Table 4 Network performance on various staining and scanning conditions

From: Artificial intelligence in neuropathology: deep learning-based assessment of tauopathy

 

Background

Metrics

High (Aperio)

Low (Philips)

True positives (TP)

329

244

False positives (FP)

98

122

False negatives (FN)

14

45

Precision

0.77

0.67

Recall

0.96

0.85

F1 score

0.85

0.75

  1. The FCN was validated in various staining conditions and in various backgrounds intensity. The TP, FP, and FN values in high background WSI (Aperio scanned AD and PART cases from UTSW) were 329, 98, and 14, respectively. The TP, FP, and FN values in low background WSI (Philips scanned PSP and CTE cases from MSSM) were 244, 122, and 45, respectively. Overall FCN performance exemplified with an F1 score was higher in the high-background high-tau-burden AD/PART cohort (0.85) versus in the low-background and low-to-moderate tau-burden PSP/CTE cohort (0.75). An F1 score harmonic mean is calculated as 2*Precision*Recall/(Precision + Recall)
Back to article page