Table 3 Ablation study on the effect of our proposed masked entity modeling loss and disease probability soft guidance on model performance.
From: Disease probability-enhanced follow-up chest X-ray radiology report summary generation
Method | MEM\(^{\dagger }\) | ESG\(^{\ddagger }\) | Natural language processing metric | Clinical metric | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
B1 | B2 | B3 | B4 | M | R | C | Acc5 | Acc14 | F1-5 | F1-14 | Rad-F1 | |||
EKAID7 | Â | Â | 62.8 | 55.3 | 49.1 | 43.4 | 33.9 | 57.7 | 1.027 | \(\text {68.2}\) | \(\text {81.0}\) | 50.1 | 46.6 | 31.6 |
DPE-base | Â | Â | 57.4 | 51.0 | 45.6 | 40.7 | 41.6 | 59.0 | 0.968 | 59.9 | 77.6 | 41.9 | 35.7 | 32.7 |
DPE-mem | \(\checkmark\) | Â | 61.4 | 54.7 | 49.2 | 44.1 | 40.5 | 61.6 | 1.421 | 65.9 | 79.1 | 43.3 | 42.3 | 29.5 |
DPE-esg | Â | \(\checkmark\) | 63.8 | 57.3 | 51.9 | 46.9 | 39.9 | 63.6 | 1.615 | 68.3 | 81.6 | 49.3 | 44.4 | 27.6 |
DPE-esg14 | Â | \(\checkmark\) | 64.5 | 58.2 | 52.9 | 48.0 | 40.8 | 65.6 | 1.684 | 69.9 | 82.7 | 51.6 | 46.2 | 25.4 |
DPE-esg26 | Â | \(\checkmark\) | 67.8 | 61.2 | 55.7 | 50.7 | 42.2 | 68.4 | 2.025 | 75.4 | 84.5 | 56.9 | 54.9 | 36.2 |
DPE-all | \(\checkmark\) | \(\checkmark\) | 65.6 | 58.8 | 53.1 | 48.0 | 39.4 | 62.9 | 1.680 | 69.7 | 81.9 | 49.9 | 45.9 | 28.1 |
DPE-all14 | \(\checkmark\) | \(\checkmark\) | 66.2 | 59.8 | 54.3 | 49.2 | 40.0 | 64.9 | 1.762 | 71.3 | 82.8 | 49.6 | 46.4 | 27.1 |
DPE-all26 | \(\checkmark\) | \(\checkmark\) | 69.0 | 62.6 | 57.2 | 52.4 | 42.2 | 69.2 | 2.118 | 76.5 | 85.2 | 58.5 | 55.9 | 36.7 |
