{
  "selection": {
    "selection_data": "BUS-BRA validation cases only",
    "search_seed": 17,
    "selected": {
      "classification": "classification_224",
      "segmentation": "segmentation_resnet34_384"
    },
    "candidates": [
      {
        "model": "classification_224",
        "task": "classification",
        "architecture": "resnet18",
        "image_size": 224,
        "batch_size": 16,
        "seed": 17,
        "best_epoch": 17,
        "validation_metric": "ap_case",
        "validation_score": 0.9018921198061237,
        "checkpoint": "artifacts/classification_224/seed_17/best.pt",
        "checkpoint_sha256": "b99cfddaa9c31a38183b449de55af3c49d50e7f83dd94c39dce77fe6598a43d2"
      },
      {
        "model": "classification_384",
        "task": "classification",
        "architecture": "resnet18",
        "image_size": 384,
        "batch_size": 8,
        "seed": 17,
        "best_epoch": 25,
        "validation_metric": "ap_case",
        "validation_score": 0.8948301941259348,
        "checkpoint": "artifacts/classification_384/seed_17/best.pt",
        "checkpoint_sha256": "0a16d78f5854b5e90d55e6343822cf14964bcd092fdd2bde524e75f6284e8e8b"
      },
      {
        "model": "segmentation_224",
        "task": "segmentation",
        "architecture": "resnet18",
        "image_size": 224,
        "batch_size": 16,
        "seed": 17,
        "best_epoch": 28,
        "validation_metric": "dice_case",
        "validation_score": 0.8913458150547526,
        "checkpoint": "artifacts/segmentation_224/seed_17/best.pt",
        "checkpoint_sha256": "aa76c82fbe17fdf14e3d55b6d56a34525dc749294a0fe7f5eaeaa772eedfc09b"
      },
      {
        "model": "segmentation_384",
        "task": "segmentation",
        "architecture": "resnet18",
        "image_size": 384,
        "batch_size": 8,
        "seed": 17,
        "best_epoch": 18,
        "validation_metric": "dice_case",
        "validation_score": 0.8978688167147346,
        "checkpoint": "artifacts/segmentation_384/seed_17/best.pt",
        "checkpoint_sha256": "d28b012e47f88ea6319b9677c69464b0c03e9f032d5859489b9a7a1d2b4c3260"
      },
      {
        "model": "segmentation_resnet34_384",
        "task": "segmentation",
        "architecture": "resnet34",
        "image_size": 384,
        "batch_size": 8,
        "seed": 17,
        "best_epoch": 14,
        "validation_metric": "dice_case",
        "validation_score": 0.9028113453097003,
        "checkpoint": "artifacts/segmentation_resnet34_384/seed_17/best.pt",
        "checkpoint_sha256": "164cbb6e52e1be933aee14b6a28e3a309584ecd62ec98f9610c60d01d2436650"
      }
    ],
    "tie_rule": "lower resolution, smaller encoder, lexical model name",
    "splits_sha256": "31dde07a218615b7e4f3be47f26bb6069344f526c567c13c3e76bf71567f323c"
  },
  "aggregates": [
    {
      "model": "classification_224",
      "label": "ResNet18 classifier · 224 px",
      "partition": "test",
      "n_runs": 3,
      "n_cases": 160,
      "classification_auroc": {
        "mean": 0.9031339031339032,
        "sd": 0.015954256969089874
      },
      "classification_average_precision": {
        "mean": 0.8646848856329944,
        "sd": 0.017098646368812156
      },
      "classification_sensitivity": {
        "mean": 0.8012820512820512,
        "sd": 0.09679274917013782
      },
      "classification_specificity": {
        "mean": 0.8518518518518517,
        "sd": 0.05555555555555558
      },
      "classification_precision": {
        "mean": 0.7292305694856939,
        "sd": 0.053730295599373167
      },
      "classification_f1": {
        "mean": 0.7588759431962101,
        "sd": 0.015759493323652878
      },
      "classification_balanced_accuracy": {
        "mean": 0.8265669515669516,
        "sd": 0.021044567952555705
      },
      "classification_mcc": {
        "mean": 0.6416236819463607,
        "sd": 0.018493643568217717
      },
      "classification_brier": {
        "mean": 0.12828312688626356,
        "sd": 0.03233840137830768
      },
      "calibration_secondary_brier": {
        "mean": 0.12081225832677443,
        "sd": 0.029519163439675297
      },
      "segmentation_dice": null,
      "segmentation_iou": null,
      "segmentation_bbox_iou": null,
      "segmentation_dice_median": null,
      "segmentation_dice_p10": null,
      "segmentation_bbox_success_image_macro": null
    },
    {
      "model": "classification_224",
      "label": "ResNet18 classifier · 224 px",
      "partition": "external",
      "n_runs": 3,
      "n_cases": 252,
      "classification_auroc": {
        "mean": 0.8467179079423977,
        "sd": 0.012537089858395155
      },
      "classification_average_precision": {
        "mean": 0.7790053194670589,
        "sd": 0.023050484208147936
      },
      "classification_sensitivity": {
        "mean": 0.8333333333333334,
        "sd": 0.041239304942116105
      },
      "classification_specificity": {
        "mean": 0.6904761904761906,
        "sd": 0.026243194054073903
      },
      "classification_precision": {
        "mean": 0.6316216002530045,
        "sd": 0.02139334685494215
      },
      "classification_f1": {
        "mean": 0.7183350811669396,
        "sd": 0.024836823043792124
      },
      "classification_balanced_accuracy": {
        "mean": 0.7619047619047619,
        "sd": 0.02280444535249514
      },
      "classification_mcc": {
        "mean": 0.5112895313689897,
        "sd": 0.044409886371261045
      },
      "classification_brier": {
        "mean": 0.22018465089300554,
        "sd": 0.07611746809584964
      },
      "calibration_secondary_brier": {
        "mean": 0.19910058319065085,
        "sd": 0.06792539411515534
      },
      "segmentation_dice": null,
      "segmentation_iou": null,
      "segmentation_bbox_iou": null,
      "segmentation_dice_median": null,
      "segmentation_dice_p10": null,
      "segmentation_bbox_success_image_macro": null
    },
    {
      "model": "control",
      "label": "Prevalence / mean mask",
      "partition": "test",
      "n_runs": 1,
      "n_cases": 160,
      "classification_auroc": {
        "mean": 0.5,
        "sd": null
      },
      "classification_average_precision": {
        "mean": 0.325,
        "sd": null
      },
      "classification_sensitivity": {
        "mean": 1.0,
        "sd": null
      },
      "classification_specificity": {
        "mean": 0.0,
        "sd": null
      },
      "classification_precision": {
        "mean": 0.325,
        "sd": null
      },
      "classification_f1": {
        "mean": 0.49056603773584906,
        "sd": null
      },
      "classification_balanced_accuracy": {
        "mean": 0.5,
        "sd": null
      },
      "classification_mcc": {
        "mean": 0.0,
        "sd": null
      },
      "classification_brier": {
        "mean": 0.2193891634871083,
        "sd": null
      },
      "calibration_secondary_brier": {
        "mean": 0.21941406292406002,
        "sd": null
      },
      "segmentation_dice": {
        "mean": 0.25997629326964544,
        "sd": null
      },
      "segmentation_iou": {
        "mean": 0.16387721162657978,
        "sd": null
      },
      "segmentation_bbox_iou": {
        "mean": 0.1874098910098664,
        "sd": null
      },
      "segmentation_dice_median": {
        "mean": 0.26511068524274267,
        "sd": null
      },
      "segmentation_dice_p10": {
        "mean": 0.00285532994923858,
        "sd": null
      },
      "segmentation_bbox_success_image_macro": {
        "mean": 0.025,
        "sd": null
      }
    },
    {
      "model": "control",
      "label": "Prevalence / mean mask",
      "partition": "external",
      "n_runs": 1,
      "n_cases": 252,
      "classification_auroc": {
        "mean": 0.5,
        "sd": null
      },
      "classification_average_precision": {
        "mean": 0.3888888888888889,
        "sd": null
      },
      "classification_sensitivity": {
        "mean": 1.0,
        "sd": null
      },
      "classification_specificity": {
        "mean": 0.0,
        "sd": null
      },
      "classification_precision": {
        "mean": 0.3888888888888889,
        "sd": null
      },
      "classification_f1": {
        "mean": 0.56,
        "sd": null
      },
      "classification_balanced_accuracy": {
        "mean": 0.5,
        "sd": null
      },
      "classification_mcc": {
        "mean": 0.0,
        "sd": null
      },
      "classification_brier": {
        "mean": 0.24223115870813577,
        "sd": null
      },
      "calibration_secondary_brier": {
        "mean": 0.24257378948110545,
        "sd": null
      },
      "segmentation_dice": {
        "mean": 0.25577647200597586,
        "sd": null
      },
      "segmentation_iou": {
        "mean": 0.16370166397862118,
        "sd": null
      },
      "segmentation_bbox_iou": {
        "mean": 0.19498914650624263,
        "sd": null
      },
      "segmentation_dice_median": {
        "mean": 0.22788059732207727,
        "sd": null
      },
      "segmentation_dice_p10": {
        "mean": 0.0,
        "sd": null
      },
      "segmentation_bbox_success_image_macro": {
        "mean": 0.047619047619047616,
        "sd": null
      }
    },
    {
      "model": "frozen_resnet18",
      "label": "Frozen ResNet18 + logistic regression",
      "partition": "test",
      "n_runs": 1,
      "n_cases": 160,
      "classification_auroc": {
        "mean": 0.7767094017094017,
        "sd": null
      },
      "classification_average_precision": {
        "mean": 0.6934482181797306,
        "sd": null
      },
      "classification_sensitivity": {
        "mean": 0.5769230769230769,
        "sd": null
      },
      "classification_specificity": {
        "mean": 0.9074074074074074,
        "sd": null
      },
      "classification_precision": {
        "mean": 0.75,
        "sd": null
      },
      "classification_f1": {
        "mean": 0.6521739130434783,
        "sd": null
      },
      "classification_balanced_accuracy": {
        "mean": 0.7421652421652422,
        "sd": null
      },
      "classification_mcc": {
        "mean": 0.5238835186571609,
        "sd": null
      },
      "classification_brier": {
        "mean": 0.1656095802078144,
        "sd": null
      },
      "calibration_secondary_brier": {
        "mean": 0.16499273835941164,
        "sd": null
      },
      "segmentation_dice": null,
      "segmentation_iou": null,
      "segmentation_bbox_iou": null,
      "segmentation_dice_median": null,
      "segmentation_dice_p10": null,
      "segmentation_bbox_success_image_macro": null
    },
    {
      "model": "frozen_resnet18",
      "label": "Frozen ResNet18 + logistic regression",
      "partition": "external",
      "n_runs": 1,
      "n_cases": 252,
      "classification_auroc": {
        "mean": 0.679035250463822,
        "sd": null
      },
      "classification_average_precision": {
        "mean": 0.59977926900708,
        "sd": null
      },
      "classification_sensitivity": {
        "mean": 0.25510204081632654,
        "sd": null
      },
      "classification_specificity": {
        "mean": 0.9090909090909091,
        "sd": null
      },
      "classification_precision": {
        "mean": 0.6410256410256411,
        "sd": null
      },
      "classification_f1": {
        "mean": 0.36496350364963503,
        "sd": null
      },
      "classification_balanced_accuracy": {
        "mean": 0.5820964749536178,
        "sd": null
      },
      "classification_mcc": {
        "mean": 0.22131242582435412,
        "sd": null
      },
      "classification_brier": {
        "mean": 0.2661932485813797,
        "sd": null
      },
      "calibration_secondary_brier": {
        "mean": 0.26908433549443284,
        "sd": null
      },
      "segmentation_dice": null,
      "segmentation_iou": null,
      "segmentation_bbox_iou": null,
      "segmentation_dice_median": null,
      "segmentation_dice_p10": null,
      "segmentation_bbox_success_image_macro": null
    },
    {
      "model": "segmentation_resnet34_384",
      "label": "U-Net / ResNet34 · 384 px",
      "partition": "test",
      "n_runs": 3,
      "n_cases": 160,
      "classification_auroc": null,
      "classification_average_precision": null,
      "classification_sensitivity": null,
      "classification_specificity": null,
      "classification_precision": null,
      "classification_f1": null,
      "classification_balanced_accuracy": null,
      "classification_mcc": null,
      "classification_brier": null,
      "calibration_secondary_brier": null,
      "segmentation_dice": {
        "mean": 0.8998659336843033,
        "sd": 0.0019134970585975373
      },
      "segmentation_iou": {
        "mean": 0.8297098021357949,
        "sd": 0.003054365191273963
      },
      "segmentation_bbox_iou": {
        "mean": 0.8390835874562258,
        "sd": 0.0009059325612342005
      },
      "segmentation_dice_median": {
        "mean": 0.9252492883561061,
        "sd": 0.001691408488151876
      },
      "segmentation_dice_p10": {
        "mean": 0.810057050419348,
        "sd": 0.01665625479521064
      },
      "segmentation_bbox_success_image_macro": {
        "mean": 0.9635416666666666,
        "sd": 0.0078644108700737
      }
    },
    {
      "model": "segmentation_resnet34_384",
      "label": "U-Net / ResNet34 · 384 px",
      "partition": "external",
      "n_runs": 3,
      "n_cases": 252,
      "classification_auroc": null,
      "classification_average_precision": null,
      "classification_sensitivity": null,
      "classification_specificity": null,
      "classification_precision": null,
      "classification_f1": null,
      "classification_balanced_accuracy": null,
      "classification_mcc": null,
      "classification_brier": null,
      "calibration_secondary_brier": null,
      "segmentation_dice": {
        "mean": 0.7345256801834174,
        "sd": 0.011584988216767374
      },
      "segmentation_iou": {
        "mean": 0.6320046982069911,
        "sd": 0.011053240206468204
      },
      "segmentation_bbox_iou": {
        "mean": 0.6500006414774819,
        "sd": 0.019200693368674828
      },
      "segmentation_dice_median": {
        "mean": 0.8398744604458849,
        "sd": 0.005633537609942984
      },
      "segmentation_dice_p10": {
        "mean": 0.3222156357810033,
        "sd": 0.01156027896874636
      },
      "segmentation_bbox_success_image_macro": {
        "mean": 0.7883597883597884,
        "sd": 0.03687132239356178
      }
    }
  ],
  "paired_comparisons": [
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "test",
      "metric": "auroc",
      "difference": 0.12642450142450146,
      "ci95": [
        0.05371260683760681,
        0.20038580246913582
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "test",
      "metric": "average_precision",
      "difference": 0.1712366674532637,
      "ci95": [
        0.0755717558162096,
        0.2639616648166056
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "test",
      "metric": "brier",
      "difference": -0.03732645332155085,
      "ci95": [
        -0.06861248834746729,
        -0.0031194810144547667
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "external",
      "metric": "auroc",
      "difference": 0.1676826574785757,
      "ci95": [
        0.10689382896015555,
        0.23260170951497466
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "external",
      "metric": "average_precision",
      "difference": 0.17922605045997886,
      "ci95": [
        0.10894265528670605,
        0.25273111890105293
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "frozen_resnet18",
      "partition": "external",
      "metric": "brier",
      "difference": -0.046008597688374175,
      "ci95": [
        -0.08369079192830188,
        -0.011144736447280942
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "control",
      "partition": "test",
      "metric": "auroc",
      "difference": 0.4031339031339032,
      "ci95": [
        0.34520269468186154,
        0.4475338319088319
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "control",
      "partition": "test",
      "metric": "average_precision",
      "difference": 0.5396848856329943,
      "ci95": [
        0.471260164039954,
        0.5968767602319391
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "control",
      "partition": "external",
      "metric": "auroc",
      "difference": 0.34671790794239765,
      "ci95": [
        0.3015052124746002,
        0.38788927908825865
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "classification_224",
      "second": "control",
      "partition": "external",
      "metric": "average_precision",
      "difference": 0.39011643057816997,
      "ci95": [
        0.3285810373262467,
        0.4537480441718623
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "segmentation_resnet34_384",
      "second": "control",
      "partition": "test",
      "metric": "dice",
      "difference": 0.6398896404146581,
      "ci95": [
        0.6086078159055708,
        0.6694955903774837
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "segmentation_resnet34_384",
      "second": "control",
      "partition": "test",
      "metric": "iou",
      "difference": 0.6658325905092152,
      "ci95": [
        0.6385072297866801,
        0.690733121686799
      ],
      "n_cases": 160,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "segmentation_resnet34_384",
      "second": "control",
      "partition": "external",
      "metric": "dice",
      "difference": 0.4787492081774416,
      "ci95": [
        0.44168584743171113,
        0.5140487096529407
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    },
    {
      "first": "segmentation_resnet34_384",
      "second": "control",
      "partition": "external",
      "metric": "iou",
      "difference": 0.46830303422836994,
      "ci95": [
        0.43384695647436766,
        0.5013924341197129
      ],
      "n_cases": 252,
      "n_seeds": 3,
      "replicates": 2000,
      "seed": 2026,
      "method": "paired case percentile bootstrap of mean fixed-seed metric differences; class-stratified for classification"
    }
  ],
  "runs": {
    "classification_224": [
      17,
      29,
      43
    ],
    "control": [
      17
    ],
    "frozen_resnet18": [
      17
    ],
    "segmentation_resnet34_384": [
      17,
      29,
      43
    ]
  }
}