ESTRO 2024 - Abstract Book

S3086

Physics - Autosegmentation

ESTRO 2024

120 data sets were analysed, equally divided by the two sub-groups (AUTO-GUIDED and NO AUTO-GUIDED) and by treating physician (ALF, ADF and GS). Mean values for DSC, JC, HD and MDA of the bladder and for the rectum are reported in Table I. When comparing the whole group of patients, we observed a smaller difference between the clinical contours and the autosegmented ones for both the rectum and the bladder in the AUTO-GUIDED over the NO AUTO-GUIDED subgroup. The difference, albeit small, was statistically significant for all indexes but HD for both the rectum and the bladder. Stratification by physician showed the same trend for all the metrics although the pattern was variable among the observers.

Bladder

Rectum

DSC

JC

HD

MDA

DSC

JC

HD

MDA

ALL

AUTO-GUIDED (60)

0.94

0.90

11.9

0.97

0.88

0.80

13.4

1.35

NO AUTO-GUIDED (60)

0.92

0.86

13.5

1.38

0.84

0.72

18

1.89

p

<0.001

<0.001

0.21

<0.001

<0.001

<0.001

0.23

<0.001

ADF

AUTO-GUIDED (20)

0.96

0.93

11.1

0.67

0.89

0.81

13.9

1.28

NO AUTO-GUIDED (20)

0.91

0.84

15.3

1.60

0.84

0.72

16.5

1.83

p

0.004

0.002

0.28

<0.001

0.001

0.001

0.30

0.004

ALF

AUTO-GUIDED (20)

0.96

0.93

11.4

1.48

0.87

0.78

16.9

1.48

NO AUTO-GUIDED (20)

0.93

0.87

15.1

1.57

0.85

0.74

16.9

1.57

p

0.004

0.005

0.19

0.002

0.36

0.33

0.27

0.64

GS

AUTO-GUIDED (20)

0.93

0.87

10.2

1.17

0.90

0.82

15.4

1.29

NO AUTO-GUIDED (20)

0.90

0.85

13.2

1.60

0.82

0.70

22.9

2.27

p

0.15

0.11

0.92

0.05

0.003

0.003

0.008

0.001

Conclusion:

The present data prove that, even for well-known contours, the availability of a preliminary AC leads to different than usual contours in clinical practice, suggesting a potential automation bias that needs to be further investigated.

Keywords: preliminary autocontours, automation bias

References: