Algorithmic Approach for Prediction a nd Early Detection of Diseases Using Reti nal Images Gurudatha Pai K
Raghu g Rajj P
Prof. Shylaja S S
Dept of Electronics and Comm. Engineering
[email protected]
Dept of Information Science and Engineering
[email protected]
Dept of Information Science and Engineering
[email protected]
PES Institute of Technology, Bangalore, India
http://www.pes.edu
Thursday, April 17, 2008,
Overview
Introduction
Motivation
A Look at Human Eye
Thee Typical yp c Eye ye
The Abnormal Eye
Our Approach
Results
Conclusion
Future Work
Acknowledgements
References
2
Introduction Classical approach to disease diagnosis Invasive Eg: Blood sampling sampling. Non-Invasive Eg: Ultra-Sound, Laser. Ophthalmology Diseases that can be diagnosed using Retina Diabetes Mellitus Hypertension Cataract Certain types of Cancers Symptoms Automation
3
Motivation Unsatisfactory Medical Attention Geographical Barriers Economic Barriers Biological Barriers Unavailability of Experts
4
A Look at Human Eye Typical Eye
5
A Look at Human Eye A Typical Retinal Image
6
A Look at Human Eye The Abnormal Eye
7
Our Approach Patie nt Came ra Image Processing Subsystem
Datab ase
Image Databas e M Manage ment Subsyste
User
Report Generator and Query Facilitator Analyzer
8
Our Approach Camera
9
Our Approach
Image Processing Subsystem Conversion C i off Image I Format F Extraction of Cotton Wool Spots Extraction i off Hemorrhages h Extraction of Veins Bottom Hat Transform Eroding
Extraction of Cholesterol Deposits Cataract Detection
10
Extraction of Cotton Wool Spots and Hemorrha ges CWS occur as a dull white patch in the retina Hemorrhages occur as a red patch in the retina due to the leakage of blood in the retina CWS indicate the possibility of serious organ failure from diabetes. Hemorrhages indicate the hypertension and stress levels Color Slicing Technique
11
Extraction of Cotton Wool Spots and Hemorrha ges R
B
G
12
Extraction of Cotton Wool Spots and Hemorrha ges ⎧ 0 if ⎪ Si = ⎨ ⎪ ri ⎩
[
r j - a j > W / 2 ]⎫ ⎪ j = 1, 2 ,3 ⎬ ⎪ otherwise ⎭ i = 1, 2 ,3
⎧ ⎪0 if Si = ⎨ ⎪⎩ ri
⎡ 3 ( rj - aj ) ^2 > R o ^2 ⎤ ⎫ ⎢⎣ ∑ ⎥⎦ ⎪⎬ j =1 ⎪⎭ otherwise i = 1, 2,3 13
Vein Extraction Focal Arteriolar Narrowing, Arterio-Venious, Nicking, Micro aneurysms are few of the early pathological changes in a patient of Diabetes Mellitus. Bottom Hat Transform. Gradient operation does not yield good results. All other features including macula and iris are highlighted
Eroding. The boundary β (A) an image A can also be determined by first eroding A by a structuring element B and then performing the set difference between A and its erosion.
β (Α) = A − (AΘB)
14
Overview of Mathematical Morphology 1/3 Dilation With A and B as sets in Z2 , then Dilation of A by B
) A⊕B = {z | [(B) z I A ≠ φ ] )
Where (B) is the reflection of B about its origin shifted by z On other words, dilation of A by B is, the set of all displacements, z, s ) uch that (B) and A overlap by at least one element. Or
) A ⊕ B = {z | [( B ) z I A] ⊆ A
15
Overview of Mathematical Morphology 2/3
Erosion With A and B as sets in Z2 , then Dilation of A by yB
) A - B = {z | [(B) z ⊆ A]
)
Where (B) is the reflection of B about its origin shifted by z On other words, erosion of A by B is, the set of all points, z, such tha t B, translated by z is contained in A. Dilation in effect, expands an image where as erosion shrinks it. Though they are complementary to each other, the effect of one can not be restored by the other operation
16
Overview of Mathematical Morphology 3/3
Morphological Opening and Closing of an Image
A o B = ( A − B) ⊕ B A • B = ( A ⊕ B) − B
Morphological Gradient
g = ( A ⊕ B) − ( A − B)
17
Results
18
Results
19
Results
20
Conclusion Today’s fast changing world demands for automated and fairly accurate diagnosis processes. Systematic Image Processing techniques can aid early identification of diseases.
21
Future Work Detect cardiac diseases and cataracts at an early stage. Stratify risks in given patient or appropriate population through analysis of its database. Aid the medical experts in targeting the problem zones to provide appropriate treatment treatment. E.g.: Narrowed retinal arteries are associated with long term risks of hypertension.
Provide multi-database networking and telemedicine provision.
22
Reference [1] Digital Clinical Photography: Choosing a Digital Camera for Clinical Work http://www.pulsemagazine.com.au/index.php?option=com_content&task=view&i d=67 [2] Fogla R, Rao Srinivas K, Ophthalmic Photography using a Digital Camera, Sankara Nethralaya, Medical and Vision Research Foundation, Chennai, http://www.ijo.in/article.asp?issn=03014738 ;year=2003;volume=51;issue=3;spage=269;epage=272;aulast=Fogla [3] Francisco F i Gómez-Ulla, Gó Ull Maria M i I. I Fernandez,Francisco F d F i G Gonzalez, l Pablo P bl R Rey, M Marta t Rodriguez, Maria J. Rodriguez-Cid, Felipe F. Casanueva, Maria A. Tome, Javier Garcia-Tobio, and Francisco Gude, Digital Retinal Images and Tele-Ophthalmology for Detecting and Grading Di b i Retinopathy Diabetic R i h http://care.diabetesjournals.org/ [[4]] Gonzalez and Woods Digital Image Processing, Pearson Education, 2002 23
Reference [5] Michal Softa and Charles V. Stewart, Retinal Vessel Centerline Extraction Using Multiscale Matched Filters, C fid Confidence andd Edge Ed Measures, M IEEE transactions On Medical Imaging Vol. 25, No.12, December 2006, 10.1109/TMI.2006.884190 [6] Tien Yin Wong, Knudtson Michael D, Klein Ronald, Klein Barbara E. K, Meuer Stacy M and Hubbard Larry D. Computer-Assisted Measurement of Retinal Vessel Diameters in the Beaver Dam Eye Study: Methodology, correlation between eyes and effect of refractive errors, http://www.cat.inist.fr [7] Thomas Hugh, Robin Martin, Nicole Bordes, Bernard Pailthorpe Semi-Automatic Semi Automatic Feature Delineation In Medical Images. Images The University of Queensland, Australia.
24
Acknowledgements Dr. Bhujanga Shetty and Staff Narayana Nethralaya, Nethralaya Eye Hospital, Hospital Research Centre Bangalore, India. Dr. K. N. B. S Murthy Principal Prof. S Natarajan Dept. of Information Science and Engineering
http://www.pes.edu
PES Institute of Technology, Bangalore, India.
25
Thank You
