Pilot study of machine learning in the task of distinguishing high and low-grade pediatric hydronephrosis on ultrasound

PURPOSE

Hydronephrosis is a common pediatric urological condition, characterized by dilation of the renal collecting system. Accurate identification of the severity of hydronephrosis is crucial in clinical management, as high-grade hydronephrosis can cause significant damage to the kidney. In this pilot study, we demonstrate the feasibility of machine learning in differentiating between high and low-grade hydronephrosis in pediatric patients.

MATERIALS AND METHODS

We retrospectively reviewed 592 images from 90 unique patients ages 0-8 years diagnosed with hydronephrosis at the University of Chicago's Pediatric Urology Clinic. The study included 74 high-grade hydronephrosis (145 images) and 227 low-grade hydronephrosis (447 images). Patients were excluded if they had less than 2 studies prior to surgical intervention or had structural abnormalities. We developed a radiomic-based artificial intelligence algorithm incorporating computerized texture analysis and machine learning (support-vector machine) to yield a predictor of hydronephrosis grade.

RESULTS

Receiver operating characteristic analysis of the classifier output yielded an area under the curve value of 0.86 (95% CI 0.81-0.92) in the task of distinguishing between low and high-grade hydronephrosis using a five-fold cross-validation by kidney. In addition, a Mann-Kendall trend test between computer output and clinical hydronephrosis grade yielded a statistically significant upward trend (p<0.001).

CONCLUSIONS

Our findings demonstrate the potential of machine learning in the differentiation between low and high-grade hydronephrosis. Further studies are warranted to validate our findings and their generalizability for use in clinical practice as a means to predict clinical outcomes and the resolution of hydronephrosis.

A machine-learning algorithm for distinguishing malignant from benign indeterminate thyroid nodules using ultrasound radiomic features

Background: Ultrasound (US)-guided fine needle aspiration (FNA) cytology is the gold standard for the evaluation of thyroid nodules. However, up to 30% of FNA results are indeterminate, requiring further testing. In this study, we present a machine-learning analysis of indeterminate thyroid nodules on ultrasound with the aim to improve cancer diagnosis. Methods: Ultrasound images were collected from two institutions and labeled according to their FNA (F) and surgical pathology (S) diagnoses [malignant (M), benign (B), and indeterminate (I)]. Subgroup breakdown (FS) included: 90 BB, 83 IB, 70 MM, and 59 IM thyroid nodules. Margins of thyroid nodules were manually annotated, and computerized radiomic texture analysis was conducted within tumor contours. Initial investigation was conducted using five-fold cross-validation paradigm with a two-class Bayesian artificial neural networks classifier, including stepwise feature selection. Testing was conducted on an independent set and compared with a commercial molecular testing platform. Performance was evaluated using receiver operating characteristic analysis in the task of distinguishing between malignant and benign nodules. Results: About 1052 ultrasound images from 302 thyroid nodules were used for radiomic feature extraction and analysis. On the training/validation set comprising 263 nodules, five-fold cross-validation yielded area under curves (AUCs) of 0.75 [Standard Error (SE) = 0.04; P < 0.001 ] and 0.67 (SE = 0.05; P = 0.0012 ) for the classification tasks of MM versus BB, and IM versus IB, respectively. On an independent test set of 19 IM/IB cases, the algorithm for distinguishing indeterminate nodules yielded an AUC value of 0.88 (SE = 0.09; P < 0.001 ), which was higher than the AUC of a commercially available molecular testing platform (AUC = 0.81, SE = 0.11; P < 0.005 ). Conclusion: Machine learning of computer-extracted texture features on gray-scale ultrasound images showed promising results classifying indeterminate thyroid nodules according to their surgical pathology.

A comparative study of pre-trained convolutional neural networks for semantic segmentation of breast tumors in ultrasound

The automatic segmentation of breast tumors in ultrasound (BUS) has recently been addressed using convolutional neural networks (CNN). These CNN-based approaches generally modify a previously proposed CNN architecture or they design a new architecture using CNN ensembles. Although these methods have reported satisfactory results, the trained CNN architectures are often unavailable for reproducibility purposes. Moreover, these methods commonly learn from small BUS datasets with particular properties, which limits generalization in new cases. This paper evaluates four public CNN-based semantic segmentation models that were developed by the computer vision community, as follows: (1) Fully Convolutional Network (FCN) with AlexNet network, (2) U-Net network, (3) SegNet using VGG16 and VGG19 networks, and (4) DeepLabV3+ using ResNet18, ResNet50, MobileNet-V2, and Xception networks. By transfer learning, these CNNs are fine-tuned to segment BUS images in normal and tumoral pixels. The goal is to select a potential CNN-based segmentation model to be further used in computer-aided diagnosis (CAD) systems. The main significance of this study is the comparison of eight well-established CNN architectures using a more extensive BUS dataset than those used by approaches that are currently found in the literature. More than 3000 BUS images acquired from seven US machine models are used for training and validation. The F1-score (F1s) and the Intersection over Union (IoU) quantify the segmentation performance. The segmentation models based on SegNet and DeepLabV3+ obtain the best results with F1s>0.90 and IoU>0.81. In the case of U-Net, the segmentation performance is F1s=0.89 and IoU=0.80, whereas FCN-AlexNet attains the lowest results with F1s=0.84 and IoU=0.73. In particular, ResNet18 obtains F1s=0.905 and IoU=0.827 and requires less training time among SegNet and DeepLabV3+ networks. Hence, ResNet18 is a potential candidate for implementing fully automated end-to-end CAD systems. The CNN models generated in this study are available to researchers at https://github.com/wgomezf/CNN-BUS-segment, which attempts to impact the fair comparison with other CNN-based segmentation approaches for BUS images.

Methods Used in Computer-Aided Diagnosis for Breast Cancer Detection Using Mammograms: A Review

According to the American Cancer Society's forecasts for 2019, there will be about 268,600 new cases in the United States with invasive breast cancer in women, about 62,930 new noninvasive cases, and about 41,760 death cases from breast cancer. As a result, there is a high demand for breast imaging specialists as indicated in a recent report for the Institute of Medicine and National Research Council. One way to meet this demand is through developing Computer-Aided Diagnosis (CAD) systems for breast cancer detection and diagnosis using mammograms. This study aims to review recent advancements and developments in CAD systems for breast cancer detection and diagnosis using mammograms and to give an overview of the methods used in its steps starting from preprocessing and enhancement step and ending in classification step. The current level of performance for the CAD systems is encouraging but not enough to make CAD systems standalone detection and diagnose clinical systems. Unless the performance of CAD systems enhanced dramatically from its current level by enhancing the existing methods, exploiting new promising methods in pattern recognition like data augmentation in deep learning and exploiting the advances in computational power of computers, CAD systems will continue to be a second opinion clinical procedure.

Fuzzy entropy based on differential evolution for breast gland segmentation

For the diagnosis and treatment of breast tumors, the automatic detection of glands is a crucial step. The true segmentation of the gland is directly related to effective treatment effect of the patient. Therefore, it is necessary to propose an automatic segmentation algorithm based on mammary gland features. A segmentation method of differential evolution (DE) fuzzy entropy based on mammary gland is proposed in the paper. According to the image fuzzy entropy, the evaluation function of image segmentation is constructed in the first step. Then, the method adopts DE, the image fuzzy entropy parameter is regard as the initial population of individual. After the mutation, crossover and selection of three evolutionary processes to search for the maximum fuzzy entropy of parameters, the optimal threshold of the segmented gland is achieved. Finally, the mammary gland is segmented by the threshold method of maximum fuzzy entropy. Eight breast images with four tissue types are tested 100 times, with accuracy (Acc), sensitivity (Sen), specificity (Spe), positive predictive value (PPV), negative predicted value (NPV), and average structural similarity (Mssim) to measure the segmentation result. The Acc of the proposed algorithm is 98.46 ± 8.02E-03%, 95.93 ± 2.38E-02%, 93.88 ± 6.59E-02%, 94.73 ± 1.82E-01%, 96.19 ± 1.15E-02%, and 97.51 ± 1.36E-02%, 96.64 ± 6.35E-02%, and 94.76 ± 6.21E-02%, respectively. The mean Mssim values of the 100 tests were 0.985, 0.933, 0.924, 0.907, 0.984, 0.928, 0.938, and 0.941, respectively. Our proposed algorithm is more effective and robust in comparison to the other fuzzy entropy based on swarm intelligent optimization algorithms. The experimental results show that the proposed algorithm has higher accuracy in the segmentation of mammary glands, and may serve as a gold standard in the analysis of treatment of breast tumors.

Detection of mass regions in mammograms by bilateral analysis adapted to breast density using similarity indexes and convolutional neural networks

BACKGROUND AND OBJECTIVE

The processing of medical image is an important tool to assist in minimizing the degree of uncertainty of the specialist, while providing specialists with an additional source of detect and diagnosis information. Breast cancer is the most common type of cancer that affects the female population around the world. It is also the most deadly type of cancer among women. It is the second most common type of cancer among all others. The most common examination to diagnose breast cancer early is mammography. In the last decades, computational techniques have been developed with the purpose of automatically detecting structures that maybe associated with tumors in mammography examination. This work presents a computational methodology to automatically detection of mass regions in mammography by using a convolutional neural network.

METHODS

The materials used in this work is the DDSM database. The method proposed consists of two phases: training phase and test phase. The training phase has 2 main steps: (1) create a model to classify breast tissue into dense and non-dense (2) create a model to classify regions of breast into mass and non-mass. The test phase has 7 step: (1) preprocessing; (2) registration; (3) segmentation; (4) first reduction of false positives; (5) preprocessing of regions segmented; (6) density tissue classification (7) second reduction of false positives where regions will be classified into mass and non-mass.

RESULTS

The proposed method achieved 95.6% of accuracy in classify non-dense breasts tissue and 97,72% accuracy in classify dense breasts. To detect regions of mass in non-dense breast, the method achieved a sensitivity value of 91.5%, and specificity value of 90.7%, with 91% accuracy. To detect regions in dense breasts, our method achieved 90.4% of sensitivity and 96.4% of specificity, with accuracy of 94.8%.

CONCLUSIONS

According to the results achieved by CNN, we demonstrate the feasibility of using convolutional neural networks on medical image processing techniques for classification of breast tissue and mass detection.

Classification of breast masses in ultrasound images using self-adaptive differential evolution extreme learning machine and rough set feature selection

A method using rough set feature selection and extreme learning machine (ELM) whose learning strategy and hidden node parameters are optimized by self-adaptive differential evolution (SaDE) algorithm for classification of breast masses is investigated. A pathologically proven database of 140 breast ultrasound images, including 80 benign and 60 malignant, is used for this study. A fast nonlocal means algorithm is applied for speckle noise removal, and multiresolution analysis of undecimated discrete wavelet transform is used for accurate segmentation of breast lesions. A total of 34 features, including 29 textural and five morphological, are applied to a [Formula: see text]-fold cross-validation scheme, in which more relevant features are selected by quick-reduct algorithm, and the breast masses are discriminated into benign or malignant using SaDE-ELM classifier. The diagnosis accuracy of the system is assessed using parameters, such as accuracy (Ac), sensitivity (Se), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV), Matthew's correlation coefficient (MCC), and area ([Formula: see text]) under receiver operating characteristics curve. The performance of the proposed system is also compared with other classifiers, such as support vector machine and ELM. The results indicated that the proposed SaDE algorithm has superior performance with [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] compared to other classifiers.

Application of wavelet techniques for cancer diagnosis using ultrasound images: A Review

Ultrasound is an important and low cost imaging modality used to study the internal organs of human body and blood flow through blood vessels. It uses high frequency sound waves to acquire images of internal organs. It is used to screen normal, benign and malignant tissues of various organs. Healthy and malignant tissues generate different echoes for ultrasound. Hence, it provides useful information about the potential tumor tissues that can be analyzed for diagnostic purposes before therapeutic procedures. Ultrasound images are affected with speckle noise due to an air gap between the transducer probe and the body. The challenge is to design and develop robust image preprocessing, segmentation and feature extraction algorithms to locate the tumor region and to extract subtle information from isolated tumor region for diagnosis. This information can be revealed using a scale space technique such as the Discrete Wavelet Transform (DWT). It decomposes an image into images at different scales using low pass and high pass filters. These filters help to identify the detail or sudden changes in intensity in the image. These changes are reflected in the wavelet coefficients. Various texture, statistical and image based features can be extracted from these coefficients. The extracted features are subjected to statistical analysis to identify the significant features to discriminate normal and malignant ultrasound images using supervised classifiers. This paper presents a review of wavelet techniques used for preprocessing, segmentation and feature extraction of breast, thyroid, ovarian and prostate cancer using ultrasound images.

In the digital era, architectural distortion remains a challenging radiological task

AIM

To compare readers' performance in detecting architectural distortion (AD) compared with other breast cancer types using digital mammography.

MATERIALS AND METHODS

Forty-one experienced breast screen readers (20 US and 21 Australian) were asked to read a single test set of 30 digitally acquired mammographic cases. Twenty cases had abnormal findings (10 with AD, 10 non-AD) and 10 cases were normal. Each reader was asked to locate and rate any abnormalities. Lesion and case-based performance was assessed. For each collection of readers (US; Australian; combined), jackknife free-response receiver operating characteristic (JAFROC), figure of merit (FOM), and inferred receiver operating characteristic (ROC), area under curve (Az) were calculated using JAFROC v.4.1 software. Readers' sensitivity, location sensitivity, JAFROC, FOM, ROC, Az scores were compared between cases groups using Wilcoxon's signed ranked test statistics.

RESULTS

For lesion-based analysis, significantly lower location sensitivity (p=0.001) was shown on AD cases compared with non-AD cases for all reader collections. The case-based analysis demonstrated significantly lower ROC Az values (p=0.02) for the first collection of readers, and lower sensitivity for the second collection of readers (p=0.04) and all-readers collection (p=0.008), for AD compared with non-AD cases.

CONCLUSIONS

The current work demonstrates that AD remains a challenging task for readers, even in the digital era.

Validation of a computer-aided diagnosis system for the automatic identification of carotid atherosclerosis

Carotid atherosclerosis represents one of the most important causes of brain stroke. The degree of carotid stenosis is, up to now, considered one of the most important features for determining the risk of brain stroke. Ultrasound (US) is a non-invasive, relatively inexpensive, portable technique, which has an excellent temporal resolution. Computer-aided diagnosis (CAD) has become one of the major research fields in medical and diagnostic imaging. We studied US images of 44 patients, 22 patients with and 22 without carotid artery stenosis, by using US examination and applying a CAD system, an automatic prototype software to detect carotid plaques. We obtained 287 regions: 60 were classified as plaques, with an average signal echogenicity of 244.1 ± 20.0 and 227 were classified as non-plaques, with an average signal echogenicity of 193.8 ± 38.6 compared with the opinion of an expert neurologist (golden test). The receiver operating characteristic (ROC) analysis revealed a highly significant area under the ROC curve difference from 0.5 (null hypothesis) in the discrimination between plaques and non-plaques; the diagnostic accuracy was 89% (95% CI: 0.85-0.92), with an appropriate cut-off value of 236.8, sensitivity was 83% and specificity reached a value of 85%. The experimental results showed that the proposed method is feasible and has a good agreement with the expert neurologist. Without the need of any user-interaction, this method generates a detection out-put that may be useful in second opinion.

A Hybrid Image Filtering Method for Computer-Aided Detection of Microcalcification Clusters in Mammograms

The presence of microcalcification clusters (MCs) in mammogram is a major indicator of breast cancer. Detection of an MC is one of the key issues for breast cancer control. In this paper, we present a highly accurate method based on a morphological image processing and wavelet transform technique to detect the MCs in mammograms. The microcalcifications are firstly enhanced by using multistructure elements morphological processing. Then, the candidates of microcalcifications are refined by a multilevel wavelet reconstruction approach. Finally, MCs are detected based on their distributions feature. Experiments are performed on 138 clinical mammograms. The proposed method is capable of detecting 92.9% of true microcalcification clusters with an average of 0.08 false microcalcification clusters detected per image.

Analysis of co-occurrence texture statistics as a function of gray-level quantization for classifying breast ultrasound

In this paper, we investigated the behavior of 22 co-occurrence statistics combined to six gray-scale quantization levels to classify breast lesions on ultrasound (BUS) images. The database of 436 BUS images used in this investigation was formed by 217 carcinoma and 219 benign lesions images. The region delimited by a minimum bounding rectangle around the lesion was employed to calculate the gray-level co-occurrence matrix (GLCM). Next, 22 co-occurrence statistics were computed regarding six quantization levels (8, 16, 32, 64, 128, and 256), four orientations (0° , 45° , 90° , and 135°), and ten distances (1, 2,...,10 pixels). Also, to reduce feature space dimensionality, texture descriptors of the same distance were averaged over all orientations, which is a common practice in the literature. Thereafter, the feature space was ranked using mutual information technique with minimal-redundancy-maximal-relevance (mRMR) criterion. Fisher linear discriminant analysis (FLDA) was applied to assess the discrimination power of texture features, by adding the first m-ranked features to the classification procedure iteratively until all of them were considered. The area under ROC curve (AUC) was used as figure of merit to measure the performance of the classifier. It was observed that averaging texture descriptors of a same distance impacts negatively the classification performance, since the best AUC of 0.81 was achieved with 32 gray levels and 109 features. On the other hand, regarding the single texture features (i.e., without averaging procedure), the quantization level does not impact the discrimination power, since AUC = 0.87 was obtained for the six quantization levels. Moreover, the number of features was reduced (between 17 and 24 features). The texture descriptors that contributed notably to distinguish breast lesions were contrast and correlation computed from GLCMs with orientation of 90° and distance more than five pixels.

COMPARISON OF SUPPORT VECTOR MACHINES AND BAYESIAN NEURAL NETWORKS PERFORMANCE FOR BREAST TISSUES USING GEOSTATISTICAL FUNCTIONS IN MAMMOGRAPHIC IMAGES

Female breast cancer is a major cause of deaths in occidental countries. Computer-aided Detection (CAD) systems can aid radiologists to increase diagnostic accuracy. In this work, we present a comparison between two classifiers applied to the separation of normal and abnormal breast tissues from mammograms. The purpose of the comparison is to select the best prediction technique to be part of a CAD system. Each region of interest is classified through a Support Vector Machine (SVM) and a Bayesian Neural Network (BNN) as normal or abnormal region. SVM is a machine-learning method, based on the principle of structural risk minimization, which shows good performance when applied to data outside the training set. A Bayesian Neural Network is a classifier that joins traditional neural networks theory and Bayesian inference. We use a set of measures obtained by the application of the semivariogram, semimadogram, covariogram, and correlogram functions to the characterization of breast tissue as normal or abnormal. The results show that SVM presents best performance for the classification of breast tissues in mammographic images. The tests indicate that SVM has more generalization power than the BNN classifier. BNN has a sensibility of 76.19% and a specificity of 79.31%, while SVM presents a sensibility of 74.07% and a specificity of 98.77%. The accuracy rate for tests is 78.70% and 92.59% for BNN and SVM, respectively.

Rapid image stitching and computer-aided detection for multipass automated breast ultrasound

PURPOSE

Breast ultrasound (US) is recently becoming more and more popular for detecting breast lesions. However, screening results in hundreds of US images for each subject. This magnitude of images can lead to fatigue in radiologist, causing failure in the detection of lesions of a subtle nature. In this study, an image stitching technique is proposed for combining multipass images of the whole breast into a series of full-view images, and a fully automatic screening system that works off these images is also presented.

METHODS

Using the registration technique based on the simple sum of absolute block-mean difference (SBMD) measure, three-pass images were merged into full-view US images. An automatic screening system was then developed for detecting tumors from these full-view images. The preprocessing step was used to reduce the tumor detection time of the system and to improve image quality. The gray-level slicing method was then used to divide images into numerous regions. Finally, seven computerized features--darkness, uniformity, width-height ratio, area size, nonpersistence, coronal area size, and region continuity--were defined and used to determine whether or not each region was a part of a tumor.

RESULTS

In the experiment, there was a total of 25 experimental cases with 26 lesions, and each case was composed of 252 images (three passes, 84 images/pass). The processing time of the proposed stitching procedure for each case was within 30 s with a Pentium IV 2.0 processor, and the detection sensitivity of the proposed CAD system was 92.3% with 1.76 false positives per case.

CONCLUSIONS

The proposed automatic screening system can be applied to the whole breast images stitched together via SBMD-based registration in order to detect tumors.

Exploring nonlinear feature space dimension reduction and data representation in breast Cadx with Laplacian eigenmaps and t-SNE

PURPOSE

In this preliminary study, recently developed unsupervised nonlinear dimension reduction (DR) and data representation techniques were applied to computer-extracted breast lesion feature spaces across three separate imaging modalities: Ultrasound (U.S.) with 1126 cases, dynamic contrast enhanced magnetic resonance imaging with 356 cases, and full-field digital mammography with 245 cases. Two methods for nonlinear DR were explored: Laplacian eigenmaps [M. Belkin and P. Niyogi, "Laplacian eigenmaps for dimensionality reduction and data representation," Neural Comput. 15, 1373-1396 (2003)] and t-distributed stochastic neighbor embedding (t-SNE) [L. van der Maaten and G. Hinton, "Visualizing data using t-SNE," J. Mach. Learn. Res. 9, 2579-2605 (2008)].

METHODS

These methods attempt to map originally high dimensional feature spaces to more human interpretable lower dimensional spaces while preserving both local and global information. The properties of these methods as applied to breast computer-aided diagnosis (CADx) were evaluated in the context of malignancy classification performance as well as in the visual inspection of the sparseness within the two-dimensional and three-dimensional mappings. Classification performance was estimated by using the reduced dimension mapped feature output as input into both linear and nonlinear classifiers: Markov chain Monte Carlo based Bayesian artificial neural network (MCMC-BANN) and linear discriminant analysis. The new techniques were compared to previously developed breast CADx methodologies, including automatic relevance determination and linear stepwise (LSW) feature selection, as well as a linear DR method based on principal component analysis. Using ROC analysis and 0.632+bootstrap validation, 95% empirical confidence intervals were computed for the each classifier's AUC performance.

RESULTS

In the large U.S. data set, sample high performance results include, AUC0.632+ = 0.88 with 95% empirical bootstrap interval [0.787;0.895] for 13 ARD selected features and AUC0.632+ = 0.87 with interval [0.817;0.906] for four LSW selected features compared to 4D t-SNE mapping (from the original 81D feature space) giving AUC0.632+ = 0.90 with interval [0.847;0.919], all using the MCMC-BANN.

CONCLUSIONS

Preliminary results appear to indicate capability for the new methods to match or exceed classification performance of current advanced breast lesion CADx algorithms. While not appropriate as a complete replacement of feature selection in CADx problems, DR techniques offer a complementary approach, which can aid elucidation of additional properties associated with the data. Specifically, the new techniques were shown to possess the added benefit of delivering sparse lower dimensional representations for visual interpretation, revealing intricate data structure of the feature space.

Comparison of algorithms to enhance spicules of spiculated masses on mammography

We have developed an algorithm for enhancement of spicules of spiculated masses, which uses the discrete radon transform. Previously, we employed a commonly used method to compute the discrete radon transform, which we refer to as the DRT. Recently, a new, more exact method to compute the discrete radon transform was developed by Averbuch et al, which is called the fast slant stack (FSS) method. Our hypothesis was that this new formulation would help to improve our enhancement algorithm. To test this idea, we conducted multiple two-alternative-forced-choice observer studies and found that most observers preferred the enhanced images generated with the FSS method.

Correlations among acoustic, texture and morphological features for breast ultrasound CAD

Acoustic, textural and morphological features of the breast in ultrasound imaging were extracted for computer-aided diagnosis. In addition, correlations among different categories of features were analyzed. Clinical data from 14 patients (7 malignant and 7 benign samples) were acquired. A custom-made experimental apparatus was used for simultaneous data acquisition of B-mode ultrasound and limited-angle tomography images. Textural features were extracted from B-mode images, including five parameters derived from the gray-level concurrence matrix and five parameters derived from a nonseparable wavelet transform. Morphological features were also extracted from B-mode images, including the depth-to-width ratio and normalized radial gradient. Acoustic features were estimated using limited-angle tomography, including the sound velocity and attenuation coefficient. Generally, the correlation coefficients for features within the textural feature group were relatively high (0.48-0.79), whereas those between different feature categories were relatively low (0.17-0.40). This suggests that combining different sets of features would improve the computer-aided diagnosis of breast cancer.

A model-based framework for the detection of spiculated masses on mammography

The detection of lesions on mammography is a repetitive and fatiguing task. Thus, computer-aided detection systems have been developed to aid radiologists. The detection accuracy of current systems is much higher for clusters of microcalcifications than for spiculated masses. In this article, the authors present a new model-based framework for the detection of spiculated masses. The authors have invented a new class of linear filters, spiculated lesion filters, for the detection of converging lines or spiculations. These filters are highly specific narrowband filters, which are designed to match the expected structures of spiculated masses. As a part of this algorithm, the authors have also invented a novel technique to enhance spicules on mammograms. This entails filtering in the radon domain. They have also developed models to reduce the false positives due to normal linear structures. A key contribution of this work is that the parameters of the detection algorithm are based on measurements of physical properties of spiculated masses. The results of the detection algorithm are presented in the form of free-response receiver operating characteristic curves on images from the Mammographic Image Analysis Society and Digital Database for Screening Mammography databases.

Contrast enhancement in dense breast images to aid clustered microcalcifications detection

This paper presents a method to provide contrast enhancement in dense breast digitized images, which are difficult cases in testing of computer-aided diagnosis (CAD) schemes. Three techniques were developed, and data from each method were combined to provide a better result in relation to detection of clustered microcalcifications. Results obtained during the tests indicated that, by combining all the developed techniques, it is possible to improve the performance of a processing scheme designed to detect microcalcification clusters. It also allows operators to distinguish some of these structures in low-contrast images, which were not detected via conventional processing before the contrast enhancement. This investigation shows the possibility of improving CAD schemes for better detection of microcalcifications in dense breast images.

Characterization of difference of Gaussian filters in the detection of mammographic regions

In this article, we present a characterization of the effect of difference of Gaussians (DoG) filters in the detection of mammographic regions. DoG filters have been used previously in mammographic mass computer-aided detection (CAD) systems. As DoG filters are constructed from the subtraction of two bivariate Gaussian distributions, they require the specification of three parameters: the size of the filter template and the standard deviations of the constituent Gaussians. The influence of these three parameters in the detection of mammographic masses has not been characterized. In this work, we aim to determine how the parameters affect (1) the physical descriptors of the detected regions, (2) the true and false positive rates, and (3) the classification performance of the individual descriptors. To this end, 30 DoG filters are created from the combination of three template sizes and four values for each of the Gaussians' standard deviations. The filters are used to detect regions in a study database of 181 craniocaudal-view mammograms extracted from the Digital Database for Screening Mammography. To describe the physical characteristics of the identified regions, morphological and textural features are extracted from each of the detected regions. Differences in the mean values of the features caused by altering the DoG parameters are examined through statistical and empirical comparisons. The parameters' effects on the true and false positive rate are determined by examining the mean malignant sensitivities and false positives per image (FPpI). Finally, the effect on the classification performance is described by examining the variation in FPpI at the point where 81% of the malignant masses in the study database are detected. Overall, the findings of the study indicate that increasing the standard deviations of the Gaussians used to construct a DoG filter results in a dramatic decrease in the number of regions identified at the expense of missing a small number of malignancies. The sharp reduction in the number of identified regions allowed the identification of textural differences between large and small mammographic regions. We find that the classification performances of the features that achieve the lowest average FPpI are influenced by all three of the parameters.

The reliability of measuring physical characteristics of spiculated masses on mammography

The goal of this study was to assess the reliability of measurements of the physical characteristics of spiculated masses on mammography. The images used in this study were obtained from the Digital Database for Screening Mammography. Two experienced radiologists measured the properties of 21 images of spiculated masses. The length and width of all spicules and the major axis of the mass were measured. In addition, the observers counted the total number of spicules. Interobserver and intraobserver variability were evaluated using a hypothesis test for equivalence, the intraclass correlation coefficient (ICC) and Bland-Altman statistics. For an equivalence level of 30% of the mean of the senior radiologist's measurement, equivalence was achieved for the measurements of average spicule length (p<0.01), average spicule width (p = 0.03), the length of the major axis (p<0.01) and for the count of the number of spicules (p<0.01). Similarly, with the ICC analysis technique "excellent" inter-rater agreement was observed for the measurements of average spicule length (ICC = 0.770), the length of the major axis (ICC = 0.801) and for the count of the number of spicules (ICC = 0.780). "Fair to good" agreement was observed for the average spicule width (ICC = 0.561). Equivalence was also demonstrated for intraobserver measurements. Physical properties of spiculated masses can be measured reliably on mammography. The interobserver and intraobserver variability for this task is comparable with that reported for other measurements made on medical images.

Impact of missing data in evaluating artificial neural networks trained on complete data

This study investigated the impact of missing data in the evaluation of artificial neural network (ANN) models trained on complete data for the task of predicting whether breast lesions are benign or malignant from their mammographic Breast Imaging and Reporting Data System (BI-RADS) descriptors. A feed-forward, back-propagation ANN was tested with three methods for estimating the missing values. Similar results were achieved with a constraint satisfaction ANN, which can accommodate missing values without a separate estimation step. This empirical study highlights the need for additional research on developing robust clinical decision support systems for realistic environments in which key information may be unknown or inaccessible.

A ranklet-based image representation for mass classification in digital mammograms

Regions of interest (ROIs) found on breast radiographic images are classified as either tumoral mass or normal tissue by means of a support vector machine classifier. Classification features are the coefficients resulting from the specific image representation used to encode each ROI. Pixel and wavelet image representations have already been discussed in one of our previous works. To investigate the possibility of improving classification performances, a novel nonparametric, orientation-selective, and multiresolution image representation is developed and evaluated, namely a ranklet image representation. A dataset consisting of 1000 ROIs representing biopsy-proven tumoral masses (either benign or malignant) and 5000 ROIs representing normal breast tissue is used. ROIs are extracted from the digital database for screening mammography collected by the University of South Florida. Classification performances are evaluated using the area Az under the receiver operating characteristic curve. By achieving Az values of 0.978 +/- 0.003 and 90% sensitivity with a false positive fraction value of 4.5%, experiments demonstrate classification results higher than those reached by the previous image representations. In particular, the improvement on the Az value over that achieved by the wavelet image representation is statistically relevant with the two-tailed p value <0.0001. Besides, owing to the tolerance that the ranklet image representation reveals to variations in the ROIs' gray-level intensity histogram, this approach discloses to be robust also when tested on radiographic images having gray-level intensity histogram remarkably different from that used for training.

Diffuse boundary extraction of breast masses on ultrasound by leak plugging

We propose a semiautomated seeded boundary extraction algorithm that delineates diffuse region boundaries by finding and plugging their leaks. The algorithm not only extracts boundaries that are partially diffuse, but in the process finds and quantifies those parts of the boundary that are diffuse, computing local sharpness measurements for possible use in computer-aided diagnosis. The method treats a manually drawn seed region as a wellspring of pixel "fluid" that flows from the seed out towards the boundary. At indistinct or porous sections of the boundary, the growing region will leak into surrounding tissue. By changing the size of structuring elements used for growing, the algorithm changes leak properties. Since larger elements cannot leak as far from the seed, they produce compact, less detailed boundary approximations; conversely, growing from smaller elements results in less constrained boundaries with more local detail. This implementation of the leak plugging algorithm decrements the radius of structuring disks and then compares the regions grown from them as they increase in both area and boundary detail. Leaks are identified if the outflows between grown regions are large compared to the areas of the disks. The boundary is plugged by masking out leaked pixels, and the process continues until one-pixel-radius resolution. When tested against manual delineation on scans of 40 benign masses and 40 malignant tumors, the plugged boundaries overlapped and correlated well in area with manual tracings, with mean overlap of 0.69 and area correlation R2 of 0.86, but the algorithm's results were more reproducible.

Correspondence in texture features between two mammographic views

It is well established that radiologists are better able to interpret mammograms when two mammographic views are available. Consequently, two mammographic projections are standard: mediolateral oblique (MLO) and craniocaudal (CC). Computer-aided diagnosis algorithms have been investigated for assisting in the detection and diagnosis of breast lesions in digitized/digital mammograms. A few previous studies suggest that computer-aided systems may also benefit from combining evidence from the two views. Intuitively, we expect that there would only be value in merging data from two views if they provide complementary information. A measure of the similarity of information is the correlation coefficient between corresponding features from the MLO and CC views. The purpose of this study was to investigate the correspondence in Haralick's texture features between the MLO and CC mammographic views of breast lesions. Features were ranked on the basis of correlation values and the two-view correlation of features for subgroups of data including masses versus calcification and benign versus malignant lesions were compared. All experiments were performed on a subset of mammography cases from the Digital Database for Screening Mammography (DDSM). It was observed that the texture features from the MLO and CC views were less strongly correlated for calcification lesions than for mass lesions. Similarly, texture features from the two views were less strongly correlated for benign lesions than for malignant lesions. These differences were statistically significant. The results suggest that the inclusion of texture features from multiple mammographic views in a CADx algorithm may impact the accuracy of diagnosis of calcification lesions and benign lesions.

Self-organizing map for cluster analysis of a breast cancer database

The purpose of this study was to identify and characterize clusters in a heterogeneous breast cancer computer-aided diagnosis database. Identification of subgroups within the database could help elucidate clinical trends and facilitate future model building. A self-organizing map (SOM) was used to identify clusters in a large (2258 cases), heterogeneous computer-aided diagnosis database based on mammographic findings (BI-RADS) and patient age. The resulting clusters were then characterized by their prototypes determined using a constraint satisfaction neural network (CSNN). The clusters showed logical separation of clinical subtypes such as architectural distortions, masses, and calcifications. Moreover, the broad categories of masses and calcifications were stratified into several clusters (seven for masses and three for calcifications). The percent of the cases that were malignant was notably different among the clusters (ranging from 6 to 83%). A feed-forward back-propagation artificial neural network (BP-ANN) was used to identify likely benign lesions that may be candidates for follow up rather than biopsy. The performance of the BP-ANN varied considerably across the clusters identified by the SOM. In particular, a cluster (#6) of mass cases (6% malignant) was identified that accounted for 79% of the recommendations for follow up that would have been made by the BP-ANN. A classification rule based on the profile of cluster #6 performed comparably to the BP-ANN, providing approximately 25% specificity at 98% sensitivity. This performance was demonstrated to generalize to a large (2177) set of cases held-out for model validation.

Combining low-, high-level and empirical domain knowledge for automated segmentation of ultrasonic breast lesions

Breast cancer is the most frequently diagnosed malignancy and the second leading cause of mortality in women. In the last decade, ultrasound along with digital mammography has come to be regarded as the gold standard for breast cancer diagnosis. Automatically detecting tumors and extracting lesion boundaries in ultrasound images is difficult due to their specular nature and the variance in shape and appearance of sonographic lesions. Past work on automated ultrasonic breast lesion segmentation has not addressed important issues such as shadowing artifacts or dealing with similar tumor like structures in the sonogram. Algorithms that claim to automatically classify ultrasonic breast lesions, rely on manual delineation of the tumor boundaries. In this paper, we present a novel technique to automatically find lesion margins in ultrasound images, by combining intensity and texture with empirical domain specific knowledge along with directional gradient and a deformable shape-based model. The images are first filtered to remove speckle noise and then contrast enhanced to emphasize the tumor regions. For the first time, a mathematical formulation of the empirical rules used by radiologists in detecting ultrasonic breast lesions, popularly known as the "Stavros Criteria" is presented in this paper. We have applied this formulation to automatically determine a seed point within the image. Probabilistic classification of image pixels based on intensity and texture is followed by region growing using the automatically determined seed point to obtain an initial segmentation of the lesion. Boundary points are found on the directional gradient of the image. Outliers are removed by a process of recursive refinement. These boundary points are then supplied as an initial estimate to a deformable model. Incorporating empirical domain specific knowledge along with low and high-level knowledge makes it possible to avoid shadowing artifacts and lowers the chance of confusing similar tumor like structures for the lesion. The system was validated on a database of breast sonograms for 42 patients. The average mean boundary error between manual and automated segmentation was 6.6 pixels and the normalized true positive area overlap was 75.1%. The algorithm was found to be robust to 1) variations in system parameters, 2) number of training samples used, and 3) the position of the seed point within the tumor. Running time for segmenting a single sonogram was 18 s on a 1.8-GHz Pentium machine.

Contrast enhancement in dense breast images using the modulation transfer function

This work proposes a method aimed at enhancing the contrast in dense breast images in mammography. It includes a new preprocessing technique, which uses information on the modulation transfer function (MTF) of the mammographic system in the whole radiation field. The method is applied to improve the efficiency of a computer-aided diagnosis (CAD) scheme. Seventy-five regions of interest (ROIs) from dense mammograms were acquired in two pieces of equipment (a CGR Senographe 500t and a Philips Mammodiagnost) and were digitized in a Lumiscan 50 laser scanner. A computational procedure determines the effective focal spot size in each region of interest from the measured focal spot in the center for a given mammographic equipment. Using computational simulation the MTF is then calculated for each field region. A procedure that enlarges the high-frequency portion of this function is applied and a convolution between the resulting new function and the original image is performed. Both original and enhanced images were submitted to a processing procedure for detecting clustered microcalcifications in order to compare the performance for dense breast images. ROIs were divided into four groups, two for each piece of equipment-one with clustered microcalcifications and another without microcalcifications. Our results show that in about 10% of the enhanced images more signals were detected when compared to the results for the original dense breast images. This is important because the usual processing techniques used in CAD schemes present poor results when applied to dense breast images. Since the MTF method is a well-recognized tool in the evaluation of radiographic systems, this new technique could be used to associate quality assurance procedures with the processing schemes employed in CAD for mammography.

Differences between computer-aided diagnosis of breast masses and that of calcifications

PURPOSE

To compare the performance of a computer-aided diagnosis (CAD) system for diagnosis of previously detected lesions, based on radiologist-extracted findings on masses and calcifications.

MATERIALS AND METHODS

A feed-forward, back-propagation artificial neural network (BP-ANN) was trained in a round-robin (leave-one-out) manner to predict biopsy outcome from mammographic findings (according to the Breast Imaging Reporting and Data System) and patient age. The BP-ANN was trained by using a large (>1,000 cases) heterogeneous data set containing masses and microcalcifications. The performances of the BP-ANN on masses and microcalcifications were compared with use of receiver operating characteristic analysis and a z test for uncorrelated samples.

RESULTS

The BP-ANN performed significantly better on masses than microcalcifications in terms of both the area under the receiver operating characteristic curve and the partial receiver operating characteristic area index. A similar difference in performance was observed with a second model (linear discriminant analysis) and also with a second data set from a similar institution.

CONCLUSION

Masses and calcifications should be considered separately when evaluating CAD systems for breast cancer diagnosis.

Patient information extraction in digitized radiography

Digital imagery is gradually replacing the traditional radiograph with the development of digital radiography and film scanner. This report presents a new method to extract the patient information number (PIN) field automatically from the film-scanned image using image analysis technique. To evaluate the PIN field extraction algorithm, 2 formats of label acquired from 2 different hospitals are tested. Given the available films with no constraints on the way the labels are written and positioned, the correct extraction rates are 73% and 84%, respectively. This extracted PIN information can link with Radiology Information System (RIS) or Hospital Information System (HIS), and the image scanned from the film then can be filed into the database automatically. The efficiency this method offers can simplify greatly the image filing process and improve the user friendliness of the overall image digitization system. Moreover, compared with the bar code reader, it solves the automatic information input problem in a very economical way. The authors believe the success of this technique will benefit the development of the PACS (Picture Archiving and Communication System) and teleradiology.