Table 2.

The confusion matrix of a modified U-Net CNN whose task was to separate 2800 chest X-rays of the set as into four different classes: negative (class 1), COVID-19 (class 2), pneumonia (class 3), or tuberculosis (class 4). We see, for instance, that the CNN classified most X-rays showing tuberculosis as COVID-19.

True$\backslash$Predicted class	1	2	3	4	Sum
1	120	7	9	4	$n_{1·}$ = 140
2	15	116	3	6	$n_{2·}$ = 140
3	12	13	115	0	$n_{3·}$ = 140
4	2	96	4	38	$n_{4·}$ = 140
Sum	$n_{·1}$ = 149	$n_{·2}$ = 232	$n_{·3}$ = 131	$n_{·4}$ = 48	n = 560

Based on this confusion matrix, the CNN has accuracy of 0.847, sensitivity of 0.695, specificity of 0.898, macro-average precision of 0.744, micro-average precision of 0.695, Youden’s index of 0.593, macro-average F1-score of 0.677, micro-average F1-score of 0.695, $\kappa$ of 0.598, and MCC of 0.616.