Table 4 Meta-regression analysis of deep learning algorithm performance based on patient-based analysis in external validation cohorts for diagnosing microsatellite instability-high (MSI-H) in colorectal cancer using whole slide images

From: Systematic review and meta-analysis of deep learning for MSI-H in colorectal cancer whole slide images

Subgroup

Studies, n

Sensitivity(95%CI)

Meta-regression P-value

Specificity(95%CI)

Meta-regression P-value

Center

  

0.03

 

0.21

Single center

2

0.77 (0.51–0.1.00)

 

0.88 (0.66–1.00)

 

Multi-center

19

0.93 (0.90–0.97)

 

0.68 (0.55–0.82)

 

Optimal AI algorithm

  

0.23

 

0.06

CNN

16

0.93 (0.89–0.97)

 

0.65 (0.50–0.80)

 

Non-CNN

5

0.91 (0.83–1.00)

 

0.80 (0.69–1.00)

 

Reference standard

  

<0.001

 

0.23

Only PCR

7

0.89 (0.80–0.98)

 

0.62 (0.37–0.86)

 

non-only PCR

14

0.94 (0.91–0.98)

 

0.75 (0.61–0.89)

 

Tile size

  

0.87

 

<0.001

224*224 or 256*256

15

0.94 (0.90–0.98)

 

0.58 (0.44–0.72)

 

512*512

4

0.88 (0.75–1.00)

 

0.91 (0.81–1.00)

 

Magnification

  

0.71

 

0.53

20*

15

0.95 (0.91–0.98)

 

0.66 (0.50–0.82)

 

40*

3

0.86 (0.70–1.00)

 

0.76 (0.47–1.00)

 
  1. CNN Convolutional Neural Network, PCR polymerase chain reaction.