Table 4 Sample-level feature importance of sample pair 1 and 2 for both imaging and pathologic report provided by LIME algorithm
From: Detection of medical text semantic similarity based on convolutional neural network
Sample pair No. | Imaging report | Pathologic report | ||||
|---|---|---|---|---|---|---|
Word-Chinese | Word-English | Feature importance of word | Word-Chinese | Word-English | Feature importance of word | |
1 (Prediction probability = 0.77) | 胎膜 | Fetal membranes | 0.15 | 胎儿 | Fetus | 0.12 |
脐带 | Umbilical cord | 0.14 | 胎心 | Fetal heart | 0.03 | |
胎盘 | Placenta | 0.06 | 羊水 | Amniotic fluid | 0.03 | |
毛发 | Hair | 0.04 | 头位 | Head position | 0.02 | |
小叶 | Lobule | 0.02 | 股骨 | Femur | 0.01 | |
面灰 | Face ash | 0.01 | 单胎 | Single fetus | 0.01 | |
2 (Prediction probability = 0.83) | 甲状腺 | Thyroid | 0.19 | 甲状腺 | Thyroid | 0.16 |
结节 | Tubercle | 0.15 | 腺体 | Glandular body | 0.14 | |
滤泡 | Follicular | 0.08 | 右叶 | Right lobe | 0.07 | |
右叶 | Right lobe | 0.03 | 包膜 | Envelope | 0.03 | |
腺瘤 | Adenoma | 0.01 | 回声 | Echoes | 0.01 | |
切面 | Section | 0.01 | 血流 | Blood flow | 0.01 | |