Diagnosis of Pulmonary Embolism Using Fuzzy Inference System

1
Diagnosis of Pulmonary Embolism Using Fuzzy
Inference System

• Research Advisor Dr. Gursel Serpen
• Research Assistant Vishwanth Acharya

2
Outline of Presentation

• Approach
• Problem Statement
• Previous Research
• Overview of Software Developed
• Working of Software (Diagnosis Sub-Section)
• Working of Software (Image Analysis Sub-Section)
• Graphical User Interface
• Results/Conclusions
• Future Scope

3
Why Artificial Intelligence???

• It can offer a competent second opinion.
• It offers the expertise of an expert radiologist
  in interpreting scans when an expert radiologist
  is not available.
• It has the ability to make accurate and quick
  diagnosis.
• It has the potential to reduce inter-observer
  variability.

4
Pulmonary Embolism (PE)

• Blood clots break off from their source and
  become emboli.
• Emboli travel through the heart into the
  pulmonary arteries.
• They occlude the arteries to various anatomic
  regions of the lung.

300,000 to 600,000 hospitalizations and 50,000
People die each year from PE NIH Consensus
Statement cited August 1999
5
Groups Facing Higher Probability of Pulmonary
Embolism

• Patients Undergoing various types of surgery -
  general, urological, neuro-surgical,
  gynecological.
• Patients with orthopedic problems and chronic
  diseases.

• These groups face a higher probability of
  Pulmonary Embolism due to the high risk of
  developing deep venous thrombosis.

6
Various Diagnostic Criterias

• PIOPED - Prospective Investigation of Pulmonary
  Embolism Diagnosis 1995.
• Biellos Criteria 1979.
• Inputs from Expert Radiologists.

7
PIOPED Criteria

• Low Probability
• Multiple Matching V/Q defects.
• Corresponding V/Q defects and CXR parenchymal
  opacity in upper or middle lung zone.
• Corresponding V/Q defects and large Pleural
  Effusion.
• gt 3 Small SPD.
• Very Low Probability
• lt 3 Small SPD.
• Normal
• No perfusion defects and perfusion outlines the
  shape of the lung seen on CXR

• High Probability
• gt 2 Large segmental perfusion defects (SPD).
• 1 Large SPD and gt 2 Moderate SPD.
• gt 4 Moderate SPD.
• Intermediate Probability
• 1 Moderate to lt 2 Large SPD.
• Corresponding V/Q defect and CXR opacity in lower
  lung.
• Single moderately matched V/Q defect.
• Corresponding V/Q defect and small Pleural
  Effusion.

8
Datzs Work

• Developed an Expert System to diagnose for PE
• Automatically determine the presence of perfusion
  defects, their size and the anatomic segments
  involved.
• Data was evaluated using a set of rules derived
  from the modified Biellos criteria.
• Templates were used for the purpose of extraction
  of information.

9
Datzs Work

• Drawbacks
• The software cannot handle Chest radiograph
  abnormalities.
• Direct comparison of perfusion defects involving
  segments not visible on the posterior perfusion
  image was not possible, and always went by as
  unmatched.
• The image processing technique of stretching the
  image to fit the template led to errors in
  diagnosis.
• To reduce complexity this approach combined two
  classes (Low and Normal) which reduced accuracy.

10
Advantages Over Datzs Work

• Fuzzy Inference System inherently captures and
  models the radiologists thinking better than
  Expert Systems.
• The normalization algorithms applied to the
  images before analyzing them help in reducing the
  problems caused due to soft tissue attenuation.
• The criterion used in the Fuzzy Inference System
  (PIOPED) was the most current version.
• All categories of Diagnosis possible were
  considered.
• Chest radiograph Infiltrates were taken into
  account.

11
Image Processing and FIS
12
Fuzzy Inference System

• The three major components of the Fuzzy Inference
  System are
• Fuzzifier - Converts the input into appropriate
  fuzzy input.
• Inference Engine - Allows the application of the
  rule base to the input parameters whereby
  producing the output.
• Defuzzifier - Converts the output produced by the
  Inference Engine into user understandable terms.

13
Mamdani Model
14
Inputs to Fuzzy System for Diagnosis of
PE(According to PIOPED Criteria)

• Number of Segmental Perfusions.
• Number of Non-Segmental Perfusions.
• Ventilation/Perfusion Mismatch.
• Chest X-Ray Abnormality.
• Presence of Pleural Effusion.

15
Inputs to Fuzzy System to Diagnose for
PE(According to PIOPED Criteria)

• Wt - Weight (pre-calculation of segmental and
  non-segmental perfusion defects.
• Vqdef - Ventilation-Perfusion Defect Mismatch.
• Cxrab - Chest X-ray abnormality.
• Peff - Presence of Pleural Effusion.

16
Input Membership Functions
17
Input Membership Functions
18
Rule Base of Fuzzy System to Diagnose for
PE(Modeling of the PIOPED Criteria)
19
Outputs from Fuzzy System(According to PIOPED
Criteria)

• Output of the Fuzzy System models the
  diagnostic capabilities of the Fuzzy System.
  Hence, the various Membership Functions are
• Normal. (0 to 14)
• Very Low. (14 to 37)
• Low. (37 to 58)
• Intermediate. (58 to 76)
• High (76 to 100)
• The output of the Fuzzy System is a crisp
  value and is mapped to one of these classes
  depending upon its range.

20
Outputs from Fuzzy System(According to PIOPED
Criteria)

• Dia - Diagnosis, is the output of the Fuzzy
  System and is divided into 5 classes.

21
Image Processing

• Image Acquisition Images are procured from the
  hospitals from patients suffering/suspected of
  suffering from Pulmonary Embolism.
• Digitization Image obtained are converted to
  24-bit bitmap to facilitate image analysis.
• Normalization Images are converted to the 8-bit
  gray scale and intensity adjustment algorithms
  are applied.
• Output The image after analysis is sent to the
  Fuzzy Inference System for diagnosis.

22
Perfusion Scans

• An intravenous injection of the radioactive
  isotope Technetium-99m is given.
• Detection of any blockages of blood flow through
  the vessels of the lungs.
• If the pulmonary blood vessels are blocked by
  Pulmonary Emboli (clots), a reduced amount or no
  blood flows into the blocked area.
• If Pulmonary Embolism is present, the Image shows
  a reduced or absent distribution of
  Technetium-99m to the involved area.

23
Ventilation Scans

• A radioactive gas is inhaled. which gets into
  the airways of the lung.
• This shows the distribution of the inhaled
  radioactive gas in the airway.
• By combining the information obtained from both
  scans a more accurate diagnosis of Pulmonary
  Embolism can be made.

24
Image Enhancement

• Intensity adjustment done to raise the average
  pixel intensity in the image to a value between
  65 and 70
• Nonlinear mapping using an S curve used to
  improve the contrast of the image
• Mapped Intensity I(x,y) a m

25
Image Enhancement
Before Normalization After Normalization
26
Problems Faced in Image Enhancement

• The experts knowledge is vague, the facts and
  rules are not totally certain or consistent.
• The presence of noise in the V/Q scan images
  (tubes, needles etc.).
• The change in size of lungs with change in
  patient.

27
Graphical User Interface

• Normalization facilitates identification of the
  segments and lobes of the lung.
• Achieved by converting the images to 8-bit
  grayscale from a 24-bit bitmap.
• Following it up by intensity adjustment.

28
Graphical User Interface

• The sizes of defects are entered.
• Number of mismatches in ventilation/perfusion
  scans noted.
• Areas of chest x-ray and pleural effusion taken
  into consideration.

29
Testing/SimulationTo ensure accuracy and
usability, the software has to pass stringent
tests. These tests were applied in two phases.

• Alpha Testing
• Output data was obtained and passed to
  radiologists to check for accuracy.
• Data developed by radiologists was run through
  the system and checked to ensure that is produced
  expected results.

• Beta Testing
• In this phase the radiologist had a hands on
  experience with the software. This ensured that
  the software has a high degree of usability and
  doctors will not be intimidated by it.

30
Test Data for Prototype
31
(No Transcript)
32
Test Data for Final Version
33
Results at a Glance
34
Results

• Fuzzy Inference System did well at diagnosing for
  Pulmonary Embolism on the basis of the PIOPED
  criteria.
• The Graphical User Interface developed was
  proclaimed as being very user friendly and
  simple to use by the radiologists invited to
  perform real-time testing of the system.
• While the Fuzzy Inference System provides
  excellent results for inputs provided, the inputs
  are dependent on the skill of the human operator.
  
• Total automation of the image analysis stage was
  not achieved.

35
Conclusions

• Implementation of Artificial Intelligence
  software in the diagnosis of medical diseases is
  feasible and can be very easily extended to cover
  different diseases.
• It can be of help to medical practitioners.
• The alternative methods utilized to diagnose for
  Pulmonary Embolism effectively capture the spirit
  of the PIOPED criteria.
• This software has the ability to make accurate
  and quick diagnosis.

36
Future Scope

• Greater Automation in Image Analysis Sub-Section
• Testing with Larger Data Sets and Real Time
  Testing
• Usage of Multiple Approaches with a Committee to
  Decide Probabilities
• Applying Different Fuzzy Inference Models
  (Sugeno)

37
Thank You

• -Vishwanath Acharya