Retinal vessel segmentation for
diabetic retinopathy screening test
< irene.fidone@mail.polimi.it, lara.cavinato@mail.polimi.it, marco.bacis@mail.polimi.it >
May 23rd, 2018

Soda 430-438 Woz Lounge @UC Berkeley
Irene Fidone, Lara Cavinato, Marco Bacis
Diabetes and Diabetic Retinopathy
Metabolic disease
with high blood
sugar level
http://www.patiadiabetes.com/en/
2
https://www.dreamstime.com/stock-illustration-diabetes-mellitus-info-graphic-complications-detailed-affected-organs-beautiful-colorful-design-image62194003
Uncontrolled
diabetes
causes health
complications
2
3
Chronically high blood sugar from diabetes is associated with damage to
the tiny blood vessels in the retina, leading to diabetic retinopathy [1]
[1] https://nei.nih.gov/health/diabetic/retinopathy
Diabetes and Diabetic Retinopathy
https://www.dreamstime.com/stock-illustration-diabetes-mellitus-info-graphic-complications-detailed-affected-organs-beautiful-colorful-design-image62194003
4
Manual segmentationSpecific equipment Qualified staff
Screening Test
5Our Proposal 5
Segmentation Algorithm
Automated the retinal vessel segmentation
for diabetic retinopathy screening
66
Software Implementation Hardware accelerationSegmentation Algorithm
Beye Implementation
The segmentation was computed on the images of Drive and Stare databases, and then
compared with the respective manual segmentations
77Methodology and Benchmarking
DRIVE database [2] STARE database [3]
20 Images: 605 × 700 pixels,
8 bits per color channel
20 Images: 768 × 584pixels,
8 bits per color channel
[2] https://www.isi.uu.nl/Research/Databases/DRIVE/ [3] http://cecas.clemson.edu/~ahoover/stare/
Key Performance Indicators:
SW Evaluation HW Evaluation
Accuracy
Execution Time
Energy Consumption
8Results 8
Accuracy
DRIVE database 92.93%
STARE database 90,3%
[4] M. A. Amin and H. Yan, “High speed detection of retinal blood vessels in fundus image using phase congruency,”
[5] J. Odstrcilik.“Retinal vessel segmentation by improved matched filtering: evaluation on a new high-resolution fundus image database,”[6] M. Al-Rawi, M. Qutaishat, and M. Arrar,
“An improved matched filter for blood vessel detection of digital retinal images,”
[9] B. Zhang, L. Zhang, L. Zhang, and F. Karray, “Retinal vessel extraction by matched filter with first-order derivative of gaussian,”
[7] M. G. Cinsdikici and D. Aydın, “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony) algorithm,”
TP + TN

Tot
Accuracy =
BeyeManual
True Positive
True Negative
False Negative
False Positive
Pixel comparison between manual
segmentation and Beye
The software implementation has been compared with other algorithms of the
state of the art implemented in software
99
[4] M. A. Amin et Al. “High speed detection of retinal blood vessels in fundus image using phase congruency,”
[5] J. Odstrcilik,et Al. “Retinal vessel segmentation by improved matched filtering: evaluation on a new high-
resolution fundus image database,”
[6] M. Al-Rawi et Al. “An improved matched filter for blood vessel detection of digital retinal images,”
Execution Time [s] in DRIVE Execution Time [s] in STARE
ExecutionTime
ExecutionTime
SpeedUp
[7] M. G. Cinsdikici et Al. “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony)
algorithm,”
[8] X. Jiang et Al.“Adaptive local thresholding by verificationbased multithreshold probing with application to vessel detection
in retinal images,”
[9] B. Zhang,et Al.“Retinal vessel extraction by matched filter with first-order derivative of gaussian,” Computers in biology
and medicine,
quello, che è un tempo, e gli altri
speedup)
SpeedUp
Results
The software implementation has been compared with other algorithms of the
state of the art implemented in software
9
[4] M. A. Amin et Al. “High speed detection of retinal blood vessels in fundus image using phase congruency,”
[5] J. Odstrcilik,et Al. “Retinal vessel segmentation by improved matched filtering: evaluation on a new high-
resolution fundus image database,”
[6] M. Al-Rawi et Al. “An improved matched filter for blood vessel detection of digital retinal images,”
Execution Time [s] in DRIVE Execution Time [s] in STARE
ExecutionTime
ExecutionTime
SpeedUp
[7] M. G. Cinsdikici et Al. “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony)
algorithm,”
[8] X. Jiang et Al.“Adaptive local thresholding by verificationbased multithreshold probing with application to vessel detection
in retinal images,”
[9] B. Zhang,et Al.“Retinal vessel extraction by matched filter with first-order derivative of gaussian,” Computers in biology
and medicine,
SpeedUp
55X
137X
260X
514X
162X
47X
29X
Speed Up•
quello, che è un tempo, e gli altri
speedup)
Results
The software implementation has been compared with other algorithms of the
state of the art implemented in software
• Comparison between of our software and hardware performances, in terms of power
consumption of the target devices and execution time for a generic image from DRIVE
10
Execution Time: Speed Up 6X

Energy consumption : Speed Up 5.7X
VS.
Intel Core i7-6700 CPU Xilinx Zynq-7000
Results
• Comparison between of our software and hardware performances, in terms of power
consumption of the target devices and execution time for a generic image from DRIVE
[10] A. Nieto, V. M. Brea, and D. L. Vilari˜no, “Fpga-accelerated retinal vessel-tree extraction,” in Field Programmable Logic and Applications,2009.
[11] D. Koukounis, C. Tttofis, and T. Theocharides, “Hardware acceleration of retinal blood vasculature segmentation,”
• Comparison between our hardware implementation and state of the art hardware
solutions, assessed by accuracy and hardware performance
Work A. Nieto et Al.[10] D. Koukounis et Al.[11] Beye
Device Spartan 3 Spartan 6 Zedboard
Accuracy 0.9100 0.9007 0.9285
Execution time 1.4 s 0.03185 s 0.01041 s
Frequency 53 MHz 100 MHz 100 MHz
Execution Time: Speed Up 6X

Energy consumption : Speed Up 5.7X
VS.
Intel Core i7-6700 CPU Xilinx Zynq-7000
10Results
11Conclusion
[12] Resnikoff S, Pascolini D, Etya’ale D, et al. Global data on visual impairment in the year 2002. Bulletin of the World Health Organization 2004; 82: 844–51.
[13] The Global Economic Cost of Visual Impairment: Summary Report April 2010
It is estimated that Diabetic Retinopathy accounts for 4.8% of the number of cases of
blindness worldwide.[12]

The financial cost of visual impairment worldwide was US$2,954 billion in 2010[13]
10Conclusion
Thank you for your attention
irene.fidone@mail.polimi.it
It is estimated that Diabetic Retinopathy accounts for 4.8% of the number of cases of
blindness worldwide.[12]

The financial cost of visual impairment worldwide was US$2,954 billion in 2010[13]
May 23rd, 2018

Soda 430-438 Woz Lounge @UC Berkeley
https://www.facebook.com/beye.project/
https://www.slideshare.net/BEye_project
https://twitter.com/BEye_NECSTLab

BEye: Software Implementation and Hardware Acceleration of Retinal Vessel Segmentation for Diabetic Retinopathy Screening Tests.

  • 1.
    Retinal vessel segmentationfor diabetic retinopathy screening test < irene.fidone@mail.polimi.it, lara.cavinato@mail.polimi.it, marco.bacis@mail.polimi.it > May 23rd, 2018 Soda 430-438 Woz Lounge @UC Berkeley Irene Fidone, Lara Cavinato, Marco Bacis
  • 2.
    Diabetes and DiabeticRetinopathy Metabolic disease with high blood sugar level http://www.patiadiabetes.com/en/ 2 https://www.dreamstime.com/stock-illustration-diabetes-mellitus-info-graphic-complications-detailed-affected-organs-beautiful-colorful-design-image62194003 Uncontrolled diabetes causes health complications 2
  • 3.
    3 Chronically high bloodsugar from diabetes is associated with damage to the tiny blood vessels in the retina, leading to diabetic retinopathy [1] [1] https://nei.nih.gov/health/diabetic/retinopathy Diabetes and Diabetic Retinopathy https://www.dreamstime.com/stock-illustration-diabetes-mellitus-info-graphic-complications-detailed-affected-organs-beautiful-colorful-design-image62194003
  • 4.
    4 Manual segmentationSpecific equipmentQualified staff Screening Test
  • 5.
    5Our Proposal 5 SegmentationAlgorithm Automated the retinal vessel segmentation for diabetic retinopathy screening
  • 6.
    66 Software Implementation HardwareaccelerationSegmentation Algorithm Beye Implementation
  • 7.
    The segmentation wascomputed on the images of Drive and Stare databases, and then compared with the respective manual segmentations 77Methodology and Benchmarking DRIVE database [2] STARE database [3] 20 Images: 605 × 700 pixels, 8 bits per color channel 20 Images: 768 × 584pixels, 8 bits per color channel [2] https://www.isi.uu.nl/Research/Databases/DRIVE/ [3] http://cecas.clemson.edu/~ahoover/stare/ Key Performance Indicators: SW Evaluation HW Evaluation Accuracy Execution Time Energy Consumption
  • 8.
    8Results 8 Accuracy DRIVE database92.93% STARE database 90,3% [4] M. A. Amin and H. Yan, “High speed detection of retinal blood vessels in fundus image using phase congruency,” [5] J. Odstrcilik.“Retinal vessel segmentation by improved matched filtering: evaluation on a new high-resolution fundus image database,”[6] M. Al-Rawi, M. Qutaishat, and M. Arrar, “An improved matched filter for blood vessel detection of digital retinal images,” [9] B. Zhang, L. Zhang, L. Zhang, and F. Karray, “Retinal vessel extraction by matched filter with first-order derivative of gaussian,” [7] M. G. Cinsdikici and D. Aydın, “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony) algorithm,” TP + TN Tot Accuracy = BeyeManual True Positive True Negative False Negative False Positive Pixel comparison between manual segmentation and Beye The software implementation has been compared with other algorithms of the state of the art implemented in software
  • 9.
    99 [4] M. A.Amin et Al. “High speed detection of retinal blood vessels in fundus image using phase congruency,” [5] J. Odstrcilik,et Al. “Retinal vessel segmentation by improved matched filtering: evaluation on a new high- resolution fundus image database,” [6] M. Al-Rawi et Al. “An improved matched filter for blood vessel detection of digital retinal images,” Execution Time [s] in DRIVE Execution Time [s] in STARE ExecutionTime ExecutionTime SpeedUp [7] M. G. Cinsdikici et Al. “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony) algorithm,” [8] X. Jiang et Al.“Adaptive local thresholding by verificationbased multithreshold probing with application to vessel detection in retinal images,” [9] B. Zhang,et Al.“Retinal vessel extraction by matched filter with first-order derivative of gaussian,” Computers in biology and medicine, quello, che è un tempo, e gli altri speedup) SpeedUp Results The software implementation has been compared with other algorithms of the state of the art implemented in software
  • 10.
    9 [4] M. A.Amin et Al. “High speed detection of retinal blood vessels in fundus image using phase congruency,” [5] J. Odstrcilik,et Al. “Retinal vessel segmentation by improved matched filtering: evaluation on a new high- resolution fundus image database,” [6] M. Al-Rawi et Al. “An improved matched filter for blood vessel detection of digital retinal images,” Execution Time [s] in DRIVE Execution Time [s] in STARE ExecutionTime ExecutionTime SpeedUp [7] M. G. Cinsdikici et Al. “Detection of blood vessels in ophthalmoscope images using mf/ant (matched filter/ant colony) algorithm,” [8] X. Jiang et Al.“Adaptive local thresholding by verificationbased multithreshold probing with application to vessel detection in retinal images,” [9] B. Zhang,et Al.“Retinal vessel extraction by matched filter with first-order derivative of gaussian,” Computers in biology and medicine, SpeedUp 55X 137X 260X 514X 162X 47X 29X Speed Up• quello, che è un tempo, e gli altri speedup) Results The software implementation has been compared with other algorithms of the state of the art implemented in software
  • 11.
    • Comparison betweenof our software and hardware performances, in terms of power consumption of the target devices and execution time for a generic image from DRIVE 10 Execution Time: Speed Up 6X Energy consumption : Speed Up 5.7X VS. Intel Core i7-6700 CPU Xilinx Zynq-7000 Results
  • 12.
    • Comparison betweenof our software and hardware performances, in terms of power consumption of the target devices and execution time for a generic image from DRIVE [10] A. Nieto, V. M. Brea, and D. L. Vilari˜no, “Fpga-accelerated retinal vessel-tree extraction,” in Field Programmable Logic and Applications,2009. [11] D. Koukounis, C. Tttofis, and T. Theocharides, “Hardware acceleration of retinal blood vasculature segmentation,” • Comparison between our hardware implementation and state of the art hardware solutions, assessed by accuracy and hardware performance Work A. Nieto et Al.[10] D. Koukounis et Al.[11] Beye Device Spartan 3 Spartan 6 Zedboard Accuracy 0.9100 0.9007 0.9285 Execution time 1.4 s 0.03185 s 0.01041 s Frequency 53 MHz 100 MHz 100 MHz Execution Time: Speed Up 6X Energy consumption : Speed Up 5.7X VS. Intel Core i7-6700 CPU Xilinx Zynq-7000 10Results
  • 13.
    11Conclusion [12] Resnikoff S,Pascolini D, Etya’ale D, et al. Global data on visual impairment in the year 2002. Bulletin of the World Health Organization 2004; 82: 844–51. [13] The Global Economic Cost of Visual Impairment: Summary Report April 2010 It is estimated that Diabetic Retinopathy accounts for 4.8% of the number of cases of blindness worldwide.[12] The financial cost of visual impairment worldwide was US$2,954 billion in 2010[13]
  • 14.
    10Conclusion Thank you foryour attention irene.fidone@mail.polimi.it It is estimated that Diabetic Retinopathy accounts for 4.8% of the number of cases of blindness worldwide.[12] The financial cost of visual impairment worldwide was US$2,954 billion in 2010[13] May 23rd, 2018 Soda 430-438 Woz Lounge @UC Berkeley https://www.facebook.com/beye.project/ https://www.slideshare.net/BEye_project https://twitter.com/BEye_NECSTLab