Skip to main content

Table 2 Overall distribution of the diagnostic classes from the residual CNNs and the independent neuropathologist with the inter-rater agreement between CNNs correct predictions and the neuropathologist (NP)

From: Novel rapid intraoperative qualitative tumor detection by a residual convolutional neural network using label-free stimulated Raman scattering microscopy

   NP CNN 1 CNN 2 CNN 3
SRH images n = 592     
(All) 100%     
Tumor   462 (78.0%) 459 (77.5%) 456 (77%) 469 (79.2%)
Non-tumor   113 (19.1%) 115 (19.4%) 117 (19.8%) 110 (18.6%)
Low quality   17 (2.9%) 18 (3%) 19 (3.2%) 13 (2.2%)
SRH images (three random areas) n = 402     
   (Ground truth)    
Tumor   326 (100%) 297 (91.1%) 298 (91.4%) 305 (95.0%)
Non-tumor   67 (100%) 46 (68.7%) 48 (71.6%) 49 (73.1%)
Low quality   9 (100%) 5 (55.6%) 5 (55.6%) 6 (66.7%)