Current Medical Imaging Reviews (v.13, #3)

Meet Our Editorial Board Member by Jose Garcia-Rodriguez (221-221).

An Effective Content Based Medical Image Retrieval by Using ABC based Artificial Neural Network (ANN) by S. Perumal Sankar, N. Vishwanath, Hong Jer Lang, Karthick S. (223-230).
Background: Content Based Image Retrieval (CBIR) has always been a demanding research area as it involves searching of digital images from the collection of images. Difficulty exists in retrieving images for the query posed by concentrating on the factors; computational complexity and accuracy. Though many research works span around CBIR, Content Based Medical Image Retrieval (CBMIR) plays a vital role in the area of medical diagnosis.

Methods: This paper introduces an enhanced feature extraction and retrieval phase for MRI images using Edge based GLCM (EGLCM), and artificial bee colony (ABC) based Artificial Neural Network (ANN) to generate the best results and to yield a higher retrieval accuracy.

Discussion: Our proposed work has four important phases namely Pre-Processing, Feature Extraction, Optimized Retrieval using ABC based ANN. and GLCM. GLCM generates the co-occurrence matrix to calculate Homogeneity, Energy, Correlation, and Contrast as texture features. As a result, eight features act as a feature vector to depict images. Then, to optimize the retrieval task, conjunction with ABC based ANN uses Association rule mining with their significant features in which we carry out Training and Validation phases in 1000 and 100 MRI images.

Conclusion: The performance of the method envisaged in this paper is encouraging, and the method is cost effective as compared to the methods described in the literature, based on the performance metrics namely precision, accuracy and recall. A comparative graph is given to corroborate the same.


Securing Patient Data in Wireless Body Area Sensor Network Using Biometrics Based Key Generation by K. Geetha, S. Chitra, Madhusudhanan B., X. Z. Gao (231-236).
Introduction: A Wireless Body Area Sensor Network (WBASN) implements wearable devices interconnected with each other and connected to a central monitor acting as a Base Station. Ensuring data security and secure communication is challenging due to resource constrained environment. Secure communication is ensured with cryptography which requires good key generation techniques.

Objective: A biometrics-based cryptographic key generation using Electrocardiogram (ECG) is proposed. The focus of this work is to use biometric for securing the data transmission in WBASN and not for biometric-based authentication.

Methods: Walsh-Hadamard transform (WHT) is used for feature extraction. After the features are generated, a stable key generation mechanism generates a stable cryptographic key to secure the data transmission. Based on the distinguishability degree, a transformed feature contributes one or more bits of information during the cryptographic key generation. The generated key has an enlarged space due to which the key is more secure.

Results: Experimental results demonstrate that the proposed biometric authentication system performs satisfactorily with low FAR and FRR. Results reveal that the suggested method achieves improved security with energy savings.

Conclusion: The work includes a generic method for distinguishable feature generation, as well as a stable key generation mechanism ECG signals.


Analysis of Carotid Ultrasound Images for the Assessment of Stroke Risk Using Level Set Method by Sumathi Krishnaswamy, Mahesh Veezhinathan, Jesu Christopher (237-244).
Background: Atherosclerosis is the systemic disease responsible for most of the cardiovascular diseases. Increased Intima-Media Thickness (IMT) of carotid artery is a validated indicator of disease progression and cardiovascular risk.

Methods: In this work an automatic segmentation technique is attempted to improve and preserve the inter-region edges in B-mode longitudinal ultrasound images of Common Carotid Arteries (CCA). The edge information generated using Gaussian filter is used to set the Level set function towards the boundaries of the Intima-Media Complex (IMC). The automated analysis using a variational level set method without re-initialization procedure is used to extract texture and geometric features to analyze pathological conditions more accurately.

Result: Results show that the proposed framework is able to segment IMC and 96.7% correlation with ground truth area. It is also observed that maximum regional overlap obtained using dice similarity with average of 88%, Jaccard index 75% and volume similarity 97%.

Discussion: The texture and ratio-metric features show significant demarcation (p<0.0001) between normal and atherosclerosis subjects. The most significant feature such as autocorrelation shows mean and standard deviation values of 0.821±0.065 in normal and 0.579 ±0.143 in abnormal. Aspect ratio calculated from geometric features is found to have maximum of 7.9 for abnormal and found to decrease with severity of the disease, 12.75 for normal images of CCA. The integration of edge map in the level set framework could extract the boundaries by preserving the edge details and show good correlation with the ground truth values. Further, the group of images investigated for significant features show distinct separation between normal and atherosclerosis subjects.

Conclusion: These findings could be clinically useful in diagnosis and treatment of cardiovascular disease.


Three Dimensional Echocardiography: Recent Trends in Segmen tation Methods by Rani Chacko, Elagiri Ramlingam Rajkumar (245-250).
Background: Echocardiography is a rapidly developing area of research in medical imaging. It is also called an echo test or heart ultrasound which captures moving picture of the heart using sound waves and is a potential alternative to other imaging modalities as far as the advantage is concerned.

Discussion: The new generation echocardiography scanners have the ability to acquire dense image volumes either real time or with few heartbeats. At present, three dimensional echocardiography complements the routine two dimensional echocardiography in clinical practice by providing useful clinical information that has led to recommend its use in defining the left ventricle chamber volume in order to diagnose the various heart disorders. Segmentation of chamber is thus a challenging task, as the ultrasound data consist of speckle noise (acoustic interference) and artefacts which have to be removed.

Conclusion: This paper summarizes the literature on recent research on 3D echocardiography image segmentation methods, focusing on techniques developed for medical diagnostic.


Applications of Ultrasound Elastography in Musculoskeletal Imaging: Technical Aspects and Review of the Literature by Michele Fabio la Torre, Nicola Magarelli, Antonio Cipriani, Santi Rapisarda, Laura Filograna, Claudia Dell'Atti, Antonio Leone, Lorenzo Bonomo (251-259).
Background: Ultrasound (US) has been widely adopted as a user friendly, easily available imaging modality in the assessment of various musculoskeletal pathologies.

Discussion: Recently, a new application of US, US elastography (USE) has emerged as a novel tool to evaluate the elasticity of tissues. This technique has been widely employed in the assessment of pathologies of the thyroid, breast, liver, prostate and pancreas.

Conclusion: The aim of this review is to discuss the different USE techniques currently available and their applications in musculoskeletal imaging.


Image Guided Biopsy of Musculoskeletal Lesions with Low Diagnostic Yield by Yu-Ching Lin, Jim S. Wu, Justin W. Kung (260-267).
Background: Image guided core needle biopsy (CNB) is an important tool in the management of musculoskeletal neoplasms. Although the diagnostic yield and accuracy of this procedure are high, non-diagnostic results can occur. A non-diagnostic CNB result can cause unnecessary patient and physician anxiety and can lead to repeat biopsy and delay in treatment. Knowledge of the radiologic and histologic factors affecting diagnostic yield in CNB of musculoskeletal lesions can assist the radiologist in selecting which lesion to biopsy and help to manage physician and patient expectations of the biopsy results.

Discussion: Small, sclerotic, necrotic, and benign lesions have lower diagnostic yield than large, lytic, non-necrotic and malignant lesions. Before the biopsy is performed, the relevant imaging studies should be reviewed and the best lesion to target should be discussed with the orthopedic oncologist and pathologist. Sampling the least necrotic and least sclerotic portion of the lesion, as well as obtaining sufficient samples, and targeting the walls of cystic lesions can improve diagnostic yield.

Conclusion: Ultimately, despite optimizing CNB methods, non-diagnostic results can still occur but do not need to be considered failures. At times, they can provide evidence that a lesion is benign and helps averting additional interventions.


Purpose: The aim of this article is to provide an updated review of the literatures and give a summary of the evidences supporting the technical mechanism, indications, results and influential factors associated with automated percutaneous lumbar discectomy (APLD) for the treatment of symptomatic contained disc herniation.

Methods: We searched the databases of PubMed and EMBASE for articles written in English and published from January 1980 to December 2014. The search terms that we used included disc herniation, disc protrusion, disc extrusion, disc prolapse, automated percutaneous lumbar discectomy/ diskectomy and mechanical disc decompression. We have analysed all of the systematic reviews, meta-analyses, randomized controlled trials, nonrandomized controlled trials and observational studies in which APLD was involved for the management of contained lumbar disc herniation.

Results: There have been 4 RCTs, one comparable trial and multiple observational studies reported for APLD. APLD were widely used since a roughly 75% success rate was reported initially. But its use declined after several randomized controlled trials' low success rate of this technique. Few studies have been performed in recent years and the high-quality of information available about this technique was poor because of the lack of quality controlled, blinded and randomized trials currently. However, it is difficult to conduct a randomized trial, especially placebo controlled in interventional pain management.

Conclusion: APLD is a safe procedure with minimal complications, and it may provide appropriate choice for the patients with contained disc herniation. For the patients with recurrent disc herniation after conventional surgical procedures, APLD may be an alternative choice.


A Review on Medical Image Registration as an Optimization Problem by Guoli Song, Jianda Han, Yiwen Zhao, Zheng Wang, Huibin Du (274-283).
Objective: In the course of clinical treatment, several medical media are required by a physician in order to provide accurate and complete information about a patient. Medical image registration techniques can provide a richer diagnosis and treatment information to doctors and to provide a comprehensive reference source for the researchers involved in image registration as an optimization problem.

Methods: The essence of image registration is associating two or more different images spatial association, and getting the translation of their spatial relationship. For medical image registration, its process is not absolute. Its core purpose is finding the conversion relationship between different images.

Result: The major step of image registration includes the change of geometrical dimensions, and change of the image of the combination, image similarity measure, iterative optimization and interpolation process.

Conclusion: The contribution of this review is sort of related image registration research methods, can provide a brief reference for researchers about image registration.


Distance Based Genetic Algorithm for Feature Selection in Computer Aided Diagnosis Systems by Sunil Retmin Raj C., Khanna Nehemiah H., Shiloah Elizabeth D., Kannan A. (284-298).
Introduction: One of the most challenging tasks in computer-aided diagnosis (CAD) is mapping of radiological features to image descriptors. If the descriptors chosen are not appropriate to discriminate a disorder from another, the role of CAD system may not be appreciable.

Methods: A better idea could be to extract all the features that are felt to have significance in discrimination and then apply a feature selection algorithm to select only the actually significant features thereby overcoming the problems of high dimensional data. In this paper a distance based genetic algorithm (DGA) has been developed for feature selection. DGA uses fifty percent of the dataset used for training as training set and the remaining fifty percent as validation set in each of the classes. It then applies genetic algorithm (GA) to minimize the objective function, defined by the sum of the squared deviation of each data in the training set of each class from each data in the validation set of the corresponding class. The proposed algorithm has been tested in a CAD system.

Result: The performance of the CAD system that uses the proposed DGA for feature selection has been compared with the system that uses features selected by differential evolution and a statistical repair mechanism (DEFS) and the system that does not use feature selection.

Conclusion: It is found that the accuracy of the system that uses DGA is 88.16% against 83.47% for the system that uses DEFS algorithm and 86.46% for the system that does not use feature selection.


The Simulation and Experimental Comparison of Parameter Factors Impact on MRI Artifact by Siyuan Li, Shan Jiang, Zhiyong Yang, Zhengxing Wu (299-306).
Background: In Magnetic Resonance Imaging, inhomogeneity of the static magnetic field lead to perturbations in the resulting images, called artifacts. Artifact generated by the magnetic metal is a significant problem of medical equipment used in the nuclear magnetic environment. Our goal is to present a method, assuming the field disturbances are known, to construct the resulting images based on the actual principle of MRI by means of simulation, and then, to study the relation between artifact's size and the shape, magnetic susceptibility of material for the shape optimization and material selection of medical equipment at the start of the product development cycle.

Methods: A mathematical model of the MRI process is developed. The way the images are distorted in intensity and shape is explained and an algorithm to simulate magnetic susceptibility artifacts is deduced.

Then, we have studied the impact of length, shape of chamfer, and magnetic susceptibility on the artifact.

Results: The results of simulation have been confirmed to be consistent with the experiment, which means that MRI artifact caused by magnetic metal can be simulated by theoretical simulation. By modifying the parameters in this model, we have found that the artifact gets smaller as the magnetic susceptibility becomes smaller, as the size of chamfer gets greater, as the shape of chamfer gets smoother. The study in this paper has a great significance for the design of medical equipment and can predict MRI artifact of medical equipment before it will be produced.

Conclusion: With the increase of material's permeability, the stronger the sample internal magnetic field will be, and the stronger the external magnetic field disturbance will be; in addition, short samples have weak magnetic conductivity to disturb magnetic field, therefore, as the length or permeability of the cylinder gets greater, the artifact size gets greater. In addition, chamfer size can influence the artifact size as mentioned above, but the effect will get smaller as the distance from the chamfer is greater. The shape of longitudinal section does affect the artifact size, and the relation between them is that the smoother boundary produces smaller artifact; the cause of the phenomenon can be explained that the magnetic charge in the sample surface is not stable, because the repulsion between same magnetic charge is easy to distribute on the boundary of sample and cause the disturbance of magnetic field. In summary, small size, low permeability and smooth boundary can make a contribution to the reduction of the artifact. Moreover, the affecting factors and calculation model of MRI artifact can provide proposed indicators to the design of MRI-compatible surgical robot and medical equipment.


Background: Recently, Magnetic Resonance Imaging (MRI) of brain is used widely in the clinical applications for the detection of abnormalities such as tumor.

Methods: Accurate segmentation of the affected regions in the brain MRI image plays a vital role in the quantitative image analysis to detect the location of tumor in the brain. However, many segmentation algorithms suffer from limited accuracy, due to the presence of noise and intensity inhomogeneity in the brain MR images. This paper proposes novel Textural Pixel Connectivity (TPC) based segmentation technique to predict the location of brain tumor. The Probabilistic Neural Network (PNN) classifier is used to classify the normal and abnormal images. If the image is classified as abnormal, then TPC segmentation process is applied for clustering out the background and tumor spot in the binary segmented output. Then, the growing pattern of tumor is analyzed and represented as a binary image output.

Results & Conclusion: The proposed technique achieves superior performance in terms of sensitivity, specificity, accuracy, error rate, correct rate, inconclusive rate, Positive Predicted Values (PPV), Negative Predicted values (NPV), classified rate, prevalence, positive likelihood and negative likelihood when compared to the traditional Adaboost and Enhanced Adaboost Techniques.


Doppler Ultrasound in the Diagnosis of Portal Vein Stenosis in Adult Living Donor Liver Transplant with Left Lobe Liver Grafts by Hsien-Wen Hsu, Tung-Liang Huang, Tai-Yi Chen, Leo Leung-Chit Tsang, Hsin-You Ou, Chun-Yen Yu, Yu-Fan Cheng, Allan M. Concejero, Chao-Long Chen (326-331).
Objectives: Our aim was to identify ultrasound criteria that may be used to detect portal vein stenosis (PVS) in adult living donor liver transplantation (ALDLT) with left lobe grafts.

Methods: 171 recipients underwent primary left lobe graft aldlt and routine liver doppler ultrasound (DUS) as protocol. Another 151 right lobe liver donors who underwent pre-transplant and 6- month follow-up post-hepatectomy dus were chosen as a control group.

Results: 68.4% (117/171) of the portal vein (PV) anastomoses can be well-visualized and measured by DUS but could not be visualized in 31.6% (54/171).

Discussion: The values of portal anastomotic mean time averaged velocity (TAV) and change in anastomotic/pre-anastomotic portal velocity (Δ TAV) were 51.5 and 31.3 cm/s, respectively, for significant stenosis in patients with visualized PV anastomoses. Whether the PV anastomosis can be visualized or not by dus, the increased umbilical portal vein width indicating possible narrowing of the PV anastomosis was >2 cm. Increased TAV and Δ TAV are useful features to diagnose PVS in ALDLT.

Conclusion: Identifying dilatation of left portal vein umbilical portion helps detecting PVS when the narrowed anastomosis cannot be visualized.


Introduction: The purpose of this study was to obtain normal values for muscle thickness (MT) and echogenicity of orofacial muscles in infants between 6 months and 5 years old. The data may be compared with data of patients with a neuromuscular disorder (NMD) to differentiate between healthy infants and patients and to explain oromotor dysfunction. Aims of this study were to provide normal values for the thickness of the tongue, digastric muscles, temporalis muscles and masseter muscles and the echogenicity of the digastric muscles, masseter muscles, temporalis muscles and geniohyoid muscles, and to establish the influence of gender, age, height and bodyweight on the thickness and echogenicity of the orofacial muscles.

Methods: Ultrasound measurements of orofacial muscles were performed in 56 healthy infants. Echogenicity was determined using a gray-scale analysis. MT was measured by placing electronic calipers at predetermined locations.

Results: MT of the digastric muscles, the temporalis muscles and the tongue depended on bodyweight. Echogenicity of the temporalis muscles depended on bodyweight and age, echogenicity of the digastric muscles depended on age and echogenicity of the masseter muscles depended on bodyweight. Gender only influenced the echogenicity of the geniohyoid muscle. Equations for calculating z-scores for echogenicity and MT were determined.

Conclusion: This study provides normal values for MT and echogenicity of orofacial muscles in infants between 6 months and 5 years old. Difficulty in chewing and swallowing may be explained by comparing the data found in this study with the data of patients.


Performance Analysis of Feature-Based Lung Tumor Detection and Classification by Manoj Senthil Kailasam, Meeradevi Thiagarajan (339-347).
Background: Lung cancer is the leading cause of cancer death and it is identified at the ending stage of the severity. The differentiation between lesions and its background tissues are difficult task due to its low contrast between lesions and its background tissues. Lesion characterization is also a difficult task due to similar texture pattern between the lung tumors and normal lung tissues.

Methods: In this paper, the computer aided automatic detection and classification of lung tumor is proposed. The multi resolution Gabor transform is applied over the lung image and then features such as local derivative and local ternary patterns are extracted from the transformed image. The extracted features are optimized by Genetic Algorithm (GA) and then classified using Adaptive Neuro Fuzzy Inference System (ANFIS) classifier.

Conclusion: The proposed system for lung tumor detection system achieves 98.18% accuracy.


Background: Bone marrow edema-like lesions (BMLs) is a kind of bone disease due to excessive stress in the joints. The abnormal joint stress may be caused by pathological changes of the knee joint. Therefore, assessment of the relationship between the pathological state of the knee joints and BMLs is helpful to understand the pathological process of osteoarthritis (OA).

Objective: To investigate pathogenic factors that are related to BMLs in osteoarthritic knee, by assessing the interrelationship between subregional BMLs and pathology of the interarticular structures.

Methods: MRI and standard radiographs of the knees were obtained in 281 patients with OA. Articular cartilage morphology, meniscal degeneration, ligament integrity, and the presence of a joint effusion were assessed on MRI and were graded by a semi-quantitative method, the Whole Organ MRI Score (WORMS). Univariate and multivariate logistic regression analyses were used to evaluate the associations between BMLs in tibio-femoral compartments of the knee with the presence of pathology in the intra-articular elements of each compartment.

Results: A total of 333 BMLs was found in 207 of 281 patients. The WORMS for cartilage, menisci and osteophytes were significantly higher in OA patients who had BMLs than those without BMLs. In univariable analysis, cartilage defects, meniscal pathology, ligament lesions and osteophates were all correlated with BMLs in various extent. However, when considering all the factors together using multivariate logistic regression analysis, only cartilage defects and meniscus lesions were significantly associated with BMLs in the ipsi-compartment.

Conclusion: The presence of BMLs is associated with multiple factors in knee of OA patients. Cartilage defects appear to be the strongest correlating factor.


Histopathology Image Analysis and Classification Using ARMA Models: Application to Brain Cancer Detection by D. Vaishali, R. Ramesh, C. Gomathy, J. Anita Christaline (355-361).
Objective: To design ARMA model for the analysis of histopathology images.

Background: In traditional cancer diagnosis using clinical pathology, pathologists examine biopsy samples and analyzes it on the basis of cell structure and tissue arrangement. This has unpredictability in ontogenesis and in positioning.

Methods: These methods have been replaced by computer assisted diagnostics (CAD) that support impartial judgment. This work demonstrates the influence of 2D ARMA models in the analysis and classification of histopathology images. The parameter estimation has been done with Yule walker Least Square (LS) method. This work describes brain histopathology image by ARMA parameters which is further analyzed and classified. These features are classified into healthy and malignant tissue samples. The liner kernel and RBF kernel support vector machine (SVM) classifier and Fusion of both classifiers have been used for diagnosis.

Results: Results show that estimated ARMA parameters are excellent discriminating features for statistical study of histopathology images and valuable in cancer diagnostics. Total accuracy improvement is shown by fusing the out puts of linear kernel and RBF kernel classifier with Bayes and Decision template fusion schemes.

Conclusion: This work describes new approach of using ARMA features to extract hidden information in histopathology imagery.