Artificial intelligence performance in detecting lymphoma from medical imaging: a systematic review and meta-analysis

Table 4 Summary estimate of pooled performance of artificial intelligence in lymphoma detection

	No. of studies				P value^b				P value
Overall	16	Sensitivity	P value ^a	I²(95%CI)		Specificity	P value	I² (95%CI)
Algorithm					0.11				0.83
Deep Learning	13	0.86 (0.80–0.90)	< 0.05	99.41 (99.37–99.47)		0.94 (0.92–0.96)	< 0.05	99.71 (99.70–99.72)
Machine Learning	3	0.93 (0.88–0.95)	< 0.05	91.47 (88.74–94.21)		0.92 (0.87–0.95)	< 0.05	87.72 (83.33–92.10)
Transfer Learning Applied					0.92				0.55
Yes	6	0.88 (0.80–0.93)	< 0.05	99.67 (99.65–99.69)		0.95 (0.92–0.97)	< 0.05	99.85 (99.84–99.85)
No	10	0.85 (0.80–0.89)	< 0.05	91.29 (89.67–92.91)		0.91 (0.88–0.93)	< 0.05	92.39 (91.04–93.75)
Human Clinicians versus Algorithms				0.01				< 0.05
Clinicians	3	0.70 (0.65–0.75)	< 0.05	77.53 (69.54–85.53)		0.86 (0.82–0.89)	< 0.05	84.09 (78.94–89.23)
Algorithms	13	0.91 (0.86–0.94)	< 0.05	99.60 (99.58–99.62)		0.96 (0.93–0.97)	< 0.05	99.81 (99.80–99.82)
Sample size					0.45				0.39
≤ 200	11	0.88 (0.84–0.92)	< 0.05	98.71 (98.55–98.86)		0.91 (0.87–0.94)	< 0.05	99.02 (98.91–99.13)
> 200	5	0.86 (0.78–0.91)	< 0.05	99.47 (99.43–99.50)		0.95 (0.92–0.97)	< 0.05	99.77 (99.76–99.78)
Geographical distribution					0.67				0.51
Asia	10	0.88 (0.83–0.91)	< 0.05	99.34 (99.30–99.38)		0.94 (0.92–0.96)	< 0.05	99.71 (99.70–99.72)
Non Asia	6	0.83 (0.72–0.90)	< 0.05	99.23 (99.09–99.36)		0.91 (0.82–0.96)	< 0.05	99.40 (99.31–99.50)

ISSN: 1472-6947