn": 210
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.3444568038561364
    },
    {
      "dataset_id": "AC28_DT_RR_UPR_STD_20260828_09_21__NG_f4c68f38e467",
      "bce": 0.12856437907894078,
      "brier": 0.025457593808805506,
      "iou": 0.2,
      "dice": 0.3333333333333333,
      "precision": 0.3333333333333333,
      "recall": 0.3333333333333333,
      "nodes": 216,
      "positive_nodes": 3,
      "negative_nodes": 213,
      "confusion": {
        "tp": 1,
        "fp": 2,
        "fn": 2,
        "tn": 211
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.34395604395604396
    },
    {
      "dataset_id": "AC28_DT_RR_UPR_STD_2_20260827_18_01__NG_8df93616f7ea",
      "bce": 0.10447718074326491,
      "brier": 0.018120446993521295,
      "iou": 0.0,
      "dice": 0.0,
      "precision": null,
      "recall": 0.0,
      "nodes": 223,
      "positive_nodes": 3,
      "negative_nodes": 220,
      "confusion": {
        "tp": 0,
        "fp": 0,
        "fn": 3,
        "tn": 220
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.34404301075268817
    },
    {
      "dataset_id": "ME1a_DT_FRT_UPR_LHD_20260608_14_06__NG_ffc4fce5d996",
      "bce": 0.30804546936900035,
      "brier": 0.07507264355442292,
      "iou": 0.0,
      "dice": 0.0,
      "precision": null,
      "recall": 0.0,
      "nodes": 202,
      "positive_nodes": 16,
      "negative_nodes": 186,
      "confusion": {
        "tp": 0,
        "fp": 0,
        "fn": 16,
        "tn": 186
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.10358785241906357
    },
    {
      "dataset_id": "MX5a_PE_DT_RR_AREST_1_20260626_13_08__NG_fe0e1b9b8b53",
      "bce": 0.2875064557060664,
      "brier": 0.07021996262409116,
      "iou": 0.0,
      "dice": 0.0,
      "precision": null,
      "recall": 0.0,
      "nodes": 178,
      "positive_nodes": 12,
      "negative_nodes": 166,
      "confusion": {
        "tp": 0,
        "fp": 0,
        "fn": 12,
        "tn": 166
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.07656234737611456
    },
    {
      "dataset_id": "MX5a_PE_DT_RR_AREST_2_20260626_13_08__NG_41ec5d7a218e",
      "bce": 0.2695476792692063,
      "brier": 0.06872814119743649,
      "iou": 0.0,
      "dice": 0.0,
      "precision": 0.0,
      "recall": 0.0,
      "nodes": 195,
      "positive_nodes": 12,
      "negative_nodes": 183,
      "confusion": {
        "tp": 0,
        "fp": 3,
        "fn": 12,
        "tn": 180
      },
      "interpretation": "held-out-design weak-label agreement",
      "independently_reviewed": false,
      "average_precision": 0.1089574568589371
    }
  ]
}
