Advertisement

A Review on Computer-Aided Modelling and Quantification of PET-CT Images for Accurate Segmentation to Bring Imagination to Life

  • Ziaur Rahiman Shaik
  • Ch. Sumanth Kumar
Conference paper
Part of the Advances in Intelligent Systems and Computing book series (AISC, volume 624)

Abstract

Image segmentation is a process of dividing image into smaller parts to identify the individual objects. Often, this process helps in the quantification of digital images related to disease complications for metabolic process. This work reports on the use of computer-aided modelling tools and rapid prototyping technology to document, preserve and reproduce in three dimensions, and historic machines and mechanisms are used for accurate medical diagnosis. Epidemiological and clinical trials have confirmed the greater incidence and prevalence of deaths due to the inability to acquire qualitative information from the acquisition of images in primary stage itself. Rapid prototyping gives a better understanding of clinical and physiologic mechanisms of various disorders and pain, lesion detection. In image segmentation process, thresholding method is suitable for defining optimal value for identification and detection of region of interest. The standard uptake value (SUV) is based on selecting threshold value to utilize a similarity metric between the grey level of image and data points obtained from the threshold values. This is based on the intensities or inhomogeneity of clustering framework. Affinity propagation is used for images as a matrix by measuring the square patches from similarity texture. A major challenge in computer vision is to extract this information directly from the images available to us, help users, and to see and feel as an actual part in order to bring a computer image to life. Actually, the framework is given by PET-CT images which is used to identify and detect malignant tissues in a human body with accurate measurements of SUVs. This process involves ROI identification, segmentation, rendering and SUV functional quantification for promising results. The results obtained from computer modelling are transformed into real substance by rapid prototyping technology to feel and provide accurate diagnosis to patient.

Keywords

Quantification SUV Epidemiological Rapid prototyping 

1 Introduction

Computed tomography, magnetic resonance imaging tomography, and positron emission tomography are used for clinical practice as diagnostic devices in examining the abnormalities of the human body. These diagnostic devices are also used as a visual aid to support various diseases. These devices will make design iterations faster, produce early feedback, identify flaws, detect inflammations and define early problems to save money and future of patient by identifying the disease in early stages. It provides better surgical results with fewer procedures and also improves the functionalities to a considerable extent [1]. These include skin soft tissues, hard tissues and extra hard bones such as jaws and teeth. The significance of the imaging modalities is that they highlight the affected region and also represent the complete face after chopping the selected portion in three different dimensions such as coronal, sagittal and axial. The realistic models developed by computer modelling can also be made into real life by rapid prototyping technology. Diagnosing in 3D is not much costly affair when compared with other medical imaging modalities. This dynamic CAD model proposed makes the forecasting and treatment of a patient on priority [2]. Physicians could use the results of computer analysis as a second opinion to make the final decision. However, the pathology detection and anatomical structures are used for therapeutic and diagnostic purposes. The structural and functional description of molecular biology is easily represented by PET images with higher sensitivity and specificity. Assessing functional images is possible with uptake values and radiotracers accumulated with abnormalities [3]. Imaging modalities are combined together to accumulate brain, kidney, liver, heart with various diseases to identify and detect earlier diagnosis of diseases. The clinical studies prove that the imaging of PET-CT provides higher resolution and accurate diagnosis at different stages in different fields of medicine such as cardiology, oncology and neurology. The high dose of radiation and targeted volume area minimizes the damage to the tissues and ensures early measures by identifying and detecting the inflammation and infection in the various parts of the human body. In this work, boundary is detected and also it easily identifies the malignant tissues involved with better clinical practice involved by segmentation methods. F-FDG is the deoxyglucose fluorine which is used to identify and localize the cells of the human body affected, and information provided will be of high resolution and predictable [4].

1.1 Use of Quantification in Clinical Practice for Standardized Uptake Value

PET/CT imaging modality uses fludeoxyglucose (F-FDG) as an imaging biomarker in identifying or detecting pathology in oncology, cardiology and neurology. This modality is a view of metabolic information and anatomical structures of various organs of the human body [5]. Therefore, the SUV information provided to this modality is suitable in characterizing the lesions for better therapeutic procedures
$${\text{SUV}} = \frac{{{\text{Activity}}\;{\text{concentration}}\;\left( {\frac{\text{KBq}}{\text{ml}}} \right)}}{{{\text{Injection}}\;{\text{dose}}\;\frac{\text{MBq}}{{{\text{Bodyweight}}\;({\text{kg}})}} }}$$

1.2 Problems in Medical Image Segmentation

One of the major problems with PET/MRI/CT images is recognition, and another is outline boundary detection. These two yield a difficulty in distinguishing the malignant and nonmalignant tissues in the human body. One more aspect is defining the region of interest and uptake value of the region based on the human parts of the body. Usually, any segmentation process involves better resolution, identifying and detecting pathologies, shape, texture, position and external noise. Any algorithm suffers from imaging artefacts, motion artefacts, streaking artefacts, scanning artefacts, etc. The above artifacts make pathology detection difficult to understand and analyse, so in order to overcome these problems, a suitable algorithm is required to carry out all these tasks successful [6]. Hence, SUV measurement and quantitative calculations will overcome this problem to provide a second opinion to the doctor. The resultant image can be used for rapid prototyping of cosmetic surgical applications. Hence, there is a need for accurate and robust technique that can perform all the multimodal images in a real time for better clinical facilities with use of upgraded technology.

1.3 Patient Preparation

Usually for any imaging modalities, patient is made to check his weight for SUV measurement. The system invariability is found out with respect to his fat percentage and body mass. For PET scanning, the acquisition time will be 60 min and an injection is given for a patient with 18F-FDG and repeated twice depending on the requirement of malignancy. Glucose level and blood pressure are monitored with accurate reconstruction parameters. Monitoring glucose level is used to reduce SUV variability. The dose level defines activity concentration of a patient. These factors define the activity recovery concentration and breathing motion artefacts as the major problems in CT scanning, where many elements such as instrumentation and detector system provide the artefacts from various sources defined above. Some of the methods used are attenuation and registration for real-time pathology detection. Some of the related organs and tissues depend on Hounsfield units (HU) for tissues such as heart, liver, lungs and abdomen. The quantitative results are defined by quantification, efficiency and superior resolution. This type of scanning results is used for diagnostic confidence, treatment monitoring and therapy planning which can be seen in real time for lesion detection and accurate quantification [7].

2 Methods

2.1 Segmentation Methods Based on Region of Interest

Homogeneity in the image is the major consideration for detecting the boundaries of the defined image. Usually, segmentation methods use intensities of each region for local distribution. There are two groups that define the region while considering PET images: Two groups are graph cut and region growing, where these methods define the intensities based on histogram. The seed value is given by the user according to the standard deviation and mean. The pixels of image are connected together to define homogeneity matrix. Every pixel is identified from the statistics, and entire region is dissected from the segmented portion to highlight the exact boundary. The ROI is used to work for low-resolution and motion artefact images. Region growing is used to avoid the false positives in segmentation based on user-defined seed value. This algorithm is able to determine the boundaries of lesion over sharp regions with number of iterations. The required target structure can be chopped for the segmented areas depending on the user selection area.

Graph cut methods are used to separate foreground and background seeds based on user-defined value. This automatically locates objects in an image based on pixel similarities for optimum value [8].

2.2 Affinity Propagation

Usually, the image is divided into clusters based on the similarities as data points. These clusters provide better efficiency and low error rate as k-means clustering is used for each data point over the spatial region provided by the affinity function of radio tracer of PET. Usually, the AP divides the spatial region of the image into two data points: availability and responsibility. These points are of scalar values indicating an angle between each other as a(i, k) where k is a responsible point and i is an exemplar point. These message points send the information indicating the angle useful for serving each other. Here, r(i, k) defines the responsibility point and a(i, k) defines the availability point with respect to voxels for easy classifications and similarities between two points.
$$\begin{aligned} & r(i,k) < - s(i,k) < - {\text{Max}}\left\{ {a\left( {i,k^{\prime } } \right) + s\left( {i,k^{\prime } } \right)} \right\} \\ & a(i,k) < - \hbox{min} \left\{ {0,r(k,k)} \right. + \sum\nolimits_{{i^{\prime } (i^{\prime } \notin i,k)}} {\hbox{max} \left\{ {0,r\left( {i^{\prime } ,k} \right)} \right\}} \\ \end{aligned}.$$

2.3 Boundary Detection

Segmentation of image involves boundary detection for low resolution and noise in the images. The source statistics of the PET images define the homogeneities in locating and defining boundaries using either level set or active contours or gradient-based methods. Active contours are used to detect or identify the object of interest based on the energy functions [9]. These energies are usually classified into external and internal energies, where external energy defines the contour or shape based on the texture, edge and gradient values.

2.4 Qualitative and Quantitative Parameters

This work involves oncology, cardiology and neurology as their clinical studies which are to be very accurate for recovery and therapeutic treatment. This diagnosis is done by using PET which is very harmful to the body in extreme cases and results in cancer also. The objective is to find the lesion and malignant tissues in the human body and provide a better analytic procedure for a doctor in providing accurate second opinion. The results obtained are developed in a prototype for measurements and quantification so this step is very important to start the work. Initially, the quantification of three parts involves PET scanning in extreme case for better synthesis. Our method is used to quantify all the details of the patient such as radio tracer life of F18 injection, weight, scanning time and injection dosage [10].

2.5 Rendering and 3D Modelling

Medical imaging uses texture mapping to render the slices of a 3D volume in an aligned position with reasonable quality. This rendering is often known clearly when it is rotated with volume at every transition. Actually, the image is in the form of slices which are aligned based on the viewing angle of the user. These are sliced through the volume at every angle defined and adjusted by the user. Classification determines the attenuation in each voxel, and from the attenuation, the represented tissue is determined. The main four tissues represented in CT are fat, soft tissue, bone and contrast medium-enhanced tissue [11]. 3D modelling is the process of developing a mathematical representation of any three-dimensional surface of object (either inanimate or living) via specialized software. The product is called a 3D model. It can be displayed as a two-dimensional image through a process called 3D rendering or used in a computer simulation of physical phenomena. The model can also be physically created using 3D printing devices (Table 1) [10]. 
Table 1

CT values of abdominal tissues for Hounsfield units

 

CT value (HU)

Organ

15-year-old patient

70-year adult

Tissue

Mean

SD

Heart wall

0.29

0.22

Air

−1006

2

Kidneys

0.089

0.074

Fat

−90

18

Brain

0.072

0.070

Skin

+16

11

Bone surface

0.052

0.041

Spinal canal

+23

15

Liver

0.076

0.058

Kidney

+32

10

Lungs

0.092

0.064

Blood (aorta)

+42

18

Skin

0.037

0.030

Muscle

+44

14

Other tissues

0.052

0.042

3 Design Procedure

Usually, this work involves MATLAB software and rapid prototyping technology with 3D modelling software to bring imagination to life. It uses PET-CT imaging and performs all types of basic steps to identify the malignant tissues and lesion for performing better surgical procedures. In the first step, the image is preprocessed and its region of interest is detected for accurate segmentation. The segmented parts use boundary cut and graph cut segmentation to define the affinity propagation by similarities and level cut formulation. The second step involves the clinical procedures to avoid adverse reactions so the details of imaging are recorded and it is quantified to find out the level of segmentation. The third step shows rendering with volume and surface for the lesion detected based on user-defined viewing angle. Those surfaces are modelled based on the organs of the human body, and the dose injection is based on the Hounsfield units and basic values of PET-CT.

Finally, the obtained results are brought into real life through rapid prototyping technology, and the rendered model is built into a 3D model for better surface to have better and accurate opinion in surgical procedures. The general principle of 3D reconstruction is composed of the following steps.

4 Experimental Result and Analysis

A graphical user interface is designed using MATLAB R2015 with PET-CT images where this GUI is used to perform many operations before developing a 3D model for better accuracy and quantification. The below figure shows a PET-CT image with organs of human heart and abdomen and performs basic steps of preprocessing (Figs. 1, 2 and 3) [9].
Fig. 1

PET-CT image with a MATLAB GUI to perform preprocessing

Fig. 2

Performance of the quantitative parameters of SUV peak and SUV mean, and also, only the ROI is displayed by the user for identifying the lesion

Fig. 3

Rendered GUI with 3D using 3D slicer

5 Conclusion

Throughout the discussion, it is observed that the PET-CT imaging is the best and suitable tool for quantitative functional information on diseases. Image segmentation is basic step for image processing followed with 3D modelling for extracting the information defined by the user. In this work, different segmentation methods are used for defining boundaries and ROI for various organs of the human body. The recent advances applied in these techniques are suitable for PET, PET-CT and MRI-PET images. The investigated results of segmentation methods for smooth and sharp regions are evaluated through quantitative variations of SUV with peak and minimum values. The justified results obtained are given as input to 3D modelling software to render the identified lesions and malignant tissues involved in critical parts of the human body. We investigated different segmentation methods in detail; results were listed and compared throughout this review. These results of 3D modelled software are used to develop the imagination to life through rapid prototyping technology. Thus, this provides the clinicians and doctors a second opinion to perform therapeutic treatment. It is observed the PET image segmentation method is optimal for all applications and can compensate for all of the difficulties inherent to PET images. The changing techniques can also be easily incorporated for anatomical information in metabolic activities for same hybrid frameworks (PET-CT, PET-CT and MRI-PET-CT) which is encouraging, and it is open to further investigations.

References

  1. 1.
    Brent foster, Ulas Bagci, Awais Mansoor, Ziyue Xu, Daniel J Mollura,“A Review on segmentation of Positron Emission Tomography Images”, IEEE trans. Biomedical Engineering, Vol. 1 (3), 76–96, April, 2014.Google Scholar
  2. 2.
    Segmentation of PET images for Computer Aided Functional Quantification of Tuberculosis in Small Animal Models” IEEE trans. Biomedical Engineering, Vol. 61(3), 711–724, Nov, 2013.Google Scholar
  3. 3.
    Partha Ghosh, “Reproducible Quantification in PET-CT: clinical Relevance and Technological Approaches” White paper, SIEMENS, www.Siemens.com.
  4. 4.
    Brent Foster, Ulas Bagci, Ziyue Xu, Bappaditiya Dey, Brian Luna, William Bishai, Sanjay Jain “Segmentation of PET Images for Computer—Aided Functional Quantification of Tuberculosis in Small animal models,” IEEE Trans. Biomedical Engineering, vol. 61, No.3, March 2014.Google Scholar
  5. 5.
    W. Sun, B. Starly, J. Nam, and A. Darling “Bio-CAD modelling its applications in Computer-aided tissue engineering”, pp. 29–47.Google Scholar
  6. 6.
    M. Viceconti, C. Zonnoni, and L. Pierotti. TRI2SOLID: “An application of reverse engineering methods to the creation of cad models of bone segments”. pp. 211–220.Google Scholar
  7. 7.
    R.A. Armistead and J.H. Stanley “Computed Tomography: A Versatile Technology.” Advanced materials & Processes 151 (2). pp. 33–36.Google Scholar
  8. 8.
    Sun W, Starly B, Darling A, Gomez C. “Computer-aided tissue engineering: application to biomimetic modelling and design of tissue scaffolds”, J Biotechnol Appl Biochem 39(1). pp. 49–58., 2004;Google Scholar
  9. 9.
    Ziaur Rahiman Shaik, D. Madhavi, N. Jyothi and Ch. Sumanth Kumar: Development of 3D Model with Iso Surface Reconstruction Algorithm in Cosmetic Surgical Applications, Pak. J. Biotechnology. Vol. 13 (special issue on Innovations in information Embedded and Communication Systems) pp. 193–198 (2016).Google Scholar
  10. 10.
    Ziaur Rahaman Shaik, Ch. Sumanth Kumar “A Survey on Professional CAD software for Digital fabrication of Human Head Phantom using Rapid Prototyping system” Published by International Journal of Engineering Sciences Research-IJESR Vol. 5 Dec 2015.Google Scholar
  11. 11.
    Brent Foster, Ulas Bagci, Brian luna, Bappaditya Dey, William Bishai, Sanjay Jain, Ziyue Xu, Daniel J Mollura,” Robust Segmentation and Accurate Target definition for PET Images using Affinity Propagation, International Symposium on Biomedical imaging, San Fransco, USA,Pg.No:1461-1464, March 2013.Google Scholar

Copyright information

© Springer Nature Singapore Pte Ltd. 2018

Authors and Affiliations

  1. 1.Department of ECEK.L.M College of EngineeringKadapaIndia
  2. 2.Department of ECE, School of TechnologyGITAM UniversityBengaluruIndia

Personalised recommendations