AUTOMATIC SEGMENTATION AND CLASSIFICATION OF LIVER TUMOR FROM CT IMAGES�

GUIDE: Prabakaran R

​

TEAM MEMBERS: Sai Anand K [2012503558]

Jagan C A [2012503513]

Kandasamy S [2012503514]

CONTENTS

• Objective
• Terminology
• Types of Liver Diseases
• Input Images
• Why CT Images?
• Evaluation parameters
• Related Work
• Existing System
• Proposed System
• Conclusion
• References

​

OBJECTIVE

• Segment the liver automatically from CT(computed tomography) images
• Segment and classify tumors from the segmented liver image.
• Find the tumor stage if the tumor is found to be a cancerous one in the second step.

TERMINOLOGY

• Benign
• Not harmful in effect.
• Malignant
• Very virulent or infectious.
• Balloon Force
• A force to control the segmentation near weak boundaries.
• Elastography
• A medical imaging modality that maps the elastic properties of soft tissue.
• Mixture-of mixtures Gaussian model
• Used to remove the noise in CT images.

​

TYPES OF LIVER DISEASES

HEPATITIS

Inflammation of the liver caused by a virus, resulting in liver cell damage and destruction. Five types from Hepatitis A to E.

CIRRHOSIS

Gradual destruction of liver tissue over time. Scarring of the liver, scar tissue slowly replaces healthy functioning liver tissue, diminishing blood flow and destroying the liver's filtering and production functions. Most common cause is alcohol abuse.

​

​

TYPES OF LIVER DISEASES(Contd.)

FIBROSIS

Gradual destruction of liver tissue over time. Similar to cirrhosis, it is the buildup of scar tissue that slowly destroys the liver’s overall function. Most common causes are infection, inflammation, injury or even healing for a related condition.

TUMOR

A typical masses of tissue that form when cells begin to multiply at a rapid rate. The liver can develop benign (noncancerous) and malignant (cancerous) tumors.

​

INPUT IMAGES

​

• Abdominal CT image

without liver tumour

​

​

​

​

• Abdominal CT image

with liver tumour

​

WHY CT IMAGES

• Computed tomography (CT or CAT scan) ranks as one of the top five medical developments in the last 40 years, according to most medical surveys. CT has proven so valuable as a medical diagnostic tool that the 1979 Nobel Prize in Medicine was awarded to the inventors.

​

BENEFITS

• Determining when surgeries are necessary
• Reducing the need for exploratory surgeries
• Improving cancer diagnosis and treatment
• In an emergency room, patients can be scanned quickly so doctors can rapidly assess their condition. Emergency surgery might be necessary to stop internal bleeding.
• The risk of radiation exposure from CT is very small compared to the benefits of a well-planned surgery.
• Rapid acquisition of images
• A wealth of clear and specific information
• A view of a large portion of the body
• No other imaging procedure combines these advantages into a single session.

​

EVALUATION PARAMETERS

Accuracy:

The quality or state of being correct or precise.

Specificity:

The quality or state of being specific.

Sensitivity:

The ability of an organism or part of an organism to react to stimuli, irritability.

Precision:

The state or quality of being precise; exactness.

​

​

RELATED WORK

A MODIFIED DISTANCE REGULARIZED LEVELS SET MODEL FOR LIVER SEGMENTATION FROM CT IMAGES

Nuseiba M. Altarawneh et al[2015] in their work have

• Proposed a modified distance regularized level set model.
• They worked with 10 volumes of abdominal CT images ,each having 64 slices of size 512x512 pixels.
• A novel balloon force was added to the existing distance regularized level set model so that we can guide the direction of the evolving contours via several desired approaches.
• The newly added balloon force discourages the evolving contour from exceeding the liver boundary or leaking at a region that is associated with a weak edge, or does not have an edge
• This model slows down the evolving contour in regions with blurred edges and dampens the evolving contour from exceeding boundaries of liver.
• This method deals with over-segmentation problems efficiently when compared to the DRLS model.

​

• RELATED WORK(Contd.)
• SUPERVISED VARIATIONAL MODEL WITH STATISTICAL INFERENCE AND ITS APPLICATION IN MEDICAL IMAGE SEGMENTATION
• Changyang Li et al[2015] in their work have
• Proposed a supervised variational level set segmentation model to harness the statistical region energy functional with a weighted probability approximation.
• Ten noisy synthetic images and 58 medical datasets with heterogeneous intensities and ill-defined boundaries.
• A weighted probability map on graphs is constructed to incorporate spatial indications from user input with a contextual constraint based on the minimization of contextual graphs energy functional.
• The mixture-of-mixtures Gaussian model is used to better approximate real intensity distributions and distinguish statistical intensity differences between foreground and background.
• The statistical functional can solve the approximation of multimodal intensity distributions for the foreground and background, and the prior probability map distinguishes the regions with marginal differences among mixture-of-mixtures model and it works well on noisy images.
• The drawback of this approach is that prior information regarding seeds is required.

​

​

• RELATED WORK(Contd.)
• STUDYING METHODS OF AUTOMATIC LIVER SEGMENTATION

​

• BUI Den Tien et al[2015] in their work have

​

• Analysed various methods of segmentation of segmentation of liver and proposed the implementation of liver volume calculation in four steps
• Noise reduction by employing anisotropic diffusion filter.
• Liver boundary enhancement by using scale-specific gradient magnitude filter.
• Liver shape approximation by roughly determining structure of liver using fast-marching algorithm.
• Refining liver shape using level set algorithm
• The method is implemented on MRI of 5 patients(3 men, 2 women; age range:57-84 years),scanning parameters- slice thickness of 5mm and reconstruction intervals of 2.5mm, in-plane pixel size range-1.17 to 1.72mm, number of slices in each case ranges from 88 to 120.
• The advantage is that all performance indicators like accuracy , specificity , sensitivity are high and error is low.
• Intensity of liver may be quite similar to intensity of organs surrounding it.
• The edge information in abdominal image is very complicated.

​

​

RELATED WORK(Contd.)

• SEGMENTATION OF CANCEROUS REGIONS IN LIVER USING EDGE-BASED AND PHASE CONGRUENT REGION ENHANCEMENT METHOD
• Gaurav Sethi et al (2015) in their work have

​

• Proposed an edge-based phase congruent region enhancement method with new stopping function for segmenting low contrast cancerous regions from CT images.
• The stages of the proposed method are :
• Separating Region of Interest using phase information of the image.
• Enhancing boundary features of the Region of Interest and creating a new stopping function.
• Edge-based Distance Regularised Level Set Evolution.
• 20 2D CT images and 4 3D CT images are used for experimenting this method
• This proposed method is immune to the location,shape and number of contours to be initialization.
• Manual intervention is required at many levels of the algorithm.

​

• RELATED WORK(Contd.)
• DETECTING SMALL LIVER TUMORS WITH IN-PENTETREOTIDE SPECT—A COLLIMATOR STUDY BASED ON MONTE CARLO SIMULATIONS
• Emma Mähler et al[2012] in their work have
• Proposed a In-pentetreotide single photon emission computed tomography to detect small tumors.
• A set of 20 anonymized patient images from the Nuclear Medicine Department at the University Hospital of Umea was selected.
• The noise is, however, also usually on a high level, and in combination with the low spatial resolution of SPECT, this may lead to difficulties in the detection of small tumors.
• The collimators are of type low-energy general-purpose (LEGP), extended LEGP (ELEGP), and medium-energy general-purpose (MEGP) were used to detect small tumors.
• Contrast as a function of noise and the corresponding CNR as a function of iteration number for the conventional reconstruction technique are used.
• The ELEGP collimator has the highest contrast for a certain noise level for the S12 spheres, which also results in the highest CNRs.

​

​

• RELATED WORK(Contd.)
• TUMOR BURDEN ANALYSIS ON COMPUTED TOMOGRAPHY BY AUTOMATED LIVER AND TUMOR SEGMENTATION
• Marius George Linguraruet al[2012] in their work have
• Proposed the automated computation of hepatic tumor burden from abdominal computed tomography(CT) images of diseased populations with images with inconsistent enhancement.
• Taken 101 CT scans from 68 patients from a variety of public and clinical sources.
• A novel 3-D affine invariant shape parameterization is employed to compare local shape across organs.
• A geodesic active contour corrects locally the segmentations of the livers in abnormal images.
• Liver shape approximation by roughly determining structure of liver using fast-marching algorithm.
• Graph cuts segment the hepatic tumors using shape and enhancement constraints.
• Support vector machines and feature selection are employed to reduce the number of false tumor detections.
• The tumor detection true position fraction of 100% is achieved at 2.3 false positives/case and the tumor burden is estimated with 0.9% error.
• Feature selection resulted in 24% sensitivity increase at 1.6 FP/case.
• Doctors have to help in giving training sets.

• RELATED WORK(Contd.)
• NEW IMAGING TECHNIQUES FOR LIVER DISEASES

Bernard E. Van Beers et al (2015) in their work have

​

• Reviewed various imaging techniques of liver which include:
• Dynamic Contrast Enhanced Ultrasonography which is performed after intravenous injection of ultrasound contrast agents
• Dynamic ultrasound Elastography which is based on assessment of propagation of shear waves within tissues to calculate viscous-elastic properties.
• Diffusion Weighted MR imaging which markedly improves the detection of solid liver tumours relative to T2-weighted fast spin-echo imaging, with lower to comparable accuracy compared with DCE MR imaging
• Dynamic Contrast enhanced MR imaging which is an is an integral part of liver MR imaging for detection and characterisation of liver tumours..
• MR Elastography which uses compression or shear waves which are generated by external vibrators.

​

RELATED WORK(Contd.)

• FULLY AUTOMATIC SEGMENTATION OF LIVER AND HEPTATIC TUMORS FROM 3-D COMPUTED TOMOGRAPHY ABDOMINAL IMAGES: COMPARATIVE EVALUATION OF TWO AUTOMATIC METHOD
• Sergio Casciaro et al[2012] in their work have
• Fully automatic processing frameworks based on graph-cut and gradient flow active contour algorithms.
• Segmentation accuracy is assessed through the following evaluation framework: dice similarity coefficient(DSC), false negative ratio (FNR), false positive ratio (FPR) and processing time.
• Regarding liver surfaces, graph-cuts achieved a DSC of 95.49% (FPR=2.35% and FNR 5.10%), while active contours reached a DSC of 96.17% ( FPR=3.35% and FNR=3.37%)
• Graph cut algorithm is the primary algorithm used in automated segmentation of liver tumor.

RELATED WORK(Contd.)

• A 3-D LIVER SEGMENTATION METHOD WITH PARALLE COMPUTING FOR SELECTIVE INTERNAL RADIATION THERAPY
• Mohammed Goryawala et al[2012] in their work have proposed
• Computing for Selective Internal Radiation Therapy
• 3-D segmentation is based on coupling a modified k-means segmentation method with a special localized contouring algorithm.
• provide fast and accurate liver segmentation and 3-D rendering as well as in delineating tumor region(s), all with minimal user interaction.
• Reduction in processing time.
• Accuracy, less than 2% error.

​

�RELATED WORK(Contd.)��BEST METHOD OF RADIOLOGICAL/CLINICAL MODALITY IS BEST FOR STAGING HEPATIC FIBROSIS

• As per the works of Adrian Huber, Lukas Ebner, Johannes T. Heverhagen, Andreas Christe the MRE appears to be the most effective method for grading liver fibrosis. The following results can be drawn from their study:

​

• Ultrasonography:
• The accuracy, sensitivity and specificity of regular B-mode sonography for diagnosing liver cirrhosis=64–79%, 52–69% and 74–89%.
• An irregular or nodular surface and blunt edges or morphological changes in the liver are the most specific signs of cirrhosis on ultrasound.
• Computer-tomography (CT)-
• The diagnostic accuracy, sensitivity and specificity of CT for hepatic cirrhosis were 67–86%, 77–84% and 53–68%, respectively.
• The diagnostic accuracy of CT scans remains disappointing. One limitation of diagnostic CT is the dose of radiation associated with its use.

​

​

​

​

RELATED WORK(Contd.)�

• Magnetic Resonance Imaging:
• The sensitivity and specificity of classic contrast-enhanced MRI for liver cirrhosis are 87% and 54%, respectively, which are similar to the sensitivity and specificity of CT.
• Double-contrast MRI uses super paramagnetic iron oxide (SPIO) particles that accumulate in hepatic Kupffer cells and lead to a shortening of the relaxation time, producing a dark liver background.
• High-resolution multifrequency MRE(Magnetic Resonance Elastography):
• Multifrequency MRE is preferable for detecting liver fibrosis because mono-frequency MRE produces standing waves with non-undulating wave nodes that do not permit elastographic measurement.
• Spatial resolution is better in Multifrequency MRE.
• MRE has been successfully used to diagnose diastolic dysfunction by identifying abnormalities in myocardial relaxation.

​

​

​

​

EXISTING SYSTEM(PHASE 1)

CT SCAN

PREPROCESSING

SEGMENTATION OF LIVER

CLASSIFICATION

FEATURE EXTRACTION

SEGMENTATION OF TUMOR

NORMAL

ABNORMAL

PROPOSED SYSTEM

• Segmentation of Liver Tumors from CT Scan using Level Set Model.
• Hepatocellular carcinoma is the most frequent ,malignant, primary liver cancer and will be classified using Feature Extraction.
• To identify the specific features which are effective for the classification task.

​

​

​

​

PHASE 2

CT SCAN

PREPROCESSING

SEGMENTATION OF LIVER

CLASSIFICATION

FEATURE EXTRACTION

SEGMENTATION OF TUMOR

2D images

Remove Noise

Segment and save

Portion of liver alone

Detect tumour

Texture,

Shape

NORMAL

ABNORMAL

IDENTIFICATION OF STAGES

LEVEL SET ALGORITHM

• Level Set Method was devised by Osher and Sethian as an improvement over the Fast Marching methods.
• Fast Marching method resulted in over segmentation in regions with broken boundaries.
• Fast Marching Methods are designed for problems in which the speed function never changes sign, so that the front is always moving forward or backward. 
• Level Set Methods are designed for problems in which the speed function can be positive in some places are negative in others, so that the front can move forwards in some places and backwards in others.
• While significantly slower than Fast Marching Methods, embedding the problem in one higher dimension gives the method tremendous generality.
• Level Set used contour based approach wherein the boundary of the expansion was more or less like an elastic band and so the over segmentation problems could be brought down significantly.

WORKING

• In Level Set Method ,
• The Level Set approach instead of following the interface takes the original curve and develops it into a surface.
• Level set method is based on the progressive evaluation of the differences among the neighbouring pixels to find the object boundaries.
• The Level Set uses the zero level set approach for calculating the changes in boundary.
• Level Set uses re-initialization to avoid crossing over of lines when two immiscible moving objects are near the interface.
• Time-Dependent problems uses level set model and these type of problems require initial value formulation.

• The Level set model divides the CT image into a series of slides.
• These slides can be compared with the initial point(zero level set) and the modifications can be stored and can be used for calculations.
• Since the slides vary minimally from each other, the changes can be made note of according to time and this temporal values increase the efficiency of segmentation.
• These values are used by the level set function and the intensity differences are evaluated.
• Sometimes for moving bodies which are near the interface, re-initialization is required.

​

APPROACH

• The level set approach takes the original curve (Red one) and builds it into a surface.
• The cone-shaped surface, which is shown in blue-green has a great property; it intersects the xy plane exactly where the curve sits.
• The blue-green surface on the right below is called the level set function, because it accepts as input any point in the plane and hands back its height as output.
• The red front is called the zero level set, because it is the collection of all points that are at height zero. �

DISTANCE REGULARIZED LEVEL SET MODEL

• Distance Regularized Level Set model is an improvement over Level set model and it uses Edge based Contour .
• Distance Regularized Level Set model uses a Balloon force which is used for calculating the growth function,the level set function.
• The major improvement in this model is the doing away of reinitialization and since a particular point is fixed for calculations, the accuracy can highly be increased.
• This uses a edge contour based segmentation which calculates the growth function and segments the edges.
• Sometimes this method may result in over segmentation at weak edges.

MODIFIED DISTANCE REGULARIZED MODEL

• To avoid over segmentation, extra balloon force has been added which keeps track of the edge distance.
• This balloon force decreases the speed of expansion of the contour when the contour is nearer to the boundary and then increase the speed of expansion when the contour is away from the boundary.
• So the efficiency of segmentation near weak and broken edges is increased.

​

​

​

EXISTING SYSTEM

• The system used currently can perform efficiently only on a certain group of images.
• The efficiency is based on the starting point of the contour.
• It is not generic.

​

PROPOSED ALGORITHM

• The Medical systems in particular need high accuracy.
• So, the main aim of this system is to develop an Application with high accuracy for detection and classification of liver tumors from CT images using Matlab which will serve as a secondary opinion for the doctors.
• The balloon forces can be optimized further to make the application generic and to process a wide range of images efficiently.
• After the detection is done, a multi class classifier is used to classify the type the tumor belongs to.
• This can be further extended to 3D images .

​

ARCHITECTURE DIAGRAM

�CONCLUSION

• The evaluation parameters values have been increased for the segmentation and classification of liver tumors from CT images.

• REFERENCES

​

• Nuseiba M. Altarawneh, Suhuai Luo, Brian Regan, Changming Sun,“A MODIFIED DISTANCE REGULARIZED LEVEL SET MODEL FOR LIVER SEGMENTATION FROM CT IMAGES ,” Signal & Image Processing : An International Journal (SIPIJ) Vol.6, No.1, February 2015.
• Changyang Li*, Member, IEEE, Xiuying Wang, Stefan Eberl, Member, IEEE, Michael Fulham, Yong Yin,and David Dagan Feng, Fellow, IEEE,“Supervised Variational Model With Statistical Inference and Its Application in Medical Image Segmentation,”IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 62, NO. 1, JANUARY 2015.
• Bernard E. Van Beers, Jean Luc Daire,Philippe Garteiser,et al. New imaging techniques of liver disease, Journal of hepatology 2015vol. 62| 690-700
• BUI Dinh Tien et al. Studying methods of automatic liver segmentation from MRI image, Internship Report, University of Bordeaux.
• Adrian Huber, Lukas Ebner, Johannes T. Heverhagen,“State-of-the-art imaging of liver fibrosis and cirrhosis: A comprehensive review of current applications and future perspectives”, Andreas Christe European Journal of Radiology Open 2 (2015) 90–100.
• Emma Mähler, Torbjörn Sundström, Jan Axelsson, and Anne Larsson, Member, IEEE “DETECTING SMALL LIVER TUMORS WITH IN-PENTETREOTIDE SPECT—A COLLIMATOR STUDY BASED ON MONTE CARLO SIMULATIONS ,” IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 59, NO. 1, FEBRUARY 2012 47.

​

​

​

• REFERENCES(Contd..)

​

• Marius George Linguraru*, William J. Richbourg, Jianfei Liu, Jeremy M. Watt, Vivek Pamulapati, Shijun Wang, and Ronald M. Summers,”TUMOR BURDEN ANALYSIS ON COMPUTED TOMOGRAPHY BY AUTOMATED LIVER AND TUMOR SEGMENTATION,” IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 31, NO. 10, OCTOBER 2012.
• Sergio Casciaro, Roberto Franchini, Laurent Massoptier, Ernesto Casciaro, Francesco Conversano, Antonio Malvasi, and Aimè Lay-Ekuakille, “Fully Automatic Segmentations of Liver and Hepatic Tumors From 3-D Computed Tomography Abdominal Images: Comparative Evaluation of Two Automatic Methods”, VOL. 12, NO. 3, MARCH 2012.
• Mohammed Goryawala,Magno R. Guillen, Mercedes Cabrerizo, Armando Barreto, Seza Gulec, Tushar C. Barot, Rekha R. Suthar, Ruchir N. Bhatt, AnthonyMcgoron, andMalek Adjouadi.,V “A 3-D Liver Segmentation Method with Parallel Computing for Selective Internal Radiation Therapy”, OL.16 NO.1 JAN 2012.
• Gaurav Sethi, B.S.Saini, Dilbag Singh, “Segmentation of cancerous regions in liver using an edge-based and phase congruent region enhance method”,Computers and Electrical Engineering (2015).

​

​

​

​

​

​