Ana Claudia Patrocinio

Investigation of clustered microcalcification features for an automated classifier as part of a mammography CAD scheme

Classification of breast microcalcifications and clusters depends characteristics selected to be the input for an automated classifier. Artificial neural networks have been used to aid in classification of structures on mammograms images. However, to achieve the classification, some attributes have to be adequately extracted from the images in the database used for tests. As a part of a CAD scheme in development, our intention is to establish a ANN-based classifier, intended to distribute detected clustered microcalcifications in one of 5 classes, according to BI-RADS classification (normal, benign, probably benign, suspicious and probably malignant). This work reports a part of this procedure, by extracting and selecting most of significant characteristics regarding digitized mammography images containing clustered microcalcifications. Two distinct classes-probably benign and suspicious-are considered in order to compare the selected characteristics incidence distribution. Distance between both classes could be estimated by using Gaussian curves. Images used for the tests were from a database composed by mammograms digitized with 600 dpi of spatial resolution in a andbit grayscale. The regions of interest were selected based on physicians reports on the existence of a cluster. This study has shown that characteristics just as irregularity, number of microcalcifications in a cluster, and cluster area are already enough to separate the processed images in two very distinct classes-suspicious and probably benign, although other features could be necessary for a more detailed classification.

Evaluation of Bayesian network to classify clustered microcalcifications

The purpose of this work is the evaluation and analysis of Bayesian network models in order to classify clusters of microcalcifications to supply a second opinion to the specialists in the detection of breast diseases by mammography. Bayesian networks are statistics techniques, which provide explanation about the inferences and influences among features and classes of a determinated problem. Therefore, the technique investigation will aid in obtaining more detailed information to the diagnosis in a CAD scheme. From regions of interest (ROI), containing clusters of microcalcifications, detailed image analysis, pixel to pixel; in this step shape using geometric descriptors (Hu Invariant Moments, second and third order moments and radius gyration); irregularity measure; compactness; area and perimeter extracted descriptors. By using software of Bayesian network models construction, different Bayesian network classifier models could be generated, using the extracted features mentioned above in order to verify their behavior and probabilistic influences and used as the input to Bayesian network, some tests were performed in order to build the classifier. The results of generated nets models validation correspond to an average of 10 tests made with 6 different database sub-groups. The first results of validation have shown 83.17% of correct results.

Comparing Mammographic Images

The main motivation in using intensity attributes to analyze and compare the technologies of mammography images acquisition is that such images are intrinsically of low contrast, which hinders the lesions detection and interpretation. Considering the viewing conditions, the study is under development to validate the influence of the contrast variation of many types of equipment on the radiologists diagnosis. However, for the image processing, the contrast variation can be considered a problem affecting the detection of important structures. A large-intensity variation in a digitized image can be manipulated by image-processing techniques and aid in lesion detection. However, low-intensity variation, as observed in the images of full-field digital mammography systems, can impede the detection of some lesions types, as masses are structures of low contrast. Therefore, the new technologies should be tested and validated for an adequate calibration according to the procedures adjacent to the image acquisition itself.

Preprocess Enhancement of CT Image for Liver Segmentation with Region Growing Algorithm

Liver cancer, although not being among the most frequent type of cancer in Brazil, but is considered of high complexity to be diagnosed and treated. In order to get a high hit rate of liver segmentation in CT images with the algorithm of region growing are compared several preprocess enhancement techniques, which are: contrast stretching, gamma transformation, laplacian operator, sobel operator. As the result, the better technique is the gamma transformation, reaching 99.99% of correct rate from exam 1, and 78.04% of correct rate from exam 2 (doing comparison with manual segmentation) and using mean squared error rate on same technique, was observed 0.33 for exam 1 and for exam 2 was of 12.05. Doing the comparison of this techniques for enhancement of CT images can be observed which one is better for do the enhancement of liver CT images before the use of segmentation technique, and for slices which radiological contrast reached a great performance.

Breast density pattern characterization by histogram features and texture descriptors

Introduction Breast cancer is the first leading cause of death for women in Brazil as well as in most countries in the world. Due to the relation between the breast density and the risk of breast cancer, in medical practice, the breast density classification is merely visual and dependent on professional experience, making this task very subjective. The purpose of this paper is to investigate image features based on histograms and Haralick texture descriptors so as to separate mammographic images into categories of breast density using an Artificial Neural Network. Methods We used 307 mammographic images from the INbreast digital database, extracting histogram features and texture descriptors of all mammograms and selecting them with the K-means technique. Then, these groups of selected features were used as inputs of an Artificial Neural Network to classify the images automatically into the four categories reported by radiologists. Results An average accuracy of 92.9% was obtained in a few tests using only some of the Haralick texture descriptors. Also, the accuracy rate increased to 98.95% when texture descriptors were mixed with some features based on a histogram. Conclusion Texture descriptors have proven to be better than gray levels features at differentiating the breast densities in mammographic images. From this paper, it was possible to automate the feature selection and the classification with acceptable error rates since the extraction of the features is suitable to the characteristics of the images involving the problem.

Feature extraction using convolutional neural network for classifying breast density in mammographic images

Breast cancer is the leading cause of death for women in most countries. The high levels of mortality relate mostly to late diagnosis and to the direct proportionally relationship between breast density and breast cancer development. Therefore, the correct assessment of breast density is important to provide better screening for higher risk patients. However, in modern digital mammography the discrimination among breast densities is highly complex due to increased contrast and visual information for all densities. Thus, a computational system for classifying breast density might be a useful tool for aiding medical staff. Several machine-learning algorithms are already capable of classifying small number of classes with good accuracy. However, machinelearning algorithms main constraint relates to the set of features extracted and used for classification. Although well-known feature extraction techniques might provide a good set of features, it is a complex task to select an initial set during design of a classifier. Thus, we propose feature extraction using a Convolutional Neural Network (CNN) for classifying breast density by a usual machine-learning classifier. We used 307 mammographic images downsampled to 260x200 pixels to train a CNN and extract features from a deep layer. After training, the activation of 8 neurons from a deep fully connected layer are extracted and used as features. Then, these features are feedforward to a single hidden layer neural network that is cross-validated using 10-folds to classify among four classes of breast density. The global accuracy of this method is 98.4%, presenting only 1.6% of misclassification. However, the small set of samples and memory constraints required the reuse of data in both CNN and MLP-NN, therefore overfitting might have influenced the results even though we cross-validated the network. Thus, although we presented a promising method for extracting features and classifying breast density, a greater database is still required for evaluating the results.

Classification of nodules in mammograms image by using wavelet transform

This work presents a classifier for mammographic masses using the wavelet transform as characteristics generator. It considers the BI-RADS classification, dividing mass according to their shapes: circulate, nodular and speculate. We developed procedures with two steps: the first involves a model applying one wavelet technique performing the contours analysis with simulated mass images. This procedure was used to choose the best wavelet that could generate the desired characteristics. The second procedure had the objective of applying the chosen wavelet to masses from segmented images. Both methods have as answers three classes of shape. A root-mean-square function is applied to obtain the energy measure for each level of wavelet decomposition. Thus the shape feature vectors are formed with the coefficients of the details and coefficients of approximation extracted by the energy of wavelet decomposition levels. Linear Discriminan Analysis (LDA) by using Fischer Discriminant was used to reduce the number of characteristics for the feature vector. The Mahalanobis distance was used by the classifier to verify aimed the pertinence of the images for each one the previously given classes. To test actual images, the leave-one-out method was used to the classifier training. The classifier has registered good results, compared to others reports in the corresponding literature.

Novel Mahalanobis-based feature selection improves one-class classification of early hepatocellular carcinoma

Detection of early hepatocellular carcinoma (HCC) is responsible for increasing survival rates in up to 40%. One-class classifiers can be used for modeling early HCC in multidetector computed tomography (MDCT), but demand the specific knowledge pertaining to the set of features that best describes the target class. Although the literature outlines several features for characterizing liver lesions, it is unclear which is most relevant for describing early HCC. In this paper, we introduce an unconstrained GA feature selection algorithm based on a multi-objective Mahalanobis fitness function to improve the classification performance for early HCC. We compared our approach to a constrained Mahalanobis function and two other unconstrained functions using Welch’s t-test and Gaussian Data Descriptors. The performance of each fitness function was evaluated by cross-validating a one-class SVM. The results show that the proposed multi-objective Mahalanobis fitness function is capable of significantly reducing data dimensionality (96.4%) and improving one-class classification of early HCC (0.84 AUC). Furthermore, the results provide strong evidence that intensity features extracted at the arterial to portal and arterial to equilibrium phases are important for classifying early HCC.

Three-dimensional reconstruction and surface extraction of lower limbs as visualization methodologies of ecchymosis

This paper presents a computational system for three-dimensional reconstruction and surface extraction of the human lower limb as a new methodology of visualizing images of multifaceted ecchymosis on the lower limbs. Through standardization of image acquisition by a mechanical system, an algorithm was developed for three-dimensional and surface reconstruction based on the extraction of depth from silhouettes. In order to validate this work, a three-dimensional model of the human lower limb was used inside a virtual environment. At this environment the mechanical procedure of image acquisition was simulated, resulting in 100 images which was later submitted to all algorithms developed. It was observed that the systems for three-dimensional reconstruction and surface extraction of the object were able to generate a new visualization method of the lesion. The results allow us to conclude that the developed systems provided adequate three-dimensional and two-dimensional visualization of the surface of the simulated model. Despite the lack of experiments with real ecchymoses, the systems developed in this work show great potential to be included in the standard methods for the visualization of ecchymoses.

Avaliação de desempenhos de esquema de diagnóstico auxiliado por computador (CAD) para diferentes grupos de imagens mamográficas

Este trabalho descreveu os testes de um esquema de diagnostico auxiliado por computador (CAD) com dois diferentes grupos de imagens mamograficas e comparou com o desempenho das respostas dos especialistas. Foram utilizadas imagens com comprovacoes patologicas com regioes de interesse (RIs) com nodulos benignos e malignos. O grupo 1 de imagens foi composto por 102 RIs apenas com nodulos malignos, e o grupo 2 por 50 RIs, contendo nodulos benignos e malignos. As imagens do grupo 1 passaram por dupla leitura de especialistas e suas respostas foram comparadas com as do CAD. O CAD apresentou area sob a curva ROC (AZ ) de 0,94 e 0,84 para os grupos 1 e 2 respectivamente. Enquanto os especialistas apresentaram AZ de 0,85 para o grupo1.