Table 3 Comparison of machine learning variants in YOLOv8.

From: Task design for crowdsourced glioma cell annotation in microscopy images

 

Astrocytes

Tumor cells

Mean

TPR

PPV

AP50

AP@

\(F_1\) score

TPR

PPV

AP50

AP@

\(F_1\) score

AP50

AP@

\(F_1\) score

Var1

0.52

0.27

0.14

0.01

0.35

0.55

0.90

0.16

0.03

0.68

0.15

0.02

0.52

Var2

0.57

0.36

0.20

0.03

0.44

0.53

0.93

0.17

0.03

0.67

0.19

0.03

0.56

Var3

0.33

0.36

0.09

0.01

0.35

0.50

0.93

0.23

0.05

0.65

0.16

0.03

0.50

Var4

0.40

0.63

0.21

0.03

0.49

0.51

0.92

0.17

0.03

0.66

0.19

0.03

0.58

Var5

0.59

0.44

0.25

0.04

0.51

0.54

0.76

0.30

0.06

0.63

0.27

0.05

0.57

Var6

0.55

0.51

0.23

0.04

0.53

0.50

0.78

0.25

0.04

0.61

0.24

0.04

0.57

Var7

0.56

0.38

0.19

0.03

0.45

0.57

0.86

0.22

0.04

0.69

0.20

0.03

0.57

Var8

0.42

0.37

0.13

0.02

0.40

0.51

0.86

0.24

0.04

0.64

0.19

0.03

0.52

Var9

0.52

0.33

0.14

0.02

0.40

0.51

0.88

0.31

0.07

0.65

0.22

0.04

0.53

  1. TPR refers to the true positive rate and PPV to positive predictive value. TPR, PPV, and \(F_1\) score are based on an intersection over union of at least 0.35 between overlapping bounding boxes. AP50 and AP@ referring to AP@\([0.5:0.05:0.95]\) include the average precision. Var1–4 contain the markers DAPI, ATRX, and GFAP; Var5–6 DAPI and GFAP; Var7–8 DAPI, GFAP and other; Var9 DAPI, ATRX, GFAP, and other; also differing in their colors and intensities (details in Fig. 1).
  2. Best variant highlighted in bold.
Back to article page