Computer-aided Diagnosis Models Research Articles

Heart Rate Variability Classification using Support Vector Machine and Genetic Algorithm

Electrocardiogram (ECG) is defined as an electrical signal, which represents cardiac activity. Heart rate variability (HRV) as the variation of interval between two consecutive heartbeats represents the balance between the sympathetic and parasympathetic branches of the autonomic nervous system. In this study, we aimed to evaluate the efficiency of discrete wavelet transform (DWT) based features extracted from HRV which were further selected by genetic algorithm (GA), and were deployed by support vector machine to HRV classification. In this paper, 53 ECGs including 3 different beat types (ventricular fibrillation (VF), atrial fibrillation (AF) and also normal sinus rhythm (NSR)), were selected from the MIT/BIH arrhythmia database. The approach contains 4 stages including HRV signal extraction from each ECG signal, feature extraction using DWT (entropy, mean, variance, kurtosis and spectral component β), best features selection by GA and classification of normal and abnormal ECGs using the selected features by support vector machine (SVM). The performance of the classification procedure employing the combination of selected features were evaluated using several measures including accuracy, sensitivity, specificity and precision which resulted in 97.14%, 97.54%, 96.9% and 97.64%, respectively. A comparative analysis with the related existing methods illustrates the proposed method has a higher potential in the classification of AF and VF. The attempt to classify the ECG signal has been successfully achieved. The proposed method has shown a promising sensitivity of 97.54% which indicates that this technique is an excellent model for computer-aided diagnosis of cardiac arrhythmias.

Mammogram classification using contourlet features with forest optimization-based feature selection approach

Breast cancer continues to be one of the major health issues across the world and it is mostly observed in females. However, the actual cause of this cancer is still an ongoing research topic. Hence, early detection and diagnosis of breast cancer are considered to be an effective and reliable solution. Mammography is one of the most efficacious medical tools for early detection of breast cancer. The radiologists identify the suspicious regions in the breast by carefully examining the mammograms. However, mammograms are sometimes difficult to analyze when the breast tissues are dense. Therefore, a computer-aided diagnosis (CAD) system is adopted which can improve the decisions of the radiologists. This paper proposes a hybrid CAD framework to classify the suspicious regions into either normal or abnormal, and further, benign or malignant. The proposed framework constitutes four computational modules, namely, ROI generation using cropping operation, texture feature extraction using contourlet transformation, a wrapper-based feature selection algorithm, namely, forest optimization algorithm to select the optimal features, and finally different classifiers like SVM, k-NN, Naive Bayes, and C4.5 that are employed to classify the inputs into normal or abnormal, and again benign or malignant. The proposed framework is examined on two widely used standard datasets, namely, MIAS and DDSM. The performance measures are computed with respect to normal vs. abnormal, and benign vs. malignant for four different hybrid CAD models, namely, (Contourlet + FOA + SVM), (Contourlet + FOA + k-NN), (Contourlet + FOA + Naive Bayes), and (Contourlet + FOA + C4.5). The highest classification accuracy of 100% is achieved for normal vs. abnormal classification in case of both MIAS and DDSM. The performance of the proposed hybrid scheme demonstrates its effectiveness with the other state-of-the-art schemes. Experimental results reveal that the proposed hybrid scheme is accurate and robust. Finally, the suggested scheme is considered as a reliable CAD framework to help the physicians for better diagnosis.

Detecting Premature Ventricular Contraction in Children with Deep Learning

Premature ventricular contractions (PVCs) are abnormal heart beats that indicate potential heart diseases. Diagnosis of PVCs is made by physicians examining long recordings of electrocardiogram (ECG), which is onerous and time-consuming. In this study, deep learning was applied to develop models that can detect PVCs in children automatically. This computer-aided diagnosis model achieved high accuracy while sustained stable performance. It could save time and repeated efforts for physicians, enabling them to focus on more complicated tasks.This study is a first step toward children’s PVC auto-detection in clinics. Further study will improve the model’s performance with optimized structure and more data in different sources, while facing the challenges of the variety and uncertainty of children’s ECG with heart diseases.

Computer aided diagnosis model for lung tumor based on ensemble convolutional neural network

The convolutional neural network (CNN) could be used on computer-aided diagnosis of lung tumor with positron emission tomography (PET)/computed tomography (CT), which can provide accurate quantitative analysis to compensate for visual inertia and defects in gray-scale sensitivity, and help doctors diagnose accurately. Firstly, parameter migration method is used to build three CNNs (CT-CNN, PET-CNN, and PET/CT-CNN) for lung tumor recognition in CT, PET, and PET/CT image, respectively. Then, we aimed at CT-CNN to obtain the appropriate model parameters for CNN training through analysis the influence of model parameters such as epochs, batchsize and image scale on recognition rate and training time. Finally, three single CNNs are used to construct ensemble CNN, and then lung tumor PET/CT recognition was completed through relative majority vote method and the performance between ensemble CNN and single CNN was compared. The experiment results show that the ensemble CNN is better than single CNN on computer-aided diagnosis of lung tumor.

Use of pattern recognition and neural networks for non-metric sex diagnosis from lateral shape of calvarium: an innovative model for computer-aided diagnosis in forensic and physical anthropology.

Sex determination on skeletal remains is one of the most important diagnosis in forensic cases and in demographic studies on ancient populations. Our purpose is to realize an automatic operator-independent method to determine the sex from the bone shape and to test an intelligent, automatic pattern recognition system in an anthropological domain. Our multiple-classifier system is based exclusively on the morphological variants of a curve that represents the sagittal profile of the calvarium, modeled via artificial neural networks, and yields an accuracy higher than 80 %. The application of this system to other bone profiles is expected to further improve the sensibility of the methodology.

Multispectral medical image fusion in Contourlet domain for computer based diagnosis of Alzheimer's disease.

Computer based diagnosis of Alzheimer's disease can be performed by dint of the analysis of the functional and structural changes in the brain. Multispectral image fusion deliberates upon fusion of the complementary information while discarding the surplus information to achieve a solitary image which encloses both spatial and spectral details. This paper presents a Non-Sub-sampled Contourlet Transform (NSCT) based multispectral image fusion model for computer-aided diagnosis of Alzheimer's disease. The proposed fusion methodology involves color transformation of the input multispectral image. The multispectral image in YIQ color space is decomposed using NSCT followed by dimensionality reduction using modified Principal Component Analysis algorithm on the low frequency coefficients. Further, the high frequency coefficients are enhanced using non-linear enhancement function. Two different fusion rules are then applied to the low-pass and high-pass sub-bands: Phase congruency is applied to low frequency coefficients and a combination of directive contrast and normalized Shannon entropy is applied to high frequency coefficients. The superiority of the fusion response is depicted by the comparisons made with the other state-of-the-art fusion approaches (in terms of various fusion metrics).

Analysis of Machine Learning Techniques Applied to the Classification of Masses and Microcalcification Clusters in Breast Cancer Computer-Aided Detection

Breast cancer is one of the most common and deadliest types of cancer among women and early detection is of major importance to decrease mortality rates. Microcalcification clusters and masses are two major indicators of malignancy in the early stages of this disease, when mammography is typically used as the screening technology. Computer-Aided Diagnosis (CAD) systems can support the radiologists’ work, by performing a double-reading process, which provides a second opinion that the physician can take into account in the detection process. This paper presents a CAD model based on computer vision procedures for locating suspicious regions that are later analyzed by artificial neural networks, support vector machines and linear discriminant analysis, to classify them into benign or malignant, based on a set of features that are extracted from lesions to characterize their visual content. A genetic algorithm is used to find the subset of features that provide the greatest discriminant power. Our results show that the SVM presented the highest overall accuracy and specificity for classifying microcalcification clusters, while the NN outperformed the rest for mass-classification in the same parameters. Overall accuracy, sensitivity and specificity were measured.

Noninvasive functional imaging of P-glycoprotein-mediated doxorubicin resistance in a mouse model of hereditary breast cancer to predict response, and assign P-gp inhibitor sensitivity

Using a "spontaneous" mammary mouse tumor model we set out to develop diagnostic approaches for non-invasive P-glycoprotein (P-gp) staging and response prediction. (99m)Tc-MIBI efflux rates were measured using a gamma camera in three Brca1 (-/-); p53 (-/-) mouse mammary tumors that have different Mdr1a/b expression levels. The efflux rates were quantified in the 10-30-min period after injection. In addition to the P-gp-mediated efflux measurements in untreated tumors, efflux measurements were performed in the presence of the P-gp inhibitor tariquidar. Volumetric doxorubicin response patterns for the different tumors were determined and correlated with the efflux rates. Combined pre- and post-inhibitor treatment imaging of P-gp-mediated efflux correlated with Mdr1a/b expression: basal (0.0026, p = 0.16), 3-fold Mdr1a/b (0.0074, p = 0.02), and 17-fold Mdr1a and 46-fold Mdr1b (0.012, p = 0.002). Based on the doxorubicin response of these tumors, we generated a computer-aided diagnosis model that predicts the likelihood of drug resistance. Quantified (99m)Tc-MIBI efflux has potential to: (1) noninvasively assign Mdr1 expression levels, (2) predict the therapeutic impact of a P-gp inhibitor, and (3) noninvasively assess the probability of drug resistance.

Breast Mass Lesions: Computer-aided Diagnosis Models with Mammographic and Sonographic Descriptors

To retrospectively develop and evaluate computer-aided diagnosis (CAD) models that include both mammographic and sonographic descriptors. Institutional review board approval was obtained for this HIPAA-compliant study. A waiver of informed consent was obtained. Mammographic and sonographic examinations were performed in 737 patients (age range, 17-87 years), which yielded 803 breast mass lesions (296 malignant, 507 benign). Radiologist-interpreted features from mammograms and sonograms were used as input features for linear discriminant analysis (LDA) and artificial neural network (ANN) models to differentiate benign from malignant lesions. An LDA with all the features was compared with an LDA with only stepwise-selected features. Classification performances were quantified by using receiver operating characteristic (ROC) analysis and were evaluated in a train, validate, and retest scheme. On the retest set, both LDAs were compared with radiologist assessment score of malignancy. Both the LDA and ANN achieved high classification performance with cross validation (area under the ROC curve [A(z)] = 0.92 +/- 0.01 [standard deviation] and (0.90)A(z) = 0.54 +/- 0.08 for LDA, A(z) = 0.92 +/- 0.01 and (0.90)A(z) = 0.55 +/- 0.08 for ANN). Results of both models generalized well to the retest set, with no significant performance differences between the validate and retest sets (P > .1). On the retest set, there were no significant performance differences between LDA with all features and LDA with only the stepwise-selected features (P > .3) and between either LDA and radiologist assessment score (P > .2). Results showed that combining mammographic and sonographic descriptors in a CAD model can result in high classification and generalization performance. On the retest set, LDA performance matched radiologist classification performance.

Application of the mutual information criterion for feature selection in computer-aided diagnosis.

The purpose of this study was to investigate an information theoretic approach to feature selection for computer-aided diagnosis (CAD). The approach is based on the mutual information (MI) concept. MI measures the general dependence of random variables without making any assumptions about the nature of their underlying relationships. Consequently, MI can potentially offer some advantages over feature selection techniques that focus only on the linear relationships of variables. This study was based on a database of statistical texture features extracted from perfusion lung scans. The ultimate goal was to select the optimal subset of features for the computer-aided diagnosis of acute pulmonary embolism (PE). Initially, the study addressed issues regarding the approximation of MI in a limited dataset as it is often the case in CAD applications. The MI selected features were compared to those features selected using stepwise linear discriminant analysis and genetic algorithms for the same PE database. Linear and nonlinear decision models were implemented to merge the selected features into a final diagnosis. Results showed that the MI is an effective feature selection criterion for nonlinear CAD models overcoming some of the well-known limitations and computational complexities of other popular feature selection techniques in the field.