Classification of breast ultrasound images using fractal feature

Classification of multi-feature fusion ultrasound images of breast tumor within category 4 using convolutional neural networks

BACKGROUND

Breast tumor is a fatal threat to the health of women. Ultrasound (US) is a common and economical method for the diagnosis of breast cancer. Breast imaging reporting and data system (BI-RADS) category 4 has the highest false-positive value of about 30% among five categories. The classification task in BI-RADS category 4 is challenging and has not been fully studied.

PURPOSE

This work aimed to use convolutional neural networks (CNNs) for breast tumor classification using B-mode images in category 4 to overcome the dependence on operator and artifacts. Additionally, this work intends to take full advantage of morphological and textural features in breast tumor US images to improve classification accuracy.

METHODS

First, original US images coming directly from the hospital were cropped and resized. In 1385 B-mode US BI-RADS category 4 images, the biopsy eliminated 503 samples of benign tumor and left 882 of malignant. Then, K-means clustering algorithm and entropy of sliding windows of US images were conducted. Considering the diversity of different characteristic information of malignant and benign represented by original B-mode images, K-means clustering images and entropy images, they are fused in a three-channel form multi-feature fusion images dataset. The training, validation, and test sets are 969, 277, and 139. With transfer learning, 11 CNN models including DenseNet and ResNet were investigated. Finally, by comparing accuracy, precision, recall, F1-score, and area under curve (AUC) of the results, models which had better performance were selected. The normality of data was assessed by Shapiro-Wilk test. DeLong test and independent t-test were used to evaluate the significant difference of AUC and other values. False discovery rate was utilized to ultimately evaluate the advantages of CNN with highest evaluation metrics. In addition, the study of anti-log compression was conducted but no improvement has shown in CNNs classification results.

RESULTS

With multi-feature fusion images, DenseNet121 has highest accuracy of 80.22 ± 1.45% compared to other CNNs, precision of 77.97 ± 2.89% and AUC of 0.82 ± 0.01. Multi-feature fusion improved accuracy of DenseNet121 by 1.87% from classification of original B-mode images (p < 0.05).

CONCLUSION

The CNNs with multi-feature fusion show a good potential of reducing the false-positive rate within category 4. The work illustrated that CNNs and fusion images have the potential to reduce false-positive rate in breast tumor within US BI-RADS category 4, and make the diagnosis of category 4 breast tumors to be more accurate and precise.

Joint Localization and Classification of Breast Cancer in B-Mode Ultrasound Imaging via Collaborative Learning with Elastography

Convolutional neural networks (CNNs) have been successfully applied in the computer-aided ultrasound diagnosis for breast cancer. Up to now, several CNN-based methods have been proposed. However, most of them consider tumor localization and classification as two separate steps, rather than performing them simultaneously. Besides, they suffer from the limited diagnosis information in the B-mode ultrasound (BUS) images. In this study, we develop a novel network ResNet-GAP that incorporates both localization and classification into a unified procedure. To enhance the performance of ResNet-GAP, we leverage stiffness information in the elastography ultrasound (EUS) modality by collaborative learning in the training stage. Specifically, a dual-channel ResNet-GAP is developed, one channel for BUS and the other for EUS. In each channel, multiple class activity maps (CAMs) are generated using a series of convolutional kernels of different sizes. The multi-scale consistency of the CAMs in both channels are further considered in network optimization. Experiments on 264 patients in this study show that the newly developed ResNet-GAP achieves an accuracy of 88.6%, a sensitivity of 95.3%, a specificity of 84.6%, and an AUC of 93.6% on the classification task, and a 1.0NLF of 87.9% on the localization task, which is better than some state-of-the-art approaches.

A Modified HSIFT Descriptor for Medical Image Classification of Anatomy Objects

Modeling low level features to high level semantics in medical imaging is an important aspect in filtering anatomy objects. Bag of Visual Words (BOVW) representations have been proven effective to model these low level features to mid level representations. Convolutional neural nets are learning systems that can automatically extract high-quality representations from raw images. However, their deployment in the medical field is still a bit challenging due to the lack of training data. In this paper, learned features that are obtained by training convolutional neural networks are compared with our proposed hand-crafted HSIFT features. The HSIFT feature is a symmetric fusion of a Harris corner detector and the Scale Invariance Transform process (SIFT) with BOVW representation. The SIFT process is enhanced as well as the classification technique by adopting bagging with a surrogate split method. Quantitative evaluation shows that our proposed hand-crafted HSIFT feature outperforms the learned features from convolutional neural networks in discriminating anatomy image classes.

Detection and recognition of ultrasound breast nodules based on semi-supervised deep learning: a powerful alternative strategy

BACKGROUND

The successful recognition of benign and malignant breast nodules using ultrasound images is based mainly on supervised learning that requires a large number of labeled images. However, because high-quality labeling is expensive and time-consuming, we hypothesized that semi-supervised learning could provide a low-cost and powerful alternative approach. This study aimed to develop an accurate semi-supervised recognition method and compared its performance with supervised methods and sonographers.

METHODS

The faster region-based convolutional neural network was used for nodule detection from ultrasound images. A semi-supervised classifier based on the mean teacher model was proposed to recognize benign and malignant nodule images. The general performance of the proposed method on two datasets (8,966 nodules) was reported.

RESULTS

The detection accuracy was 0.88±0.03 and 0.86±0.02, respectively, on two testing sets (1,350 and 2,220 nodules). When 800 labeled training nodules were available, the proposed semi-supervised model plus 4,396 unlabeled nodules performed better than the supervised learning model (area under the curve (AUC): 0.934±0.026 vs. 0.83±0.050; 0.916±0.022 vs. 0.815±0.049). The performance of the semi-supervised model trained on 800 labeled and 4,396 unlabeled nodules was close to that of the supervised learning model trained on a massive number of labeled nodules (n=5,196) (AUC: 0.934±0.026 vs. 0.952±0.027; 0.916±0.022 vs. 0.918±0.017). Moreover, the semi-supervised model was better than the average accuracy of five human sonographers (AUC: 0.922 vs. 0.889).

CONCLUSIONS

The semi-supervised model can achieve excellent performance for nodule recognition and be useful for medical sciences. The method reduced the number of labeled images required for training, thus significantly alleviating the difficulty in data preparation of medical artificial intelligence.

Quantitative Evaluation of Rotator Cuff Tears Based on Non-linear Statistical Analysis of Ultrasound Radiofrequency Signals

There is increasing clinical requirement for early and accurate ultrasound diagnosis of rotator cuff tears (RCTs). A method based on non-linear statistical analysis was proposed for the detection of RCTs using ultrasound radiofrequency (RF) signals. One hundred fifty-two patients with shoulder pain were first examined with ultrasound and then diagnosed with magnetic resonance imaging (MRI) as the ground truth. By comparison of the region of interest (ROI) with a part of the supraspinatus with no pathologic change part in the same RF signal frame, the relative P_ks value (viz., rP_ks value) was evaluated to quantify the pathophysiologic changes. The results indicated that the rP_ks values of all RCTs are <0.7, and the accuracy, sensitivity and specificity of the proposed method can reach 97.5%, 100% and 91.4%, respectively. This computer-aided method was found to perform better diagnostic than the results reported by an experienced radiologist (accuracy = 75.7%, sensitivity = 72.6%, and specificity = 85.7%). The high sensitivity advantage of this method indicates that the prospects for its application in the computer-aided diagnosis of RCTs are good.

Detecting Brain Tumor using Machines Learning Techniques Based on Different Features Extracting Strategies

BACKGROUND

Brain tumor is the leading cause of death worldwide. It is obvious that the chances of survival can be increased if the tumor is identified and properly classified at an initial stage. MRI (Magnetic Resonance Imaging) is one source of brain tumors detection tool and is extensively used in the diagnosis of brain to detect blood clots. In the past, many researchers developed Computer-Aided Diagnosis (CAD) systems that help the radiologist to detect the abnormalities in an efficient manner.

OBJECTIVE

The aim of this research is to improve the brain tumor detection performance by proposing a multimodal feature extracting strategy and employing machine learning techniques.

METHODS

In this study, we extracted multimodal features such as texture, morphological, entropybased, Scale Invariant Feature Transform (SIFT), and Elliptic Fourier Descriptors (EFDs) from brain tumor imaging database. The tumor was detected using robust machine learning techniques such as Support Vector Machine (SVM) with kernels: polynomial, Radial Base Function (RBF), Gaussian; Decision Tree (DT), and Naïve Bayes. Most commonly used Jack-knife 10-fold Cross- Validation (CV) was used for testing and validation of dataset.

RESULTS

The performance was evaluated in terms of specificity, sensitivity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), False Positive Rate (FPR), Total Accuracy (TA), Area under the receiver operating Curve (AUC), and P-value. The highest performance of 100% in terms of Specificity, Sensitivity, PPV, NPV, TA, AUC using Naïve Bayes classifiers based on entropy, morphological, SIFT and texture features followed by Decision Tree classifier with texture features (TA=97.81%, AUC=1.0) and SVM polynomial kernel with texture features (TA=94.63%). The highest significant p-value was obtained using SVM polynomial with texture features (P-value 2.65e-104) followed by SVM RB with texture features (P-value 1.96e-98).

CONCLUSION

The results reveal that Naïve Bayes followed by Decision Tree gives highest detection accuracy based on entropy, morphological, SIFT and texture features.

Computer-aided diagnosis of breast ultrasound images using ensemble learning from convolutional neural networks

Breast ultrasound and computer aided diagnosis (CAD) has been used to classify tumors into benignancy or malignancy. However, conventional CAD software has some problems (such as handcrafted features are hard to design; conventional CAD systems are difficult to confirm overfitting problems, etc.). In our study, we propose a CAD system for tumor diagnosis using an image fusion method combined with different image content representations and ensemble different CNN architectures on US images. The CNN-based method proposed in this study includes VGGNet, ResNet, and DenseNet. In our private dataset, there was a total of 1687 tumors that including 953 benign and 734 malignant tumors. The accuracy, sensitivity, specificity, precision, F1 score and the AUC of the proposed method were 91.10%, 85.14%, 95.77%, 94.03%, 89.36%, and 0.9697 respectively. In the open dataset (BUSI), there was a total of 697 tumors that including 437 benign lesions, 210 malignant tumors, and 133 normal images. The accuracy, sensitivity, specificity, precision, F1 score, and the AUC of the proposed method were 94.62%, 92.31%, 95.60%, 90%, 91.14%, and 0.9711. In conclusion, the results indicated different image content representations that affect the prediction performance of the CAD system, more image information improves the prediction performance, and the tumor shape feature can improve the diagnostic effect.

Texture Analysis Based on Auto-Mutual Information for Classifying Breast Lesions with Ultrasound

Described here is a novel texture extraction method based on auto-mutual information (AMI) for classifying breast lesions. The objective is to extract discriminating information found in the non-linear relationship of textures in breast ultrasound (BUS) images. The AMI method performs three basic tasks: (i) it transforms the input image using the ranklet transform to handle intensity variations of BUS images acquired with distinct ultrasound scanners; (ii) it extracts the AMI-based texture features in the horizontal and vertical directions from each ranklet image; and (iii) it classifies the breast lesions into benign and malignant classes, in which a support-vector machine is used as the underlying classifier. The image data set is composed of 2050 BUS images consisting of 1347 benign and 703 malignant tumors. Additionally, nine commonly used texture extraction methods proposed in the literature for BUS analysis are compared with the AMI method. The bootstrap method, which considers 1000 bootstrap samples, is used to evaluate classification performance. The experimental results indicate that the proposed approach outperforms its counterparts in terms of area under the receiver operating characteristic curve, sensitivity, specificity and Matthews correlation coefficient, with values of 0.82, 0.80, 0.85 and 0.63, respectively. These results suggest that the AMI method is suitable for breast lesion classification systems.

Automated diagnosis of breast ultrasonography images using deep neural networks

Ultrasonography images of breast mass aid in the detection and diagnosis of breast cancer. Manually analyzing ultrasonography images is time-consuming, exhausting and subjective. Automated analyzing such images is desired. In this study, we develop an automated breast cancer diagnosis model for ultrasonography images. Traditional methods of automated ultrasonography images analysis employ hand-crafted features to classify images, and lack robustness to the variation in the shapes, size and texture of breast lesions, leading to low sensitivity in clinical applications. To overcome these shortcomings, we propose a method to diagnose breast ultrasonography images using deep convolutional neural networks with multi-scale kernels and skip connections. Our method consists of two components: the first one is to determine whether there are malignant tumors in the image, and the second one is to recognize solid nodules. In order to let the two networks work in a collaborative way, a region enhance mechanism based on class activation maps is proposed. The mechanism helps to improve classification accuracy and sensitivity for both networks. A cross training algorithm is introduced to train the networks. We construct a large annotated dataset containing a total of 8145 breast ultrasonography images to train and evaluate the models. All of the annotations are proven by pathological records. The proposed method is compared with two state-of-the-art approaches, and outperforms both of them by a large margin. Experimental results show that our approach achieves a performance comparable to human sonographers and can be applied to clinical scenarios.

A computer-aided diagnosis system for breast ultrasound based on weighted BI-RADS classes

BACKGROUND AND OBJECTIVE

Conventional computer-aided diagnosis (CAD) systems for breast ultrasound (BUS) are trained to classify pathological classes, that is, benign and malignant. However, from a clinical perspective, this kind of classification does not agree totally with radiologists' diagnoses. Usually, the tumors are assessed by using a BI-RADS (Breast Imaging-Reporting and Data System) category and, accordingly, a recommendation is emitted: annual study for category 2 (benign), six-month follow-up study for category 3 (probably benign), and biopsy for categories 4 and 5 (suspicious of malignancy). Hence, in this paper, a CAD system based on BI-RADS categories weighted by pathological information is presented. The goal is to increase the classification performance by reducing the common class imbalance found in pathological classes as well as to provide outcomes quite similar to radiologists' recommendations.

METHODS

The BUS dataset considers 781 benign lesions and 347 malignant tumors proven by biopsy. Moreover, every lesion is associated to one BI-RADS category in the set {2, 3, 4, 5}. Thus, the dataset is split into three weighted classes: benign, BI-RADS 2 in benign lesions; probably benign, BI-RADS 3 and 4 in benign lesions; and malignant, BI-RADS 4 and 5 in malignant lesions. Thereafter, a random forest (RF) classifier, denoted by RF_w, is trained to predict the weighted BI-RADS classes. In addition, for comparison purposes, a RF classifier is trained to predict pathological classes, denoted as RF_p.

RESULTS

The ability of the classifiers to predict the pathological classes is measured by the area under the ROC curve (AUC), sensitivity (SEN), and specificity (SPE). The RF_w classifier obtained AUC=0.872,SEN=0.826, and SPE=0.919, whereas the RF_p classifier reached AUC=0.868,SEN=0.808, and SPE=0.929. According to a one-way analysis of variance test, the RF_w classifier statistically outperforms (p < 0.001) the RF_p classifier in terms of the AUC and SEN. Moreover, the classification performance of RF_w to predict weighted BI-RADS classes is given by the Matthews correlation coefficient that obtained 0.614.

CONCLUSIONS

The division of the classification problem into three classes reduces the imbalance between benign and malignant classes; thus, the sensitivity is increased without degrading the specificity. Therefore, the CAD based on weighted BI-RADS classes improves the classification performance of the conventional CAD systems. Additionally, the proposed approach has the advantage of being capable of providing a multiclass outcome related to radiologists' recommendations.

A nonlinear approach to identify pathological change of thyroid nodules based on statistical analysis of ultrasound RF signals

In order to reassure the majority of patients with benign nodules from unnecessary needle biopsy, there is an increasing clinical requirement to identify benign and malignant thyroid nodules during ultrasound diagnosis. A nonlinear approach based on statistical analysis of ultrasound radio-frequency (RF) signals was developed for differential diagnosing the thyroid nodules to improve the diagnostic accuracy. Data from 44 patients with solitary thyroid nodules were collected, following with the ultrasound-guided fine needle aspiration (FNA) as the ground truth. The relative P-value (rP-value) was estimated to quantify the pathophysiologic changes by comparing the region of interest (ROI) with the no pathological change part in the thyroid gland using only one frame of raw RF data. The malignant nodules were distinguished from benign ones with high accuracy and high credibility (sensitivity = 100%, specificity = 80%). Suspicious nodules (rP-value < 0.5) could be picked out for FNA with no additional instruments. This method shows promising in differentiating malignant from benign thyroid nodules, subsequently avoiding unnecessary biopsies.

Sparse Representation Based Multi-Instance Learning for Breast Ultrasound Image Classification

We propose a novel method based on sparse representation for breast ultrasound image classification under the framework of multi-instance learning (MIL). After image enhancement and segmentation, concentric circle is used to extract the global and local features for improving the accuracy in diagnosis and prediction. The classification problem of ultrasound image is converted to sparse representation based MIL problem. Each instance of a bag is represented as a sparse linear combination of all basis vectors in the dictionary, and then the bag is represented by one feature vector which is obtained via sparse representations of all instances within the bag. The sparse and MIL problem is further converted to a conventional learning problem that is solved by relevance vector machine (RVM). Results of single classifiers are combined to be used for classification. Experimental results on the breast cancer datasets demonstrate the superiority of the proposed method in terms of classification accuracy as compared with state-of-the-art MIL methods.

A contrast enhancement method for improving the segmentation of breast lesions on ultrasonography

PURPOSE

This paper presents an adaptive contrast enhancement method based on sigmoidal mapping function (SACE) used for improving the computerized segmentation of breast lesions on ultrasound.

METHODS

First, from the original ultrasound image an intensity variation map is obtained, which is used to generate local sigmoidal mapping functions related to distinct contextual regions. Then, a bilinear interpolation scheme is used to transform every original pixel to a new gray level value. Also, four contrast enhancement techniques widely used in breast ultrasound enhancement are implemented: histogram equalization (HEQ), contrast limited adaptive histogram equalization (CLAHE), fuzzy enhancement (FEN), and sigmoid based enhancement (SEN). In addition, these contrast enhancement techniques are considered in a computerized lesion segmentation scheme based on watershed transformation. The performance comparison among techniques is assessed in terms of both the quality of contrast enhancement and the segmentation accuracy. The former is quantified by the measure, where the greater the value, the better the contrast enhancement, whereas the latter is calculated by the Jaccard index, which should tend towards unity to indicate adequate segmentation.

RESULTS

The experiments consider a data set with 500 breast ultrasound images. The results show that SACE outperforms its counterparts, where the median values for the measure are: SACE: 139.4, SEN: 68.2, HEQ: 64.1, CLAHE: 62.8, and FEN: 7.9. Considering the segmentation performance results, the SACE method presents the largest accuracy, where the median values for the Jaccard index are: SACE: 0.81, FEN: 0.80, CLAHE: 0.79, HEQ: 77, and SEN: 0.63.

CONCLUSION

The SACE method performs well due to the combination of three elements: (1) the intensity variation map reduces intensity variations that could distort the real response of the mapping function, (2) the sigmoidal mapping function enhances the gray level range where the transition between lesion and background is found, and (3) the adaptive enhancing scheme for coping with local contrasts. Hence, the SACE approach is appropriate for enhancing contrast before computerized lesion segmentation.

A Fusion-Based Approach for Breast Ultrasound Image Classification Using Multiple-ROI Texture and Morphological Analyses

Ultrasound imaging is commonly used for breast cancer diagnosis, but accurate interpretation of breast ultrasound (BUS) images is often challenging and operator-dependent. Computer-aided diagnosis (CAD) systems can be employed to provide the radiologists with a second opinion to improve the diagnosis accuracy. In this study, a new CAD system is developed to enable accurate BUS image classification. In particular, an improved texture analysis is introduced, in which the tumor is divided into a set of nonoverlapping regions of interest (ROIs). Each ROI is analyzed using gray-level cooccurrence matrix features and a support vector machine classifier to estimate its tumor class indicator. The tumor class indicators of all ROIs are combined using a voting mechanism to estimate the tumor class. In addition, morphological analysis is employed to classify the tumor. A probabilistic approach is used to fuse the classification results of the multiple-ROI texture analysis and morphological analysis. The proposed approach is applied to classify 110 BUS images that include 64 benign and 46 malignant tumors. The accuracy, specificity, and sensitivity obtained using the proposed approach are 98.2%, 98.4%, and 97.8%, respectively. These results demonstrate that the proposed approach can effectively be used to differentiate benign and malignant tumors.

New Fully Automated Method for Segmentation of Breast Lesions on Ultrasound Based on Texture Analysis

The study described here explored a fully automatic segmentation approach based on texture analysis for breast lesions on ultrasound images. The proposed method involves two main stages: (i) In lesion region detection, the original gray-scale image is transformed into a texture domain based on log-Gabor filters. Local texture patterns are then extracted from overlapping lattices that are further classified by a linear discriminant analysis classifier to distinguish between the "normal tissue" and "breast lesion" classes. Next, an incremental method based on the average radial derivative function reveals the region with the highest probability of being a lesion. (ii) In lesion delineation, using the detected region and the pre-processed ultrasound image, an iterative thresholding procedure based on the average radial derivative function is performed to determine the final lesion contour. The experiments are carried out on a data set of 544 breast ultrasound images (including cysts, benign solid masses and malignant lesions) acquired with three distinct ultrasound machines. In terms of the area under the receiver operating characteristic curve, the one-way analysis of variance test (α=0.05) indicates that the proposed approach significantly outperforms two published fully automatic methods (p<0.001), for which the areas under the curve are 0.91, 0.82 and 0.63, respectively. Hence, these results suggest that the log-Gabor domain improves the discrimination power of texture features to accurately segment breast lesions. In addition, the proposed approach can potentially be used for automated computer diagnosis purposes to assist physicians in detection and classification of breast masses.

Computer Aided Theragnosis Using Quantitative Ultrasound Spectroscopy and Maximum Mean Discrepancy in Locally Advanced Breast Cancer

A noninvasive computer-aided-theragnosis (CAT) system was developed for the early therapeutic cancer response assessment in patients with locally advanced breast cancer (LABC) treated with neoadjuvant chemotherapy. The proposed CAT system was based on multi-parametric quantitative ultrasound (QUS) spectroscopic methods in conjunction with advanced machine learning techniques. Specifically, a kernel-based metric named maximum mean discrepancy (MMD), a technique for learning from imbalanced data based on random undersampling, and supervised learning were investigated with response-monitoring data from LABC patients. The CAT system was tested on 56 patients using statistical significance tests and leave-one-subject-out classification techniques. Textural features using state-of-the-art local binary patterns (LBP), and gray-scale intensity features were extracted from the spectral parametric maps in the proposed CAT system. The system indicated significant differences in changes between the responding and non-responding patient populations as well as high accuracy, sensitivity, and specificity in discriminating between the two patient groups early after the start of treatment, i.e., on weeks 1 and 4 of several months of treatment. The proposed CAT system achieved an accuracy of 85%, 87%, and 90% on weeks 1, 4 and 8, respectively. The sensitivity and specificity of developed CAT system for the same times was 85%, 95%, 90% and 85%, 85%, 91%, respectively. The proposed CAT system thus establishes a noninvasive framework for monitoring cancer treatment response in tumors using clinical ultrasound imaging in conjunction with machine learning techniques. Such a framework can potentially facilitate the detection of refractory responses in patients to treatment early on during a course of therapy to enable possibly switching to more efficacious treatments.

An efficient fractal method for detection and diagnosis of breast masses in mammograms

In this paper, we present an efficient fractal method for detection and diagnosis of mass lesion in mammogram which is one of the abnormalities in mammographic images. We used 110 images that were carefully selected by a radiologist, and their abnormalities were also confirmed by biopsy. These images included circumscribed benign, ill-defined, and spiculated malignant masses. Firstly, we discriminated lesions automatically using new fractal dimensions. The results which were examined by different types of breast density showed that the proposed method was able to yield quite satisfactory detection results. Secondly, noting that contours of masses playing the most important role in diagnosis of different mass types, we defined new fractal features based on information extraction from the contours. This information is able to identify the roughness in mass contours and determines the extent of spiculation or smoothness of the masses. In this manner, in classification of the spiculated malignant masses from the circumscribed benign tumors, we achieved highly satisfactory results, i.e., 0.98 measured in terms of area under ROC curve (AUC). In this paper, it is also shown that the roughness in contours is a suitable characteristic feature for diagnosis of ill-defined malignant tumors with AUC equal to 0.94 in their classification. The extracted information was also found to be useful in the classification of early malignancies whereas in the classification of spiculated and ill-defined malignant masses in their early stage from those of benign tumors, we achieved high accuracy of 0.99 and 0.90 for AUC, respectively.

Ultrasound RF time series for classification of breast lesions

This work reports the use of ultrasound radio frequency (RF) time series analysis as a method for ultrasound-based classification of malignant breast lesions. The RF time series method is versatile and requires only a few seconds of raw ultrasound data with no need for additional instrumentation. Using the RF time series features, and a machine learning framework, we have generated malignancy maps, from the estimated cancer likelihood, for decision support in biopsy recommendation. These maps depict the likelihood of malignancy for regions of size 1 mm(2) within the suspicious lesions. We report an area under receiver operating characteristics curve of 0.86 (95% confidence interval [CI]: 0.84%-0.90%) using support vector machines and 0.81 (95% CI: 0.78-0.85) using Random Forests classification algorithms, on 22 subjects with leave-one-subject-out cross-validation. Changing the classification method yielded consistent results which indicates the robustness of this tissue typing method. The findings of this report suggest that ultrasound RF time series, along with the developed machine learning framework, can help in differentiating malignant from benign breast lesions, subsequently reducing the number of unnecessary biopsies after mammography screening.

Classification of breast tumors using sonographic texture analysis

OBJECTIVES

The purpose of this study was to evaluate a computer-aided diagnostic system with texture analysis to improve radiologists' accuracy in identification of breast tumors as malignant or benign.

METHODS

The database included 20 benign and 12 malignant tumors. We extracted 300 statistical texture features as descriptors for each selected region of interest in 3 normalization schemes (default, μ - 3σ, and μ + 3σ, where μ and σ were the mean value and standard deviation, respectively, of the gray-level intensity and 1%-99%). Then features determined by the Fisher coefficient and the lowest probability of classification error + average correlation coefficient yielded the 10 best and most effective features. We analyzed these features under 2 standardization states (standard and nonstandard). For texture analysis of the breast tumors, we applied principle component, linear discriminant, and nonlinear discriminant analyses. First-nearest neighbor classification was performed for the features resulting from the principle component and linear discriminant analyses. Nonlinear discriminant analysis features were classified by an artificial neural network. Receiver operating characteristic curve analysis was used for examining the performance of the texture analysis methods.

RESULTS

Standard feature parameters extracted by the Fisher coefficient under the default and 3σ normalization schemes via nonlinear discriminant analysis showed high performance for discrimination between benign and malignant tumors, with sensitivity of 94.28%, specificity of 100%, accuracy of 97.80%, and an area under the receiver operating characteristic curve of 0.9714.

CONCLUSIONS

Texture analysis is a reliable method and has the potential to be used effectively for classification of benign and malignant tumors on breast sonography.

Segmentation of MR images by using grow and learn network on FPGAs

Image segmentation is one of the mostly used procedures in the medical image processing applications. Due to the high resolution characteristics of the medical images and a large amount of computational load in mathematical methods, medical image segmentation process has an excessive computational complexity. Recently, FPGA implementation has been applied in many areas due to its parallel processing capability. In this study, neighbor-pixel-intensity based method for feature extraction and Grow and Learn (GAL) network for segmentation process are proposed. The proposed method is comparatively examined on both PC and FPGA platforms.

Breast ultrasound image classification based on multiple-instance learning

Breast ultrasound (BUS) image segmentation is a very difficult task due to poor image quality and speckle noise. In this paper, local features extracted from roughly segmented regions of interest (ROIs) are used to describe breast tumors. The roughly segmented ROI is viewed as a bag. And subregions of the ROI are considered as the instances of the bag. Multiple-instance learning (MIL) method is more suitable for classifying breast tumors using BUS images. However, due to the complexity of BUS images, traditional MIL method is not applicable. In this paper, a novel MIL method is proposed for solving such task. First, a self-organizing map is used to map the instance space to the concept space. Then, we use the distribution of the instances of each bag in the concept space to construct the bag feature vector. Finally, a support vector machine is employed for classifying the tumors. The experimental results show that the proposed method can achieve better performance: the accuracy is 0.9107 and the area under receiver operator characteristic curve is 0.96 (p < 0.005).

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.

Self-organized crystallization patterns from evaporating droplets of common wheat grain leakages as a potential tool for quality analysis

We studied the evaporation-induced pattern formation in droplets of common wheat kernel leakages prepared out of ancient and modern wheat cultivars as a possible tool for wheat quality analysis. The experiments showed that the substances which passed into the water during the soaking of the kernels created crystalline structures with different degrees of complexity while the droplets were evaporating. The forms ranged from spots and simple structures with single ramifications, through dendrites, up to highly organized hexagonal shapes and fractal-like structures. The patterns were observed and photographed using dark field microscopy in small magnifications. The evaluation of the patterns was performed both visually and by means of the fractal dimension analysis. From the results, it can be inferred that the wheat cultivars differed in their pattern-forming capacities. Two of the analyzed wheat cultivars showed poor pattern formation, whereas another two created well-formed and complex patterns. Additionally, the wheat cultivars were analyzed for their vigor by means of the germination test and measurement of the electrical conductivity of the grain leakages. The results showed that the more vigorous cultivars also created more complex patterns, whereas the weaker cultivars created predominantly poor forms. This observation suggests a correlation between the wheat seed quality and droplet evaporation patterns.

Automatic detection and classification of breast tumors in ultrasonic images using texture and morphological features

Due to severe presence of speckle noise, poor image contrast and irregular lesion shape, it is challenging to build a fully automatic detection and classification system for breast ultrasonic images. In this paper, a novel and effective computer-aided method including generation of a region of interest (ROI), segmentation and classification of breast tumor is proposed without any manual intervention. By incorporating local features of texture and position, a ROI is firstly detected using a self-organizing map neural network. Then a modified Normalized Cut approach considering the weighted neighborhood gray values is proposed to partition the ROI into clusters and get the initial boundary. In addition, a regional-fitting active contour model is used to adjust the few inaccurate initial boundaries for the final segmentation. Finally, three textures and five morphologic features are extracted from each breast tumor; whereby a highly efficient Affinity Propagation clustering is used to fulfill the malignancy and benign classification for an existing database without any training process. The proposed system is validated by 132 cases (67 benignancies and 65 malignancies) with its performance compared to traditional methods such as level set segmentation, artificial neural network classifiers, and so forth. Experiment results show that the proposed system, which needs no training procedure or manual interference, performs best in detection and classification of ultrasonic breast tumors, while having the lowest computation complexity.

Computer-aided classification of breast masses: performance and interobserver variability of expert radiologists versus residents

PURPOSE

To evaluate the interobserver variability in descriptions of breast masses by dedicated breast imagers and radiology residents and determine how any differences in lesion description affect the performance of a computer-aided diagnosis (CAD) computer classification system.

MATERIALS AND METHODS

Institutional review board approval was obtained for this HIPAA-compliant study, and the requirement to obtain informed consent was waived. Images of 50 breast lesions were individually interpreted by seven dedicated breast imagers and 10 radiology residents, yielding 850 lesion interpretations. Lesions were described with use of 11 descriptors from the Breast Imaging Reporting and Data System, and interobserver variability was calculated with the Cohen κ statistic. Those 11 features were selected, along with patient age, and merged together by a linear discriminant analysis (LDA) classification model trained by using 1005 previously existing cases. Variability in the recommendations of the computer model for different observers was also calculated with the Cohen κ statistic.

RESULTS

A significant difference was observed for six lesion features, and radiology residents had greater interobserver variability in their selection of five of the six features than did dedicated breast imagers. The LDA model accurately classified lesions for both sets of observers (area under the receiver operating characteristic curve = 0.94 for residents and 0.96 for dedicated imagers). Sensitivity was maintained at 100% for residents and improved from 98% to 100% for dedicated breast imagers. For residents, the computer model could potentially improve the specificity from 20% to 40% (P < .01) and the κ value from 0.09 to 0.53 (P < .001). For dedicated breast imagers, the computer model could increase the specificity from 34% to 43% (P = .16) and the κ value from 0.21 to 0.61 (P < .001).

CONCLUSION

Among findings showing a significant difference, there was greater interobserver variability in lesion descriptions among residents; however, an LDA model using data from either dedicated breast imagers or residents yielded a consistently high performance in the differentiation of benign from malignant breast lesions, demonstrating potential for improving specificity and decreasing interobserver variability in biopsy recommendations.

Ultrasonic Nakagami imaging: a strategy to visualize the scatterer properties of benign and malignant breast tumors

Previous studies have demonstrated the usefulness of the Nakagami parameter in characterizing breast tumors by ultrasound. However, physicians or radiologists may need imaging tools in a clinical setting to visually identify the properties of breast tumors. This study proposed the ultrasonic Nakagami image to visualize the scatterer properties of breast tumors and then explored its clinical performance in classifying benign and malignant tumors. Raw data of ultrasonic backscattered signals were collected from 100 patients (50 benign and 50 malignant cases) using a commercial ultrasound scanner with a 7.5 MHz linear array transducer. The backscattered signals were used to form the B-scan and the Nakagami images of breast tumors. For each tumor, the average Nakagami parameter was calculated from the pixel values in the region-of-interest in the Nakagami image. The receiver operating characteristic (ROC) curve was used to evaluate the clinical performance of the Nakagami image. The results showed that the Nakagami image shadings in benign tumors were different from those in malignant cases. The average Nakagami parameters for benign and malignant tumors were 0.69 +/- 0.12 and 0.55 +/- 0.12, respectively. This means that the backscattered signals received from malignant tumors tend to be more pre-Rayleigh distributed than those from benign tumors, corresponding to a more complex scatterer arrangement or composition. The ROC analysis showed that the area under the ROC curve was 0.81 +/- 0.04 and the diagnostic accuracy was 82%, sensitivity was 92% and specificity was 72%. The results showed that the Nakagami image is useful to distinguishing between benign and malignant breast tumors.

A fast automatic recognition and location algorithm for fetal genital organs in ultrasound images

Severe sex ratio imbalance at birth is now becoming an important issue in several Asian countries. Its leading immediate cause is prenatal sex-selective abortion following illegal sex identification by ultrasound scanning. In this paper, a fast automatic recognition and location algorithm for fetal genital organs is proposed as an effective method to help prevent ultrasound technicians from unethically and illegally identifying the sex of the fetus. This automatic recognition algorithm can be divided into two stages. In the 'rough' stage, a few pixels in the image, which are likely to represent the genital organs, are automatically chosen as points of interest (POIs) according to certain salient characteristics of fetal genital organs. In the 'fine' stage, a specifically supervised learning framework, which fuses an effective feature data preprocessing mechanism into the multiple classifier architecture, is applied to every POI. The basic classifiers in the framework are selected from three widely used classifiers: radial basis function network, backpropagation network, and support vector machine. The classification results of all the POIs are then synthesized to determine whether the fetal genital organ is present in the image, and to locate the genital organ within the positive image. Experiments were designed and carried out based on an image dataset comprising 658 positive images (images with fetal genital organs) and 500 negative images (images without fetal genital organs). The experimental results showed true positive (TP) and true negative (TN) results from 80.5% (265 from 329) and 83.0% (415 from 500) of samples, respectively. The average computation time was 453 ms per image.

Differentiation of urinary stone and vascular calcifications on non-contrast CT images: an initial experience using computer aided diagnosis

The purpose of this study was to develop methods for the differentiation of urinary stones and vascular calcifications using computer-aided diagnosis (CAD) of non-contrast computed tomography (CT) images. From May 2003 to February 2004, 56 patients that underwent a pre-contrast CT examination and subsequently diagnosed as ureter stones were included in the study. Fifty-nine ureter stones and 53 vascular calcifications on pre-contrast CT images of the patients were evaluated. The shapes of the lesions including disperseness, convex hull depth, and lobulation count were analyzed for patients with ureter stones and vascular calcifications. In addition, the internal textures including edge density, skewness, difference histogram variation (DHV), and the gray-level co-occurrence matrix moment were also evaluated for the patients. For evaluation of the diagnostic accuracy of the shape and texture features, an artificial neural network (ANN) and receiver operating characteristics curve (ROC) analyses were performed. Of the several shape factors, disperseness showed a statistical difference between ureter stones and vascular calcifications (p < 0.05). For the internal texture features, skewness and DHV showed statistical differences between ureter stones and vascular calcifications (p < 0.05). The performance of the ANN was evaluated by examining the area under the ROC curves (AUC, A (z)). The A (z) value was 0.85 for the shape parameters and 0.88 for the texture parameters. In this study, several parameters regarding shape and internal texture were statistically different between ureter stones and vascular calcifications. The use of CAD would make it possible to differentiate ureter stones from vascular calcifications by a comparison of these parameters.

3D mutifractal analysis: a new tool for epileptic fit sources detection in SPECT images

One of the imaging modalities used for the diagnosis of epilepsy is SPECT (Single-Photon Emission Computed Tomography). Ictal and interictal images are registered to MR images (SISCOM (Substracted Ictal Spect COregistred to MR) to delineate the sources. However, in some cases and for many reasons, the used method does not lead to precise delimitation of epileptic fit sources. In this case, works have been investigated on group's studies or in combining others modalities like EEG (Electroencephalography). This study investigates the possibility of using a mathematic model for the image texture to detect the changes on SPECT images. Beyond encouraging preliminary results concerning the multifractal analysis to distinguish volunteers and epileptic patients, our aim was to detect sources by the singularity spectrum compute. The experiment is divided into two phases. First, we developed a 3D method for the singularity spectrum compute. In the test phase, we applied this multifractal spectrum to the sources detection on SPECT images. The results obtained on a base of seven patients show that the proposed method is encouraging. Indeed, the detections of epileptic fit sources obtained were in agree with the expert diagnostic.

Quantitative analysis of invasive front in tongue cancer using ultrasonography

PURPOSE

Ultrasonography is a useful diagnostic modality for the head and neck region. The mode of invasion is one of the predictive factors for the cervical lymph node metastasis. The purpose of this study is to evaluate the correlation between quantified invasive fronts from ultrasonographic images and the pathological malignancy grading of tongue squamous cell carcinoma.

MATERIALS AND METHODS

Forty-eight previously untreated T1-2N0 tongue squamous cell carcinoma patients were enrolled. The ultrasonographic images of intraoral lesions and the excised lesions were collected. The invasive front on the ultrasonographic image was analyzed by a multipurpose software package. The length of the invasive front and the line of smoothed invasive front were measured and the invasive front ratio was calculated. The hematoxylin and eosin stained sections were assessed according to modified Jakobsson's malignancy grading as low and high malignancy grade groups. The correlation between invasive front ratio and the histological malignancy grade groups was evaluated.

RESULTS

The mean invasive front ratio for the excised lesions was 1.145 in low malignancy grade group and 1.248 in high malignancy grade group. These values as determined for the intraoral lesions were 1.074 in the low malignancy grade group and 1.174 in high malignancy grade group. Significant statistical differences were observed between the 2 groups by Mann-Whitney's U test in both images.

CONCLUSIONS

Invasive front ratio is related to the histological malignancy grade group. It can be considered to be a valuable diagnostic method which can predict neck node metastasis, and help surgeons to choose adequate neck treatment.

Texture analysis of aggressive and nonaggressive lung tumor CE CT images

This paper presents the potential for fractal analysis of time sequence contrast-enhanced (CE) computed tomography (CT) images to differentiate between aggressive and nonaggressive malignant lung tumors (i.e., high and low metabolic tumors). The aim is to enhance CT tumor staging prediction accuracy through identifying malignant aggressiveness of lung tumors. As branching of blood vessels can be considered a fractal process, the research examines vascularized tumor regions that exhibit strong fractal characteristics. The analysis is performed after injecting 15 patients with a contrast agent and transforming at least 11 time sequence CE CT images from each patient to the fractal dimension and determining corresponding lacunarity. The fractal texture features were averaged over the tumor region and quantitative classification showed up to 83.3% accuracy in distinction between advanced (aggressive) and early-stage (nonaggressive) malignant tumors. Also, it showed strong correlation with corresponding lung tumor stage and standardized tumor uptake value of fluorodeoxyglucose as determined by positron emission tomography. These results indicate that fractal analysis of time sequence CE CT images of malignant lung tumors could provide additional information about likely tumor aggression that could potentially impact on clinical management decisions in choosing the appropriate treatment procedure.

Breast mass lesions: computer-aided diagnosis models with mammographic and sonographic descriptors

PURPOSE

To retrospectively develop and evaluate computer-aided diagnosis (CAD) models that include both mammographic and sonographic descriptors.

MATERIALS AND METHODS

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.

RESULTS

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).

CONCLUSION

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.