Upload evaluation_metrics.json — RCMTUNetV4-VLM v4.0
Browse files- evaluation_metrics.json +127 -0
evaluation_metrics.json
ADDED
|
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "v4.0_n50",
|
| 3 |
+
"model": "llava-hf/llava-v1.6-mistral-7b-hf",
|
| 4 |
+
"n_patients": 50,
|
| 5 |
+
"timestamp": "2026-04-23T18:39:03.722916",
|
| 6 |
+
"seed": 42,
|
| 7 |
+
"new_metrics": [
|
| 8 |
+
"BLEU-1/2/4 (MediVLM + BTReport)",
|
| 9 |
+
"RaTEScore (Zhao EMNLP 2024)",
|
| 10 |
+
"RadGraph-F1 proxy (entit\u00e9s + relations)",
|
| 11 |
+
"RadGraph entity-F1 d\u00e9tail",
|
| 12 |
+
"TBFact simple (BTReport inspir\u00e9)",
|
| 13 |
+
"CSA Clinical Severity Agreement (ORIGINAL)"
|
| 14 |
+
],
|
| 15 |
+
"means": {
|
| 16 |
+
"rougeL": 0.3017,
|
| 17 |
+
"rouge1": 0.413,
|
| 18 |
+
"rouge2": 0.2076,
|
| 19 |
+
"bleu1": 0.1926,
|
| 20 |
+
"bleu2": 0.1176,
|
| 21 |
+
"bleu4": 0.0288,
|
| 22 |
+
"bert_F": 0.8141,
|
| 23 |
+
"radgraph_f1": 0.8713,
|
| 24 |
+
"rg_entity_f1": 0.9076,
|
| 25 |
+
"rg_relation_f1": 0.7867,
|
| 26 |
+
"medical_score": 0.8493,
|
| 27 |
+
"structure_score": 1.0,
|
| 28 |
+
"hallucination_score": 1.0,
|
| 29 |
+
"crossval_score": 1.0,
|
| 30 |
+
"completeness_score": 1.0,
|
| 31 |
+
"tbfact_score": 0.9218,
|
| 32 |
+
"global_score": 0.8522,
|
| 33 |
+
"global_score_v4": 0.8818,
|
| 34 |
+
"innovation_score": 0.8767,
|
| 35 |
+
"article_score": 0.8558
|
| 36 |
+
},
|
| 37 |
+
"radgraph": {
|
| 38 |
+
"composite_f1": 0.8713,
|
| 39 |
+
"entity_f1": 0.9076,
|
| 40 |
+
"relation_f1": 0.7867,
|
| 41 |
+
"ci_95": [
|
| 42 |
+
0.8561,
|
| 43 |
+
0.8865
|
| 44 |
+
],
|
| 45 |
+
"vs_autorg": 0.38,
|
| 46 |
+
"note": "proxy sans d\u00e9pendance externe \u2014 entity coverage dominant"
|
| 47 |
+
},
|
| 48 |
+
"csa": {
|
| 49 |
+
"pearson_r": 0.2317,
|
| 50 |
+
"pearson_p": 0.1055,
|
| 51 |
+
"spearman_r": 0.2401,
|
| 52 |
+
"spearman_p": 0.0931,
|
| 53 |
+
"agreement": 0.98,
|
| 54 |
+
"significant": false
|
| 55 |
+
},
|
| 56 |
+
"vs_articles": {
|
| 57 |
+
"your_bertscore": 0.8141,
|
| 58 |
+
"MediVLM_bertscore": 0.616,
|
| 59 |
+
"your_tbfact": 0.9218,
|
| 60 |
+
"BTReport_tbfact": 0.353,
|
| 61 |
+
"your_radgraph": 0.8713,
|
| 62 |
+
"your_rg_entity": 0.9076,
|
| 63 |
+
"AutoRG_radgraph": 0.38,
|
| 64 |
+
"your_hallu": 1.0,
|
| 65 |
+
"hallu_unique": true
|
| 66 |
+
},
|
| 67 |
+
"ci_95": {
|
| 68 |
+
"global_score": [
|
| 69 |
+
0.8506,
|
| 70 |
+
0.8538
|
| 71 |
+
],
|
| 72 |
+
"global_score_v4": [
|
| 73 |
+
0.8753,
|
| 74 |
+
0.8882
|
| 75 |
+
],
|
| 76 |
+
"medical_score": [
|
| 77 |
+
0.843,
|
| 78 |
+
0.8557
|
| 79 |
+
],
|
| 80 |
+
"hallucination_score": "stable",
|
| 81 |
+
"crossval_score": "stable",
|
| 82 |
+
"rougeL": [
|
| 83 |
+
0.2929,
|
| 84 |
+
0.3106
|
| 85 |
+
],
|
| 86 |
+
"bert_F": [
|
| 87 |
+
0.8131,
|
| 88 |
+
0.8152
|
| 89 |
+
],
|
| 90 |
+
"bleu1": [
|
| 91 |
+
0.1878,
|
| 92 |
+
0.1974
|
| 93 |
+
],
|
| 94 |
+
"bleu2": [
|
| 95 |
+
0.1145,
|
| 96 |
+
0.1208
|
| 97 |
+
],
|
| 98 |
+
"bleu4": [
|
| 99 |
+
0.027,
|
| 100 |
+
0.0307
|
| 101 |
+
],
|
| 102 |
+
"radgraph_f1": [
|
| 103 |
+
0.8561,
|
| 104 |
+
0.8865
|
| 105 |
+
],
|
| 106 |
+
"rg_entity_f1": [
|
| 107 |
+
0.9018,
|
| 108 |
+
0.9133
|
| 109 |
+
],
|
| 110 |
+
"rg_relation_f1": [
|
| 111 |
+
0.7379,
|
| 112 |
+
0.8355
|
| 113 |
+
],
|
| 114 |
+
"tbfact_score": [
|
| 115 |
+
0.9053,
|
| 116 |
+
0.9383
|
| 117 |
+
],
|
| 118 |
+
"innovation_score": [
|
| 119 |
+
0.8526,
|
| 120 |
+
0.9007
|
| 121 |
+
],
|
| 122 |
+
"article_score": [
|
| 123 |
+
0.8522,
|
| 124 |
+
0.8595
|
| 125 |
+
]
|
| 126 |
+
}
|
| 127 |
+
}
|