Transfer language space with similar domain adaptation: a case study with hepatocellular carcinoma

Table 3 Performance of language model and classifiers on structured US reports (reports with template) and unstructured US reports (reports without template). Models trained over US domain as well as cross-domain models (US-finetuned) have been tested

	Report document with template			Report document without template
	Precision	Recall	f1-score	Precision	Recall	f1-score
	US-finetuned Word2Vec+Random Forest Classifier
Malignant	0.68	0.52	0.59	0.75	0.27	0.40
Benign	0.95	0.97	0.96	0.94	0.99	0.97
	US Word2Vec+Random Forest Classifier
Malignant	0.71	0.59	0.64	0.67	0.18	0.29
Benign	0.96	0.97	0.96	0.94	0.99	0.96
	US-finetuned Word2Vec Embedding+1DCNN
Malignant	0.68	0.66	0.67	0.70	0.64	0.67
Benign	0.96	0.97	0.96	0.97	0.98	0.97
	US Word2Vec Embedding+1DCNN
Malignant	0.68	0.72	0.70	0.64	0.64	0.64
Benign	0.97	0.96	0.97	0.97	0.97	0.97
	US-finetuned BERT+Random Forest Classifier
Malignant	0.91	0.34	0.50	0.33	0.09	0.14
Benign	0.93	1.00	0.96	0.93	0.98	0.96
	US BERT+Random Forest Classifier
Malignant	0.67	0.34	0.45	0.33	0.09	0.14
Benign	0.93	0.98	0.96	0.93	0.98	0.96

ISSN: 2041-1480