Thejus Science and Engineering, Adi Shankara Institute of Engineering


Thejus Venugopal

Department of Computer Science and

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

Adi Shankara Institute of
Engineering and Technology, Kalady, Kerala, India


T.  Sobha

Assistant Proffessor,

Department of Computer Science and

 Adi Shankara Institute of Engineering and
Technology, Kalady, Kerala, India



Abstract— This study describes a computer-based approach to classify ten different
kinds of focal and diffused liver disorders using ultrasound images. The
diseased portion is isolated from the ultrasound image by applying automatic
segmentation technique. The segmented region is further decomposed into
horizontal, vertical and diagonal component images by applying bi-orthogonal
wavelet transform. The results are compared with spatial feature extraction
techniques. The proposed technique, which is an application of texture feature
extraction on transform domain images, gives a great classification accuracy
for a combination of ten classes of similar looking diseases which is
appreciable than the spatial domain only techniques for liver disease
classification from ultrasound images.

Keywords—focal liver
disorder;diffused liver disorder;automatic segmentation;wavelet
transform;random forests;spatial domain;

I.      Introduction

Medical Diagnosis has been gaining importance
in everyday life. The diseases and their symptoms are highly varying and there
is always a need for a continuous update of knowledge needed for the doctors. Liver
diseases are generally classified into two categories namely focal and diffused
diseases1. Each disorder has its own characteristic appearance with one
disease resembling another or the same disease presenting itself in visually
varying appearances. Focal
disorders have specific structures and are found  at one or more sites which are caused by
bacteria and other foreign particles. It is usually an enclosed collection of
liquified tissue called pus accompanied by sweling and inflammation. Example:
abscess, cyst . Diffused disorders do not have a specific structures and  are spread throughout the liver. It is caused
due to excessive amount of triglycerides and 
other fats.  Example: fatty

Ultrasound imaging2 is widely
used in modern medicine, as it is economical and non-invasive for final
diagnosis in many cases and primary diagnosis in some. Each disorder has its
own characteristic appearance with one disease resembling another or the same
disease presenting itself in visually varying appearances. Absence of standard
ultrasound machine, individual body conditions etc. influence appearance of


II.    Methodology
















Fig. 1.         

A.    Methodology outline

The proposed method consists of image acquisition
followed by automatic segmentation. Wavelet transform is applied on the
segmented images resulting in three components namely horizontal (H),vertical
(V) and diagonal (D) details for every segmented image which are stored as
component images. GLRLM features4 are extracted from each component image and
classified using random forests by applying ten-fold cross-validation
strategy.The classification results obtained using the proposed method are
analysed and compared with the classification results obtained using GLRLM,
invariant moments and intensity histogram extracted from the original active
contour segmented images in the spatial domain and presented.

B.    Image Acquisition and Automatic Segmentation

Ultrasound image of the diseased liver is obtained
from the patient.
Segregating an image into
distinct regions with each region containing pixels with similar attributes is
known as segmentation. Automatic segmentation3 is a method of evolving a
curve to separate object boundaries from an image automatically.

Wavelet Decomposition

A wavelet is a mathematical function which resembles a small wave like
oscillation which is time limited and helps to capture both temporal and
frequency information from a signal on a multiscale resolution. Continuous
wavelet transform (CWT) is used for analyzing signals and when it is used for
image analysis, CWT produces redundant information.

If signal is analysed using smaller number of scales and different
number of translations at each scale, it gives DWT. The DWT is obtained by
critical sampling of the CWT with the scale parameter a = 2-j and is
defined as







where j and k represent the set of discrete translations
and dilations.

The discrete wavelet transform on an
image consists of one-dimensional row-column filtering the original image
which results in four image components namely (LL),(LH),(HL) and (HH)
detail images.
The LH and HL images capture the
horizontal and vertical information. LL image is a low-resolution
decimated version of the original and HH contains the diagonal
This work analyses the suitability of bi-orthogonal
wavelet family among others, in liver ultrasound image processing for
better classification.