Table 3.

Performance of the deep learning versus radiologists in classifying compression fractures in the human and deep learning collaboration dataset.

Dataset	AUC	Accuracy (%)	Sensitivity (%)	Specificity (%)	Precision (%)	F1 Score (%)	PPV (%)	NPV (%)
Deep learning model		76.4	64.7	82.3	65.0	64.8	64.7	82.3
Acute	0.780	69.3	67.5	71.4	73.0	70.1	73.0	65.8
Chronic	0.809	71.3	68.4	73.1	60.9	64.5	60.9	79.1
Pathologic	0.734	88.7	30.8	94.2	33.3	32.0	33.3	93.5
Trainee radiologist		70.7	56.0	78.0	55.7	55.8	56.0	78.0
Acute	0.573	62.0	65.0	58.6	64.2	64.6	64.2	59.4
Chronic	0.618	64.0	52.6	71.0	52.6	52.6	52.6	71.0
Pathologic	0.541	86.0	15.4	92.7	16.7	16.0	16.7	92.0
Competent radiologist		76.9	65.3	82.7	69.9	67.5	65.3	82.7
Acute	0.701	69.3	58.8	81.4	78.3	67.1	78.3	63.3
Chronic	0.782	78.0	78.9	77.4	68.2	73.2	68.2	85.7
Pathologic	0.665	83.3	46.2	86.9	25.0	32.4	25.0	94.4
Expert radiologist		78.2	67.3	83.7	67.4	67.4	67.3	83.7
Acute	0.707	70.7	70.0	71.4	73.7	71.8	73.7	67.6
Chronic	0.732	74.0	70.2	76.3	64.5	67.2	64.5	80.7
Pathologic	0.667	90.0	38.5	94.9	41.7	40.0	41.7	94.2
Deep learning and trainee		77.8	66.7	83.3	67.9	67.3	66.7	83.3
Acute	0.722	72.0	70.0	74.3	75.7	72.7	75.7	68.4
Chronic	0.744	75.3	70.2	78.5	66.7	68.4	66.7	81.1
Pathologic	0.610	86.0	30.8	91.2	25.0	27.6	25.0	93.3
Deep learning and competent radiologist		81.6	72.7	86.0	73.4	73.1	72.2	86.3
Acute	0.767	76.7	76.3	77.1	79.2	77.7	79.2	74.0
Chronic	0.779	79.3	71.9	83.9	73.2	72.6	73.2	83.0
Pathologic	0.729	88.7	53.8	92.0	38.9	45.2	38.9	95.5
Deep learning and expert radiologist		85.3	77.6	89.1	77.4	77.5	77.6	89.1
Acute	0.801	80.0	80.5	79.5	80.5	80.5	80.5	79.5
Chronic	0.825	83.3	78.9	86.0	77.6	78.3	77.6	87.0
Pathologic	0.751	92.7	53.8	96.4	58.3	56.0	58.3	95.7