Table 2.

Odds of HCC in study populations.

Study	Arms	Odds ratio	Sample size	Sensitivity (95% CI)	Specificity (95% CI)	Accuracy (95% CI)
Kim et al. (2020)	Optimized CNN architecture	0.73	549	0.8700	0.93	0.90
	Expert radiologist			0.9800	0.92	0.91
	Less expert radiologist			0.8600	0.93	0.94
Gao et al. (2021)	STIC (Spatial Extractor-Temporal Encoder Integration-Classifier)	1.69	60	0.8650	0.87	0.73
	Doctors’ consensus			0.7840	0.95	0.71
	AI assisted doctors			0.8330	0.92	–
Hamm et al. (2019)	Concept convolutional neural network (CNN) based deep learning system (DLS)	2.08	88	0.9000	0.98	–
	Radiologist 1			0.7000	1.00	–
	Radiologist 2			0.6000	1.00	–
Urhuț et al. (2023)	Unblinded clinician	0.69	24	0.5833	1.00	0.83
	Blinded clinician			0.5000	1.00	0.80
	AI based Contrast-enhanced ultrasound (CEUS)			0.8691	0.56	0.70
Liu et al. (2023)	RCNN model	Infinite	21	0.9600	–	–
	Radiologist 1			0.9270	–	–
	Radiologist 2			0.9170	–	–
Zhen et al. (2020)	Model A: seven-way classifier with six sequences.	2.94	47	0.8720	0.92	–
	Model B: seven-way classifier with 3 unenhanced sequences			0.7450	0.86	–
	Radiologists’ consensus			0.8720	0.95	–
	Model E: three-way classifier with sixes sequences.			0.9360	0.67	–
	Model F: three-way classifier with six sequences and clinical data			0.9570	0.96	–
	Model G: three-way classifier with three sequences and clinical data.			0.9570	0.90	–
	Radiologists’ consensus			0.8910	0.90	–
Nishida et al. (2022)	CNN AI model 1	1.2	18	0.6110	0.84	0.86
	CNN AI model 2			0.7220	0.88	0.87
	CNN AI model 3			0.7780	0.90	0.93
	Physicians			0.6910	–	0.69