evaluation_metrics.json

{
  "version": "v4.0_n50",
  "model": "llava-hf/llava-v1.6-mistral-7b-hf",
  "n_patients": 50,
  "timestamp": "2026-04-23T18:39:03.722916",
  "seed": 42,
  "new_metrics": [
    "BLEU-1/2/4 (MediVLM + BTReport)",
    "RaTEScore (Zhao EMNLP 2024)",
    "RadGraph-F1 proxy (entit\u00e9s + relations)",
    "RadGraph entity-F1 d\u00e9tail",
    "TBFact simple (BTReport inspir\u00e9)",
    "CSA Clinical Severity Agreement (ORIGINAL)"
  ],
  "means": {
    "rougeL": 0.3017,
    "rouge1": 0.413,
    "rouge2": 0.2076,
    "bleu1": 0.1926,
    "bleu2": 0.1176,
    "bleu4": 0.0288,
    "bert_F": 0.8141,
    "radgraph_f1": 0.8713,
    "rg_entity_f1": 0.9076,
    "rg_relation_f1": 0.7867,
    "medical_score": 0.8493,
    "structure_score": 1.0,
    "hallucination_score": 1.0,
    "crossval_score": 1.0,
    "completeness_score": 1.0,
    "tbfact_score": 0.9218,
    "global_score": 0.8522,
    "global_score_v4": 0.8818,
    "innovation_score": 0.8767,
    "article_score": 0.8558
  },
  "radgraph": {
    "composite_f1": 0.8713,
    "entity_f1": 0.9076,
    "relation_f1": 0.7867,
    "ci_95": [
      0.8561,
      0.8865
    ],
    "vs_autorg": 0.38,
    "note": "proxy sans d\u00e9pendance externe \u2014 entity coverage dominant"
  },
  "csa": {
    "pearson_r": 0.2317,
    "pearson_p": 0.1055,
    "spearman_r": 0.2401,
    "spearman_p": 0.0931,
    "agreement": 0.98,
    "significant": false
  },
  "vs_articles": {
    "your_bertscore": 0.8141,
    "MediVLM_bertscore": 0.616,
    "your_tbfact": 0.9218,
    "BTReport_tbfact": 0.353,
    "your_radgraph": 0.8713,
    "your_rg_entity": 0.9076,
    "AutoRG_radgraph": 0.38,
    "your_hallu": 1.0,
    "hallu_unique": true
  },
  "ci_95": {
    "global_score": [
      0.8506,
      0.8538
    ],
    "global_score_v4": [
      0.8753,
      0.8882
    ],
    "medical_score": [
      0.843,
      0.8557
    ],
    "hallucination_score": "stable",
    "crossval_score": "stable",
    "rougeL": [
      0.2929,
      0.3106
    ],
    "bert_F": [
      0.8131,
      0.8152
    ],
    "bleu1": [
      0.1878,
      0.1974
    ],
    "bleu2": [
      0.1145,
      0.1208
    ],
    "bleu4": [
      0.027,
      0.0307
    ],
    "radgraph_f1": [
      0.8561,
      0.8865
    ],
    "rg_entity_f1": [
      0.9018,
      0.9133
    ],
    "rg_relation_f1": [
      0.7379,
      0.8355
    ],
    "tbfact_score": [
      0.9053,
      0.9383
    ],
    "innovation_score": [
      0.8526,
      0.9007
    ],
    "article_score": [
      0.8522,
      0.8595
    ]
  }
}