Top Banner
Pulmonary Embolism Sanjay Kumar Kulchania (LECTURER) M.M.INSTITUTE OF MEDICAL & NURSING SCIENCES
21
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
  1. 1. Pulmonary Embolism Sanjay Kumar Kulchania (LECTURER) M.M.INSTITUTE OF MEDICAL & NURSING
  2. 2. Neural Networks In Medical Diagnosis A neural network system: does not suffer from fatigue or psychological factors that can affect the reliability of the diagnosis procedure. once trained, can offer the expertise of an expert radiologist in interpreting the scans when an expert radiologist is not accessible. has the promise for a more accurate diagnosis than is possible with human interpretation.
  3. 3. 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]
  4. 4. Various Diagnostic Criterias Modified PIOPED - Prospective Investigation of Pulmonary Embolism Diagnosis [1995]. Biellos Criteria [1979]. Inputs from Expert Radiologists. The modified PIOPED criteria was followed in this project
  5. 5. Modified PIOPED Criteria High Probability > = 2 Large segmental perfusion defects (SPD). 1 Large SPD and >= 2 Moderate SPD. > = 4 Moderate SPD. Intermediate Probability 1 Moderate to < 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. 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. > 3 Small SPD. Very Low Probability < = 3 Small SPD. Normal No perfusion defects and perfusion outlines the shape of the lung seen on CXR *CXR = Chest Radiograph **V/Q = Ventilation-Perfusion
  6. 6. Architecture of the Neural Diagnosis System Architecture of the Neural Diagnosis System Output Inputs to ANN Image Processing System Artificial Neural Network Committee Machine V/Q Scans and Chest X-Ray Graphical User Interface (GUI)
  7. 7. The ANN Committee Machine Dynamic committee machine 13 MLPs to classify (divided into 5 groups for various probabilitites) 14 RBFNNs as Gating Networks (Part of Integrator) Confidence Integrator (14 RBFNNs) Output Inputs 1 perceptron 1 Perceptron High Probability Intermediate Probability 2 perceptrons MLP-1 2 hidden nodes MLP-2 3 hidden nodes Low Probability 7 perceptrons MLP 2 hidden node Very Low Probability Normal
  8. 8. Inputs to the ANN Committee Machine 1) Size of the largest perfusion defect with respect to the size of the lung. 2) Number of small (< 25% of a segment) segmental perfusion defects with a normal CXR. 3) Number of matched V/Q defects with normal CXR 4) Number of non-segmental perfusion defects 5) Number of perfusion defects surrounded by normally perfused lung 6) Number of corresponding V/Q defects with CXR parenchymal opacity in upper or middle lung zone. 7) Number of corresponding V/Q defects with large pleural effusion. 8) Number of perfusion defects with substantially larger CXR abnormality. 9) Number of moderate matched V/Q defects with normal CXR. 10) Number of corresponding V/Q defects with CXR parenchymal opacity in lower lung zone. 11) Number of corresponding V/Q defects with small pleural effusion. 12) Number of large (>75% of a segment) perfusion defect with normal CXR. 13) Number of moderate (25% - 75% of a segment) perfusion defects without CXR abnormality.
  9. 9. Outputs Classification - Normal Very Low Probability Low Probability Intermediate Probability High Probability Confidence Range 0 to 1
  10. 10. The Integrator Produces confidences in the MLP outputs Confidences depends on distance of input point from decision boundary of the particular MLP (Gaussian Function used) Confidence = |r -1| where, r= RBFNN output Distance from Decision Boundary (x) RBFNN Output (y) 1 0 RBFNN Output v/s Distance from Decision Boundaries
  11. 11. 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 0 255 * a 127 * a 0 1 0 255 * m 200 * a Mapping function (m)Image intensity range (a