Skip to main content

Table 8 Effect of the inverse regression-based prediction and measurement error correction on Ki67 dichotomisation accuracy at various reference value cutoffs (n = 164)

From: A methodology to ensure and improve accuracy of Ki67 labelling index estimation by automated digital image analysis in breast cancer tissue

Method Underestimated (%) Overestimated (%) Total misclassified (%)
Ki-67 cutoff >10%    
Ki67-VE-median 16/148 (11) 2/16 (13) 18 (11)
Ki67-DIA-2 0/148 (0) 12/16 (75) 12 (7)
Ki67-DIA-2 corrected 0/148 (0) 9/16 (56) 9 (5)
Ki67-DIA-2 corrected <40* 2/148 (1) 6/16 (38) 8 (5)
Ki-67 cutoff >15%    
Ki67-VE-median 22/136 (16) 1/28 (4) 23 (14)
Ki67-DIA-2 2/136 (1) 13/28 (46) 15 (9)
Ki67-DIA-2 corrected 3/136 (2) 11/28 (46) 14 (9)
Ki67-DIA-2 corrected <40* 5/136 (4) 6/28 (21) 11 (7)
Ki-67 cutoff >20%    
Ki67-VE-median 28/123 (23) 1/41 (2) 29 (18)
Ki67-DIA-2 2/123 (2) 9/41 (22) 11 (7)
Ki67-DIA-2 corrected 2/123 (2) 12/41 (29) 14 (9)
Ki67-DIA-2 corrected <40* 6/123 (5) 6/41 (15) 12 (7)
  1. *Ki67-DIA-2 < 40 - represents a regression model for Ki67-DIA-2 with only Ki67-Count less than 40% cases included in the analysis (n = 92). DIA, digital image analysis; VE, visual estimate.