Skip to main content

Table 3 Results (pixel level accuracy) on 12 layered U-Net architecture initialized using He Normal. a) Results on the validation slides in dataset S1. b) Results on the test slides in dataset S1

From: Semantic segmentation to identify bladder layers from H&E Images

  a b
Accuracy True Positive False Negative Accuracy True Positive False Negative
Background 0.99 91,488,149 672,063 0.99 119,216,562 45,508
Lamina Propria 0.99 27,861,365 278,336 0.98 53,285,338 1,138,536
Muscularis Propria 0.87 105,691,097 15,302,863 0.88 22,362,961 3,068,361
Mucosa 0.90 11,391,120 1,289,356 0.97 39,830,580 1,280,450
RBC 0.99 10,701,972 143,239 0.93 1,896,309 135,229
Cautery 0.64 6,612,587 3,740,058 0.41 9,610,122 13,927,259
Inflammation 0.97 41,619,905 1,393,986 0.41 7,082,849 10,235,456
Muscularis Mucosa 0.84 1,371,094 258,490 N/A N/A N/A