Use of smartphones to estimate carbohydrates in foods for

Use of smartphones to estimate
carbohydrates in foods for diabetes
management
Jurong HUANG, Hang DING, Simon MCBRIDE, David IRELAND, Mohan KARUNANITHI
Presenter: Hang Ding | [email protected] | HIC 2015
3 August 2015
HEALTH AND BIOSECURITY
Prevalence of diabetes
380 million
Adults with
diabetes worldwide
1.5 million
Deaths directly
caused by diabetes
2 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Traditional estimation of carbohydrate
3 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Smartphone Approach
Food
Classifier
OpenCV
Volume
Estimator
Camera
OS
Nutrition
Database
Carbohydrate
Calculation
4 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Food Classifier
Three Features
Shape
Colour
(scale Invariant Feature Transform)
(RGB elements)
Support Vector Machine
5 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Texture
(Local Binary Pattern)
Volume Estimator
Object with calibrated size
Food Photo
Objects extracted
6 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Estimated Volume
Evaluation of Classification
10 types of fruits, 60 photos each
(orange, apple, pear, tomato, strawberry, banana, mango,
avocado, pineapple, and kiwi fruit)
Randomized
Training Data
10 types, 50 photos each
Test Data
10 types, 10 photos each
ACC =
(TP + TN)
Optimized Classification
Parameters
Accuracy of
Classification
7 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
TP + TN + FP + FN
Classification Accuracy
Accuracy of Classification
Types of Tested Fruits
8 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Accuracy of Carbo estimation
Table 1. Summary of the volume and carbohydrate estimations, compared with the
actual values measured from the water displacement and weight scale.
Test Item
Peach
Apple1
Apple2
Apple3
Tomato1
Tomato2
Average Error
Model
Volume
(ml)
158
165
172
201
21
17
Actual
Volume
(ml)
151
173
190
198
22
19
Error Rate
(%)
4.43
4.85
13.9
1.49
4.76
11.7
6.86
Estimated
Carbs (g)
16.3
18.5
19.3
22.5
0.74
0.62
9 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Actual Carbs
(g)
15.9
21.3
22.4
23.7
0.78
0.66
Error Rate
(%)
2.45
15.1
16.1
3.56
5.41
6.45
8.18
Future work
•Improvement of the approach
•Combination with other techniques
•Evaluation through clinical studies
10 | Use of smartphones to estimate carbohydrates in foods for diabetes management | Hang Ding, 3 August 2015
Dr. David Hansen
Dr. Mohan Karunanithi
Dr. Simon McBride
Dr. David Ireland
Dr. Farhad Fatehi
Prof. Len Gray
Prof. Anthony Russell
Ms. Denise Bennetts
Ms. Dominique Bird
Mr. Jurong Huang
Thank you
Contact Us
Phone: 1300 363 400 or +61 3 9545 2176
Email: [email protected] Web: www.csiro.au