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

Table 2 Performance of language model and classifiers on structured MR reports (reports with template) and unstructured MR reports (reports without template). Models trained over MR domain as well as cross-domain models (MR-finetuned) have been tested

	Report document with template			Report document without template
	Precision	Recall	f1-score	Precision	Recall	f1-score
	MR-finetuned Word2Vec+Random Forest Classifier
Malignant	0.94	0.90	0.92	0.73	0.76	0.74
Benign	0.93	0.95	0.94	0.94	0.93	0.94
	MR Word2Vec+Random Forest Classifier
Malignant	0.95	0.93	0.94	0.70	0.76	0.73
Benign	0.95	0.96	0.95	0.94	0.92	0.93
	MR-finetuned Word2Vec Embedding+1DCNN
Malignant	0.95	0.98	0.96	0.68	0.62	0.65
Benign	0.98	0.96	0.97	0.91	0.93	0.92
	MR Word2Vec Embedding+1DCNN
Malignant	0.97	0.96	0.97	0.40	0.19	0.26
Benign	0.97	0.98	0.98	0.83	0.93	0.88
	MR-finetuned BERT+Random Forest Classifier
Malignant	0.94	0.90	0.92	0.71	0.57	0.63
Benign	0.93	0.95	0.94	0.91	0.95	0.92
	MR BERT+Random Forest Classifier
Malignant	0.94	0.91	0.92	0.77	0.48	0.59
Benign	0.94	0.95	0.94	0.89	0.97	0.93

ISSN: 2041-1480