Artificial intelligence assists identification and pathologic classification of glomerular lesions in patients with diabetic nephropathy

Table 3 The consistency between the CNN-based and pathologist-based pathologic classes

Cohen’s kappa value	CNN-based classes
	Internal application subset			External application subset
	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
Pathologist 1	0.475	0.457	0.529	0.667	0.585	0.640
Pathologist 2	0.688	0.722	0.688	0.655	0.588	0.423
Pathologist 3	0.653	0.655	0.654	0.699	0.618	0.741
Total	0.604	0.587	0.624	0.663	0.592	0.602

Model 1 was the CNN-based classification derived from combination of percent GS, the presence of KW lesions with the optimal cutoff values of average mesangial area; Model 2 derived from combination of percent GS, the presence of KW lesions with the optimal cutoff values of mesangial area/mesangial cell ratio; Model 3 derived from combination of percent GS, the presence of KW lesions with the optimal cutoff values of mesangial area fraction

ISSN: 1479-5876