Table 3.

The performance of supervised machine learning models (e.g., neural network, k nearest neighbor (kNN), and logistic regression) in comparison to 5 technicians.

Methods	AUC	CA	Sensitivity	Specificity	Precision	F1
Neural network	0.92	0.906	0.906	1	0.938	0.913
kNN	0.912	0.844	0.844	0.83	0.881	0.855
Logistic regression	0.927	0.906	0.906	0.83	0.914	0.909
Average	0.920	0.885	0.885	0.887	0.911	0.892
Technician 1	–	0.781	0.778	0.800	0.955	0.857
Technician 2	–	0.750	0.731	0.833	0.950	0.826
Technician 3	–	0.813	0.846	0.667	0.917	0.880
Technician 4	–	0.813	0.885	0.500	0.885	0.885
Technician 5	–	0.813	0.815	0.800	0.957	0.880
Average	–	0.794	0.811	0.720	0.932	0.866

The k-fold cross validation sampling methods (e.g., k = 2, 3, 5) and leave-one-out method were used to test and train the data. The performance of naïve Bayes model is well below the average human being (details in Supplementary Fig. 7). The abbreviations used were area under the curve (AUC), classification accuracy (CA), and F1-score is the harmonic mean for precision and sensitivity.