This article investigates the feasibility of isotropic super-resolution reconstruction on 4-D (3-D + time) thoracic MRI data. 4-D MRI sequences generally have high temporal resolution to characterize dynamic phenomena but poor spatial resolution, creating highly anisotropic voxels elongated in the slice-select dimension. Isotropic post-acquisition reconstruction can be obtained using super-resolution algorithms. A new MRI compatible phantom design that simulates lung tumour motion is introduced to evaluate the feasibility and performance of the proposed super-resolution algorithm in the context of 4-D MRI. Several orthogonal low-resolution acquisitions of the phantom are performed through time using a fast true 3-D gradient echo based sequence. The acquired volumes are then registered and combined using a total-variation based regularizer super-resolution algorithm to obtain the high-resolution volume. The quality of the reconstruction is evaluated by measuring the mutual information between the reconstructed volume and a direct isotropic 3-D acquisition. Subjective and objective evaluations show the superiority of our approach compared to the averaging method. This article also discusses the influence of various parameters such as the number of low-resolution scans used and the influence of automatic motion estimation versus known displacement.
Thermography provides an interesting modality for diagnosing breast cancer as it is a non-contact, non-invasive and passive technique that is able to detect small tumors, which in turn can lead to earlier diagnosis. We perform computer-aided diagnosis of breast thermograms based on image features describing bilateral differences in regions of interest and a pattern classification approach that learns from previous examples. As is often the case in medical diagnosis, such training sets are imbalanced as typically (many) more benign cases get recorded compared to malignant cases. In this paper, we address this problem and perform classification using an ensemble of one-class classifiers. One-class classification uses samples from a single distribution to derive a decision boundary, and employing this method on the minority class can significantly boost its recognition rate and hence the sensitivity of our approach. We combine several one-class classifiers using a random subspace approach and a diversity measure to select members of the committee. We show that our proposed technique works well and leads to significantly improved performance compared to a single one-class predictor as well as compared to state-of-the-art classifier ensembles for imbalanced data.
Automated lung cancer detection using computer aided diagnosis (CAD) is an important area in clinical applications. As the manual nodule detection is very time consuming and costly so computerized systems can be helpful for this purpose. In this paper, we propose a computerized system for lung nodule detection in CT scan images. The automated system consists of two stages i.e. lung segmentation and enhancement, feature extraction and classification. The segmentation process will result in separating lung tissue from rest of the image, and only the lung tissues under examination are considered as candidate regions for detecting malignant nodules in lung portion. A feature vector for possible abnormal regions is calculated and regions are classified using neuro fuzzy classifier. It is a fully automatic system that does not require any manual intervention and experimental results show the validity of our system.
Pleural thickenings are caused by asbestos exposure and may evolve into malignant pleural mesothelioma. The detection of pleural thickenings is today done by visual inspection of CT data, which is time-consuming and underlies the subjective judgment. In this work, thickenings are initially detected as the differences between the original contours and the healthy model of the pleura. A subsequent tissue-specific segmentation using the 3D Gibbs-Markov random field (GMRF) within the initially detected region-of-interest separates thickenings from thoracic tissue. Morphometric analysis leads then to 3D modeling and volumetric assessment. Both automatic detection and morphometric modeling of pleural thickenings proposed in this work assure not only reproducible detection but also precise measurement, hence this automated approach can assist physicians to diagnose pleural mesothelioma in its early stage.
Image registration (IR) - the task of aligning different images having a common content - is a fundamental problem in computer vision. In particular, IR is one of the key steps in medical imaging, with applications ranging from computer assisted diagnosis to computer aided therapy and surgery. As IR can be formulated as an optimization problem, a large family of metaheuristics methods can be used to improve the results obtained by classic gradient-based, continuous optimization techniques. In this work, we extend our previous intensity-based image registration (IR) technique based on a real-coded genetic algorithm with a more appropriate design. The performance evaluation of an heterogeneous group of state-of-the-art IR techniques is also extended to two experimental studies on both synthetic and real-word medical IR problems. The results prove the accuracy and applicability of our new method.
In this paper, a novel method for detecting the onset of Alzheimer's disease (AD) from Magnetic Resonance Imaging (MRI) scans is presented. It uses a combination of three different machine learning algorithms in order to get improved results and is based on a three-class classification problem. The three classes for classification considered in this study are normal, very mild AD and mild and moderate AD subjects. The machine learning algorithms used are: the Extreme Learning Machine (ELM) for classification, with its performance optimized by a Particle Swarm Optimization (PSO) and a Genetic algorithm (GA) used for feature selection. A Voxel-Based Morphometry (VBM) approach is used for feature extraction from the MRI images and GA is used to reduce the high dimensional features needed for classification. The GA-ELM-PSO classifier yields an average training accuracy of 94.57 % and a testing accuracy of 87.23 %, averaged across the three classes, over ten random trials. The results clearly indicate that the proposed approach can differentiate between very mild AD and normal cases more accurately, indicating its usefulness in detecting the onset of AD.
Newborn brain MR image segmentation is a crucial procedure for computer-aided diagnosis of brain disorders using MR images. We have previously proposed an automated method for segmenting parenchymal region. The method is based on a fuzzy rule based deformable surface model. In order to improve the segmentation accuracy, this paper introduces a priori knowledge represented by fuzzy object radial model called FORM. The FORM is generated from learning data set, and represents knowledge on shape and MR signal of parenchymal region in MR images. The performance of the proposed method has been validated by using 12 newborn volunteers whose revised age was between -1 month and 1 month. In comparison with the previous method, the proposed method showed the best performance, and the sensitivity was 87.6 % and false-positive-rate (FPR) was 5.68 %. And, leave-one-out cross validation (LOOCV) was conducted to evaluate the robustness. Mean sensitivity and FPR in LOOCV was 86.7 % and 12.1 %.
The automated detection and diagnosis of Diabetic Retinopathy (DR) is very critical to save the patient's vision and to help the ophthalmologists in mass screening of diabetes sufferers. DR is a progressive eye disease and should be detected as early as possible. In this paper, we present a new system for detection and classification of different DR lesions i.e. Microaneurysms (MAs), Haemorrhage (H), Hard Exudates (HE) and Cotton Wool Spots (CWS). We proposed a three stage system in which first stage extracts all possible candidate lesions present in a fundus image suing filter bank. Then feature sets are computed for each candidate lesion using different properties and features followed by classification of lesions. The evaluation of proposed system is performed using retinal image databases with the help of different performance matrices and the results show the validity of proposed system.
Glaucoma is one of the main causes of blindness worldwide. Segmentation of vascular system and optic disc is an important step in the development of an automatic retinal screening system. In this paper we present an unsupervised method for the optic disc segmentation. The main obstruction in the optic disc segmentation process is the presence of blood vessels breaking the continuity of the object. While many other methods have addressed this problem trying to eliminate the vessels, we have incorporated the blood vessel information into our formulation. The blood vessel inside of the optic disc are used to give continuity to the object to segment. Our approach is based on the graph cut technique, where the graph is constructed considering the relationship between neighboring pixels and by the likelihood of them belonging to the foreground and background from prior information. Our method was tested on two public datasets, DIARETDB1 and DRIVE. The performance of our method was measured by calculating the overlapping ratio (Oratio), sensitivity and the mean absolute distance (MAD) with respect to the manually labeled images. Experimental results demonstrate that our method outperforms other methods on these datasets.
We propose in this paper a bayesian network based similarity measure for the retrieving of magnetic resonance imaging exams containing cerebral tumors. Bayesian networks proved their efficiency and reliability in several Artificial Intelligence problems and especially in computer aided decision applications. To diagnose a cerebral tumor in a MRI exam, we need to interpret diverse sequences and to refer to visual characteristics and, also, to the patient clinical information such as age, sex, other diseases, etc. Our main idea is argued by the uncertain aspect embodied of the decision making process. This aspect will be translated as a probabilistic decision model. Our work is tested on several medical cases collected from Sahloul Hospital. The retrieval results seem to be promising.
The problem of automatic classification of brain images obtained by magnetic resonance imaging (MRI) is considered. In order to design the classification system, a three-stage approach is used. It consists of wavelet decomposition of the image under study, feature extraction from the LH and HL subbands using first order statistics, and final classification by support vector machines (SVM). The proposed approach shows higher performance than when using features extracted from the LL sub-band. It is concluded that the horizontal and vertical sub-bands of the wavelet transform can effectively encode the discriminating features of normal and pathological images.
Active contour model has been widely used in image processing applications such as boundary delineation, image segmentation, stereo matching, shape recognition and object tracking. In this paper a novel particle swarm optimization scheme has been introduced to evolve snake over time in a way to reduce time complexity while improving quality of results. Traditional active contour models converge slowly and are prone to local minima due to their complex nature. Various evolutionary techniques including genetic algorithms, particle swarm optimization and predator prey optimization have been successfully employed to tackle this problem. Most of these methods are general problem solvers that, more or less, formulate the snake model equations as a minimization problem and try to optimize it. In contrary, our proposed approach integrates concepts from active contour model into particle swarm optimization so that each particle will represent a snaxel of the active contour. Canonical velocity update equation in particle swarm algorithm is modified to embrace the snake kinematics. This new model makes it possible to have advantages of swarm based searching strategies and active contour principles all together. Aptness of the proposed approach has been examined through several experiments on synthetic and real world images of CT and MRI images of brain and the results demonstrate its promising performance particularly in handling boundary concavities and snake initialization problems.
In the last few decades, image registration (IR) has been established as a very active research area in computer vision. Over the years, it has been applied to a broad range of real-world problems ranging from remote sensing to medical imaging, artificial vision, and computer-aided design. IR has been usually tackled by iterative approaches considering numerical optimization methods which are likely to get stuck in local optima. Recently, a large number of IR methods based on the use of metaheuristics and evolutionary computation paradigms has been proposed providing outstanding results. In this contribution, we aim to develop a preliminary experimental study on some of the most recognized feature-based IR methods considering evolutionary algorithms. To do so, the IR framework is first presented and a brief description of some prominent evolutionary-based IR proposals are reviewed. Finally, a selection of some of the most representative methods are benchmarked facing challenging 3D medical image registration problem instances.
