Diapositivo 1

1
Morphologic cirrhosis diagnosis from Ultrasound
Ricardo Ribeiro1,3 Rui Marinho2 and J.M.
Sanches1 1Institute for Systems and Robotics /
Instituto Superior Técnico 2Faculdade de Medicina
da Universidade de Lisboa 3Escola Superior de
Tecnologia da Saúde de Lisboa Lisboa, Portugal

• Abstract
• Cirrhosis is an endemic diseases across the world
  that leads to observed liver contour
  irregularities in the Ultrasound images, which
  can be used to detect and confirm the pathologic
  condition.
• In this work these irregularities are
  semi-automatically segmented and quantified in
  order to help the physician in the diagnosis.
  Results obtained from real data have shown the
  ability of the method to detect the disease.
• The ultimate goal is to use these irregularity
  features jointly with other features extracted
  from the liver parenchyma to design a highly
  discriminative classifier for liver cirrhosis,
  steatosis and other diffuse liver diseases.

• Experimental Results
• Sixty-nine US liver images were obtained from 69
  patients
• 40 patients had cirrhosis (OC)
• 29 had normal liver (ON)
• Image Processing Results

(a) Normal Liver
(b) Cirrhotic liver without ascites

• Problem Formulation
• Estimation of the RF envelope and denoised
  anatomic images
• This is performed using the Bayesian methods
  proposed by 8, where the use of total variation
  techniques allows the preservation of major
  transitions, as seen in the case of liver capsule
  and overlying structures.
• Using the de-noised US image, the liver surface
  contour is obtain using a snake technique which
  computes one iteration of the energy-minimization
  of active contour models.
• Based on the detect contour, the following
  features were extracted
• Root Mean Square of the different angles produced
  by the points that characterize the contour
  (rmsa).
• Root Mean Square of the variation of the points
  of the contour in the y axis (rmsy).
• Mean and Variance of the calculated angles (ma
  and va).
• Variance of the y axis coordinates at each point
  (vy).
• Feature selection - forward selection method with
  the criterion 1 - Nearest Neighbor leave-one-out
  classification performance.
• The features extracted by the preceding methods
  are used for classification, where a support
  vector machine (SVM) classifier 3 is used.

(c) Cirrhotic liver with ascites
Gold Standard Gold Standard
ON OC
Classifier ON 17(24.6) 7(10.1)
Classifier OC 12(17.4) 33(47.9)

• Conclusions
• In this work a semi-automatic detection of liver
  surface is proposed to help in the diagnosis of
  the cirrhosis.
• Results obtained showed an overall accuracy of
  72,46, a high sensitivity, 82,5, and a low
  specificity, 58,6.
• In the future the authors intend to include other
  features to increase the accuracy of the method,
  as well as use more state-of-the-art automatic
  snakes, in order to create a fully automatic
  method.

RecPad2010 - 16th edition of the Portuguese
Conference on Pattern Recognition, UTAD
University, Vila Real city, October 29th