Mass Classification in Digitized Mammograms Using Texture Features and Artificial Neural Network

The aim of this study was to investigate the performance of texture analysis in texture classification and tissue discrimination between MS lesions, normal appearing white matter (NAWM) and normal white matter (NWM) in order to support early diagnosis of MS. T2-weighted MR images of sixteen relapsing remitting MS (RRMS) patients and sixteen healthy subjects were selected. Based on the lesion size, sixteen regions of interests (ROIs) were chosen from MS patient MR images and healthy subject MR images for MS lesions, NAWM and NWM respectively. Texture features extracted from grey level co-occurrence matrix (GLCM) were selected based on greatest feature difference. For statistical analysis, raw data analysis (RDA), principal component analysis (PCA) and nonlinear discriminant analysis (NDA) were applied to the texture features. The k-nearest neighbor (k-NN) and artificial neural network (ANN) methods were used for texture classification. Fisher coefficient and classification accuracy were used to evaluate the performance of texture analysis. The results demonstrated that (1) classification was successful (>90.00%) between MS lesions and NAWM or NWM, less successful (88.89%) among the three tissue groups and worst (66.67%) between NAWM and NWM; (2) In statistical analysis, NDA outperforms RDA and PCA; (3) ANN classified more accurately than k-NN method between NAWM and NWM, and among the three texture types. This study demonstrated that MRI texture analysis can achieve high classification accuracy in tissue discrimination between MS lesions and NAWM or NWM, which is valuable in supporting early diagnosis of MS.

We investigated the application of multiresolution global and local texture features to reduce false-positive detection in a computerized mass detection program. One hundred and sixty-eight digitized mammograms were randomly and equally divided into training and test groups. From these mammograms, two datasets were formed. The first dataset (manual) contained four regions of interest (ROIs) selected manually from each of the mammograms. One of the four ROIs contained a biopsy-proven mass and the other three contained normal parenchyma, including dense, mixed dense/fatty, and fatty tissues. The second dataset (hybrid) contained the manually extracted mass ROIs, along with normal tissue ROIs extracted by an automated Density-Weighted Contrast Enhancement (DWCE) algorithm as false-positive detections. A wavelet transform was used to decompose an ROI into several scales. Global texture features were derived from the low-pass coefficients in the wavelet transformed images. Local texture features were calculated from the suspicious object and the peripheral subregions. Linear discriminant models using effective features selected from the global, local, or combined feature spaces were established to maximize the separation between masses and normal tissue. Receiver Operating Characteristic (ROC) analysis was conducted to evaluate the classifier performance. The classification accuracy using global features were comparable to that using local features. With both global and local features, the average area, Az, under the test ROC curve, reached 0.92 for the manual dataset and 0.96 for the hybrid dataset, demonstrating statistically significant improvement over those obtained with global or local features alone. The results indicated the effectiveness of the combined global and local features in the classification of masses and normal tissue for false-positive reduction.

In this paper we will present a computer-aided detection (CAD) system for mass detection and classification. This CAD system performs mass detection on regions of interest (ROIs) and performs the normal-abnormal classification on detected masses. To enhance the ROIs, histogram equalization and average filtering were employed. ROIs or suspicious lesions were detected by an edge-base segmentation technique. Each ROI was represented by 12 textural features calculated from spatial gray-level co-occurrence matrix (GLCM). Finally, a back propagation neural network was used to classify a ROI to normal or abnormal one. Performance of the proposed system was analyzed in mini-MIAS data set by means of receiver operating characteristics curve (ROC curve) and free-response ROC curve (FROC curve). We archived area under the ROC curve AZ = 0,815. This value lied in the range from 0.7 to 0.9 so that our CAD system could be considered accurate enough.

In this paper, we propose a Computer-Aided Detection (CAD) system based on the novel Multi-Classifier Fusion-based Classification Model (MCFM) of the breast lesions using textural features in Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI). The proposed system consists of four principal phases: Region Of Interest (ROI) segmentation, feature extraction, fusion and selection and finally Classification. We have extracted a complementary set of features from the DCE-MRI database on two statistical descriptors such as the Gray Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) methods. Then, we have concatenated the extracted features using Serial method to exploit the complementary between these sets. Next, we have applied a features selection phase through Principal Component Analysis (PCA) and Feature Subset Selection (FSS) techniques to remove redundant information and improve the overall performance of breast DCE-MRI in a computer-aided system. For the classification, we have used a set of classifiers (K-Nearest Neighbors (K-NN), Support Vector Machines (SVM) and Multilayer Perceptron-Artificial Neural Network (MLP-ANN)). Then, a fusion Model between them using several metrics is applied. Experimental validation is performed over an MRI dataset, which is comprised of 286 patients with 143 Masses and 143 No Masses (Normal) breast tissues. These experiments report encouraging performances for mass detection, achieved using ten-fold cross-validation model, in terms of the averaged measurements of different metrics such as Sensitivity, Specificity, Area Under Curve (AUC) and Accuracy, which are 0.8972, 0.9572, 0.9630 and 0.9501, respectively. The proposed DCE-MRI analysis model can offer an efficient system for radiologists to identify the breast tissues.

High false-positive recall rate is an important clinical issue that reduces efficacy of screening mammography. Aiming to help improve accuracy of classification between the benign and malignant breast masses and then reduce false-positive recalls, we developed and tested a new computer-aided diagnosis (CAD) scheme for mass classification using a database including 600 verified mass regions. The mass regions were segmented from regions of interest (ROIs) with a fixed size of 512×512 pixels. The mass regions were first segmented by an automated scheme, with manual corrections to the mass boundary performed if there was noticeable segmentation error. We randomly divided the 600 ROIs into 400 ROIs (200 malignant and 200 benign) for training, and 200 ROIs (100 malignant and 100 benign) for testing. We computed and analyzed 124 shape, texture, contrast, and spiculation based features in this study. Combining with previously computed 27 regional and shape based features for each of the ROIs in our database, we built an initial image feature pool. From this pool of 151 features, we extracted 13 features by applying the Sequential Forward Floating Selection algorithm on the ROIs in the training dataset. We then trained a multilayer perceptron model using these 13 features, and applied the trained model to the ROIs in the testing dataset. Receiver operating characteristic (ROC) analysis was used to evaluate classification accuracy. The area under the ROC curve was 0.8814±0.025 for the testing dataset. The results show a higher CAD mass classification performance, which needs to be validated further in a more comprehensive study.

Image retrieval is the process of finding images in the database that are similar to the query image by measuring how close the feature values of the query image are to other images. Image retrieval is currently dominated by approaches that combine several different representations or features. The optimal weight of each feature is needed in combining the image features such as color features, texture features, and shape features. In this study, we use a multi-layer perceptron artificial neural network (MLP) method to obtain feature weights automatically and simultaneously look for optimal weights. The color moment is used to find nine color features, Gray Level Co-occurrence Matrix (GLCM) to find four texture features, and Hu Moment to find seven shape features totaling 20 neurons and all of these features become the input layer in our MLP network. Three neurons in output layers become the automatic weight of each feature. These weights are used to combine each feature's role in obtaining the relevant image. Euclidean Distance is used to measure similarity. The average precision values obtained using automatic feature weights are 93.94% for the synthetic dataset, 91.19% for the Coil-100 dataset, and 54.31% for the Wang dataset. These results have an average difference of 5.06% with the target so automatic feature weighting works well. This value is obtained at a hidden layer size of 11 and a learning rate of 0.1. In addition, the use of automatic feature weighting gives more accurate results compared to manual feature weighting.

Assessment of Neural Network Based Classifiers to Diagnose Focal Liver Lesions Using CT Images

The recovery phase following an earthquake event is essential for urban areas with a significant number of damaged buildings. A lot of changes can take place in such a landscape within the buildings’ footprints, such as total or partial collapses, debris removal and reconstruction. Remote sensing data and methodologies can considerably contribute to site monitoring. The main objective of this paper is the change detection of the building stock in the settlement of Vrissa on Lesvos Island during the recovery phase after the catastrophic earthquake of 12 June 2017, through the analysis and processing of UAV (unmanned aerial vehicle) images and the application of Artificial Neural Networks (ANNs). More specifically, change detection of the settlement’s building stock by applying an ANN on Gray-Level Co-occurrence Matrix (GLCM) texture features of orthophotomaps acquired by UAVs was performed. For the training of the ANN, a number of GLCM texture features were defined as the independent variable, while the existence or not of structural changes in the buildings were defined as the dependent variable, assigning, respectively, the values 1 or 0 (binary classification). The ANN was trained based on the Levenberg–Marquardt algorithm, and its ability to detect changes was evaluated on the basis of the buildings’ condition, as derived from the binary classification. In conclusion, the GLCM texture feature changes in conjunction with the ANN can provide satisfactory results in predicting the structural changes of buildings with an accuracy of almost 92%.

Blood oxygen level dependent magnetic resonance imaging (BOLD-MRI) is a kind of non-invasive MRI technology which reflects the tissue blood oxyen levels. This stuy aims to explore the value of radiomics based on BOLD-MRI in differentiating malignant from benign renal tumors. A total of 141 patients with renal tumors confirmed by pathology were retrospectively analyzed. Seventy-four men and sixty-seven women, aged 26-78 years, with a median age of 56, were included. In all patients, 118 with malignant tumors and 23 with benign tumors were confirmed. All the patients underwent renal T1 weighted imaging (T1WI), T2 weighted imaging (T2WI), and BOLD-MRI scan within 2 weeks before surgery. The patients were randomly assigned into a training group (benign, n=17; malignant, n=83) and a test group (benign, n=6; malignant, n=35). Two radiologists (A and B), who were blind to the pathological results, delineated the regions of interest (ROI) on the maximum axial slices of the tumors. Radiologist B delineated the ROI again at an interval of one month. The intra-class correlation coefficient (ICC) was used to evaluate inter-observer and intra-observer repeatability and ICC>0.75 represented as a good consistency. All the T2* Mapping images and the related ROI files were loaded into the Artificial Intelligence Kit software. A total of 396 texture features, which were calculated based on morphology, histogram, gray level co-occurrence matrix, gray-scale run length matrix, gray-scale area size matrix and gray-scale dependent matrix, were extracted from each ROI. The lowest redundancy and the highest correlation were filtered using minimum redundancy maximum relevance (mRMR) algorithm. Then least absolute shrinkage and selection operator (LASSO) algorithm was used to screened out the most predictive features. Multivariate logistic regression was performed to develop the prediction model after feature selection. The radiomics signature score (Radscore) of each case was calculated. The Wilcoxon test was used to compare the difference in the Radscore between benign and malignant renal tumors in the training and test groups. The diagnostic performance of the model in differentiating malignant from benign renal tumors was evaluated with receiver operating characteristic (ROC) curve and leave group out cross validation. The clinical application value of the model was evaluated by decision curve analysis (DCA). There was significant difference in the age between the patients with benign and those with malignant tumors (t=4.383, P<0.001). There were no significant differences in gender composition and in the largest tumor diameter between the 2 groups (χ2=3.452, P=0.063; t=1.432, P=0.154). The ICC values of all the texture features for the inter-observer repeatability were ranged from 0.71 to 0.87, and the ICC values for the intra-observer repeatability were ranged from 0.76 to 0.91. Thirty features with the lowest redundancy and the highest correlation were screened out. The most predictive 12 features were filtered out. The Radscores of malignant tumors in the training and test groups were higher than those of benign tumors (P<0.001 and P=0.006, respectively). The areas under the ROC curve of the model developed by multivariable logistic regression for differentiating malignant from benign renal tumors in the training and test groups were 0.881 and 0.706, with the accuracy at 82.93% and 79.00%, the sensitivity at 82.86% and 77.11%, and the specificities at 83.33% and 88.24%, respectively. The results of decision curve analysis showed that the net benefit of the radiomics model was higher than that of "all malignant" or "all benign" when the threshold was higher than 0.3. BOLD-MRI-based radiomics can be a reliable non-invasive approach for differentiating renal malignant tumors from benign tumors.

We propose the use of fuzzy membership functions to analyze images of diffuse pulmonary diseases (DPDs) based on fractal and texture features. The features were extracted from preprocessed regions of interest (ROIs) selected from high-resolution computed tomography images. The ROIs represent five different patterns of DPDs and normal lung tissue. A Gaussian mixture model (GMM) was constructed for each feature, with six Gaussians modeling the six patterns. Feature selection was performed and the GMMs of the five significant features were used. From the GMMs, fuzzy membership functions were obtained by a probability-possibility transformation and further statistical analysis was performed. An average classification accuracy of 63.5% was obtained for the six classes. For four of the six classes, the classification accuracy was superior to 65%, and the best classification accuracy was 75.5% for one class. The use of fuzzy membership functions to assist in pattern classification is an alternative to deterministic approaches to explore strategies for medical diagnosis.

Intravascular optical coherence tomography (IVOCT) is becoming an important choice for in vivo diagnosis of coronary artery diseases. The atherosclerotic disease can be detected from IVOCT images, but the segmentation of IVOCT images and identification of plaques are mainly performed manually. This process is laborious and time consuming and its accuracy relies on the expertise of the observer. To address these limitations, a semi-automated identification algorithm based on texture features is presented in this paper. Regions of interest (ROIs) in IVOCT images are firstly selected, then texture features are calculated to represent this ROI. Finally, an artificial neural network is trained to identify the types of atherosclerotic plague: calcium, lipid tissue and fibrous tissue. The effectiveness of the developed algorithm is evaluated by using hundreds of IVOCT images acquired from 46 patients. It is proven that texture can be regarded as typical features of plagues, whose number has a great influence on the accuracy. In addition, identification can be improved by adding a step of screening the neural network before testing samples.

According to the need of automatic weighing and pricing system for fruits in the supermarket and the limitation of current methods, this paper proposes a new segmentation method for fruits with plastic packing and a modified minimum distance classifier. Firstly, we segment images and obtain the region of interest (ROI) in the HSV color space. Secondly, we extract color features and texture features which can be derived from gray level co-occurrence matrix (GLCM), and saving these features into the classifier. Then, we apply modified classifier to recognize and categorize. The experiment results show that our segmentation method can separate ROI from the background with plastic packaging effectively. Furthermore, this novel method can also achieve a better recognition performance with overall recognition rate above 91%.

Texture analysis as an approach to tissue characterization based on the two-dimensional spatial distribution of ultrasound amplitude or grey-level pixels within a region of interest (ROI) is well-established and have been used extensively in Radiology and echocardiography. Objective: We performed a quantitative statistical analysis of various texture parameters of EUS images in patients with histologically confirmed pancreatic adenocarcinoma to evaluate if such statistical techniques can reliably distinguish between malignant and normal pancreatic tissue. Methods: Linear endosonographic images of 22 patients who presented with pancreatic mass and underwent EUS guided FNA for establishing diagnosis of pancreatic adenocarcinoma were analyzed with an image analyzing software (ImageJ, NIH). Multiple ROIs were digitally selected both in the area of the malignant mass and normal pancreatic tissue by an experienced endosonographer. Texture analysis was performed using: histogram (first order statistics) run-length and co-occurrence matrix (measuring features of runs of pixels and distribution of pairs of pixels), gradient analysis (spatial distribution of pixels), auto-regressive analysis (measurement of local interaction amongst pixels of different grey level values) and also wavelet analysis (which measures parameters of spatial frequency). A neural network (NN) based predictive model was built, trained and validated using the extracted texture features for classification of malignant tissue from normal. Methods: A total of 70 ROIs representing malignant tissue and 110 representing normal pancreatic tissue was available for analysis and for each ROI, a total 256 statistical parameters were initially extracted and subsequently 10 best discriminating (based on highest Fisher's co-efficient) features were retained. Overall, 3 types of features had high discriminatory power: Kurtosis (which is a measure of the steepness of the pixel distribution), angular second momentum and horizontal and vertical grey level non-uniformity (parameters estimating the second order joint conditional probability density functions of spatial distribution of two neighboring pixels). Using a random selection of the half of the data set (90 ROIs), a multi-layered perceptron NN model was built, trained using the 10 features as input variables for prediction of malignant tissue and then the model was validated on the rest of the 90 ROIs. The NN model had a very high classification performance with an area under Receiver Operating Characteristic Curve of 0.91 (PPV of 80% and NPV 90%). Conclusion: Digital image analysis of texture features on EUS images is highly accurate in differentiating malignant from normal pancreatic tissue with performance characteristics similar to aspiration cytology.

Breast cancer is becoming the leading cause of cancer deaths among women. The best way to reduce deaths due to breast cancer is early detection and treatment. Dynamic contrast enhanced (DCE) MRI has emerged as a promising new imaging modality for breast cancer screening. Currently, radiologists evaluate breast lesions based on a qualitative description of lesion morphology and contrast uptake profiles. The qualitative description of breast lesions from DCE-MRI introduces a high degree of inter-observer variability. In addition, the high sensitivity of MRI results in good specificity. A computer-assisted evaluation system that can automatically analyse lesion features to differentiate between malignant and benign lesions would be very useful. One of the major characteristics for mass classification is texture. Artificial neural networks exploit this important factor to classify the mass as benign or malignant. The selected texture features were used to classify the mass with a three-layered neural network to predict the outcome of a biopsy. The main objective of this proposed method is to increase the effectiveness, robustness, and efficiency of the classification process in an objective manner to reduce the numbers of false-positive results. The paper presents an intelligent computer assisted mass classification method for breast DCE-MR images. It uses the artificial bee colony algorithm to optimize the a neural network performing benign- malignant classification on the region of interest. A three-layer neural network with seven features was used for classifying the region of interest as benign or malignant. The network was trained and tested using the artificial bee colony algorithm and was found to yield a good diagnostic accuracy.

We investigate the reconstruction of low-count positron emission tomography (PET) projection, which is an important, but challenging, task. Using the texture feature extraction method of radiomics, i.e. the gray-level co-occurrence matrix (GLCM), texture features can be extracted from magnetic resonance imaging images with high-spatial resolution. In this work, we propose a kernel reconstruction method combining autocorrelation texture features derived from the GLCM. The new kernel function includes the correlations of both the intensity and texture features from the prior image. By regarding the GLCM as a discrete approximation of a probability density function, the asymptotically gray-level-invariant autocorrelation texture feature is generated, which can maintain the accuracy of texture features extracted from small image regions by reducing the number of quantized image gray levels. A computer simulation shows that the proposed method can effectively reduce the noise in the reconstructed image compared to the maximum likelihood expectation maximum method and improve the image quality and tumor region accuracy compared to the original kernel method for low-count PET reconstruction. A simulation study on clinical patient images also shows that the proposed method can improve the whole image quality and that the reconstruction of a high-uptake lesion is more accurate than that achieved by the original kernel method.