Table 4 Accuracy in extracting complete descriptions of breast lesions by the NLP system
Entity type | Imaging features | Recall | Precision |
|---|---|---|---|
Mass | Location | 90.1% | 95.7% |
Shape | 85.4% | 94.1% | |
Size | 90.4% | 94.2% | |
T1WI | 90.3% | 94.1% | |
T2WI | 89.6% | 94.1% | |
DWI | 88.7% | 93.2% | |
Margin | 90.9% | 95.9% | |
Internal enhancement | 91.6% | 90.2% | |
Enhancement kinetic curve | 91.6% | 95.7% | |
NME | Location | 90.9% | 95.2% |
Distribution pattern | 86.2% | 94.1% | |
Scope | 89.2% | 92.3% | |
T1WI | 91.1% | 93.6% | |
T2WI | 88.9% | 93.2% | |
DWI | 88.6% | 94.0% | |
Internal enhancement | 91.3% | 91.5% | |
Enhancement kinetic curve | 90.9% | 95.0% | |
Lymphadenopathy | 98.7% | 87.7% | |
Invasion | Nipple | 98.6% | 86.4% |
Skin | 97.4% | 86.8% | |
Chest wall | 97.7% | 86.9% | |
Pectoralis muscle | 96.2% | 85.8% | |
BI-RADS category | 96.6% | 94.8% | |
Overall | 91.5% | 92.9% |