Emerson M. Del Ponte
Sarah Pethybridge
Clive Bock
To diagram or not to diagram:
Is there a future for diagrammatic
visual severity assessments aids?
Slide 2
Slide 2 text
Visual (severity) assessment aids
Disease quantification
- Long history, extensive use
- New concepts over the decades
- Technology has key influence
- Active (peer-reviewed) research
- Terminology - not clear!
Slide 3
Slide 3 text
What is Severity?
Conceptually and operationally
Ratio (%)
Ordinal
(midpoint %)
Class
(% interval)
Ordinal
Slide 4
Slide 4 text
Where do diagrams/images fit?
International Working
group on soybean rust
Nominal
Ratio
Easier
decision
Tougher
decision
Slide 5
Slide 5 text
Most common use:
Godoy et al (2006)
Standard Area Diagram set (SADs)
Slide 6
Slide 6 text
Hybrid systems
Peterson et al. (1948)
- 0-100 score
- % severity
Ordinal (0-100)
Ratio (%)
Slide 7
Slide 7 text
Hybrid system
Tovar-soto et al (2002)
Ratio (%)
Ordinal (0-6)
Slide 8
Slide 8 text
Hybrid system
RamosandIslas (2015)
Scores and severity interval (class)
Ordinal
Class
(% interval)
Slide 9
Slide 9 text
Belan et al. (2014)
multiple systems:
Ordinal
Class
(% interval) Ratio (%)
Slide 10
Slide 10 text
Multiple interactive systems
Slide 11
Slide 11 text
Confused terminology
Are they the same?
- Standard Area Diagrams (SAD)
- Diagrammatic (nominal, ordinal) Scales
- Disease Diagrams
- Standard Area Diseased Images
Slide 12
Slide 12 text
SAD: active research last 25 years
Slide 13
Slide 13 text
Technology
for diagram
preparation
Bock et al. (2016)
Slide 14
Slide 14 text
Statistical evaluation of the tool
Research data to assess the visual aids for:
Bock et al. (2016)
Slide 15
Slide 15 text
SAD preparation and evaluation
Need to obtain "assumed actual" severity:
Repeated assessments using Assess® - Bock et al (2016)
Bock et al. (2016)
Slide 16
Slide 16 text
Methods: software dominates
Pethybridge and Nelson (2015)
Del Ponte et al. (2017)
Slide 17
Slide 17 text
Methods: Incremental scale
H-B : Horsfall-Barratt scale (1945)
Del Ponte et al. (2017)
Slide 18
Slide 18 text
Do SADs work?
Del Ponte et al (unpublished)
unaided
SAD-aided
Larger gains
Overall mean gain
in precision:
0.08
lower gains
Slide 19
Slide 19 text
Del Ponte et al (unpublished)
Lower gain
Minimal gain
Disease characteristics?
Slide 20
Slide 20 text
Evaluation of the Estimate app
Cercospora
Leaf Spot
on Table
Beets
Del Ponte EM, Pethybridge S, et al. unpublished
Slide 21
Slide 21 text
Two incremental scales
H-B
scale
10%-linear
Slide 22
Slide 22 text
Second step: pick a unitary %
Note:
Same image!
Slide 23
Slide 23 text
Experimental set:
4 methods and 2 types of variables
Slide 24
Slide 24 text
Unaided: baseline accuracy
Mean (n=30)
pc = 0.84
r = 0.78
Cb = 0.93
Del Ponte et al. (unpublished)
Slide 25
Slide 25 text
One-step: ordinal data
H-B scale 10%-linear
Del Ponte et al. (unpublished)
Slide 26
Slide 26 text
How good were the methods?
ρ
c
= 0.94
r = 0.95
C
b
= 0.99
ρ
c
= 0.85
r = 0.82
C
b
= 0.96
ρ
c
= 0.86
r = 0.82
C
b
= 0.96
ρ
c
= 0.86
r = 0.81
C
b
= 0.94
Unaided
ρ
c
= 0.84
r = 0.78
C
b
= 0.93
Del Ponte et al. (unpublished)
Slide 27
Slide 27 text
Conclusions
- H-B and Linear scales no better than no aid
- Two-stage better: Linear + direct estimate
- Estimate app needs revisions
- Single step of severity?
- Compare with standard static SAD?