Table 2.

Metrics details of the different models for predicting BI-RADS scores

	Accuracy	Precision	Recall	F1 score (weighted)
RNN-ATT	0.823 (0.802, 0.843)	0.741 (0.703, 0.781)	0.656 (0.625, 0.686)	0.811 (0.790, 0.833)
RNN-ATT_TF	0.836 (0.744, 0.816)	0.780 (0.744, 0.816)	0.678 (0.649, 0.709)	0.825 (0.805, 0.847)
RadioLOGIC	0.850 (0.832, 0.869)	0.811 (0.778, 0.844)	0.692 (0.662, 0.722)	0.838 (0.817, 0.859)
RadioLOGIC_TF	0.906 (0.890, 0.921)	0.871 (0.846, 0.896)	0.819 (0.791, 0.846)	0.903 (0.887, 0.919)

Note: values in parentheses are 95% confidence intervals. BI-RADS, breast imaging-reporting and data system; TF, transfer learning; RNN, recurrent neural network; ATT, attention mechanism; RadioLOGIC, radiological repomics-driven model incorporating medical token cognition.