Comparison of predicted t_1/2 values from the three random forest models with their true values from the test dataset. Each model is trained on properties derived from the aperture radial ranges identified by the saliency maps, as listed in Table B.1. Results for the model trained with the aperture range from CNN Model 1 are shown in the first row, CNN Model 2 in the second row, and CNN Model 3 in the third row. Left panel: Two-dimensional joint probability density functions comparing true and predicted t_1/2 values. Black points represent predicted versus true values for the random forest model trained on the saliency-identified aperture range. The dotted gray 45-degree line indicates perfect prediction. Right panel: Probability density functions of the relative errors between predicted and true t_1/2 values, highlighting the median error and standard deviation.