Prospective assessment of breast lesions AI classification model based on ultrasound dynamic videos and ACR BI-RADS characteristics.

Introduction: AI-assisted ultrasound diagnosis is considered a fast and accurate new method that can reduce the subjective and experience-dependent nature of handheld ultrasound. In order to meet clinical diagnostic needs better, we first proposed a breast lesions AI classification model based on ultrasound dynamic videos and ACR BI-RADS characteristics (hereafter, Auto BI-RADS). In this study, we prospectively verify its performance. Methods: In this study, the model development was based on retrospective data including 480 ultrasound dynamic videos equivalent to 18122 static images of pathologically proven breast lesions from 420 patients. A total of 292 breast lesions ultrasound dynamic videos from the internal and external hospital were prospectively tested by Auto BI-RADS. The performance of Auto BI-RADS was compared with both experienced and junior radiologists using the DeLong method, Kappa test, and McNemar test. Results: The Auto BI-RADS achieved an accuracy, sensitivity, and specificity of 0.87, 0.93, and 0.81, respectively. The consistency of the BI-RADS category between Auto BI-RADS and the experienced group (Kappa:0.82) was higher than that of the juniors (Kappa:0.60). The consistency rates between Auto BI-RADS and the experienced group were higher than those between Auto BI-RADS and the junior group for shape (93% vs. 80%; P = .01), orientation (90% vs. 84%; P = .02), margin (84% vs. 71%; P = .01), echo pattern (69% vs. 56%; P = .001) and posterior features (76% vs. 71%; P = .0046), While the difference of calcification was not significantly different. Discussion: In this study, we aimed to prospectively verify a novel AI tool based on ultrasound dynamic videos and ACR BI-RADS characteristics. The prospective assessment suggested that the AI tool not only meets the clinical needs better but also reaches the diagnostic efficiency of experienced radiologists.

Breast ultrasound lesion classification based on image decomposition and transfer learning.

PURPOSE: In medical image analysis, deep learning has great application potential. Discovering a method for extracting valuable information from medical images and integrating that information closely with medical treatment has recently become a major topic of interest. Because obtaining large volumes of breast lesion ultrasound image data is difficult, transfer learning is usually employed to obtain benign and malignant classification of breast lesions. However, because of blurred unclear regions of interest in breast lesion ultrasound images and severe speckle noise interference, convolutional neural networks have proven ineffective in extracting features, thus providing unreliable classification results. METHODS: This study employs image decomposition to obtain fuzzy enhanced and bilateral filtered images to enrich input information of breast lesions. Fuzzy enhanced, bilateral filtered, and original ultrasound images comprise multifeature data, which are presented as inputs to a pre-trained model to realize knowledge fusion. Therefore, effective features of breast lesions are extracted and then used to train fully connected layers with ground truths provided by a doctor to accomplish the classification. RESULTS: A pre-trained VGG16 model was used to extract features from multifeature data, and these features were fused to train the fully connected layers to realize classification. The performance score reported is as follows: accuracy of 93%, sensitivity of 95%, specificity of 88%, F1 score of 0.93, and AUC of 0.97. CONCLUSIONS: Compared with using a single original ultrasound image for feature extraction, multifeature data based on image decomposition enables the pre-trained model to extract more relevant features, thereby providing better classification results than those from traditional transfer learning techniques.

Cardiac VFM visualization and analysis based on YOLO deep learning model and modified 2D continuity equation.

In order to realize the visual analysis of cardiac fluid motion, according to the characteristics of cardiac flow field ultrasound image, a method for the cardiac Vector Flow Mapping (VFM) analysis and evaluation based on the You-Only-Look-Once (YOLO) deep learning model and the improved two-dimensional continuity equation is proposed in this paper. Firstly, based on the ultrasound Doppler data, the radial velocity values of the blood particles are obtained; due to the real-time VFM's high requirement on the computing speed, the YOLO deep learning model is combined with an improved block matching algorithm for the localization and tracking of myocardial wall, and then the azimuth velocity of myocardial wall speckles can be obtained; in addition, it is proposed in this paper to use a nonlinear weight function to fuse the radial velocity of the blood particles and azimuth velocity of myocardial wall speckles nonlinearly, and further the vortex streamline diagram in the cardiac flow field can be obtained. The results of the experiments on the evaluation of the Ultrasonic apical long-axis view show that the proposed method not only improves the accuracy of VFM, but also provides a new evaluation basis for cardiac function impairment.

A fuzzy clustering based color-coded diagram for effective illustration of blood perfusion parameters in contrast-enhanced ultrasound videos.

BACKGROUND AND OBJECTIVE: Early identification and diagnosis of tumors are of great significance to improve the survival rate of patients. Amongst other techniques, contrast-enhanced ultrasound is an important means to help doctors diagnose tumors. Due to the advantages of high efficiency, accuracy and objectivity, more and more computer-aided methods are used in medical diagnosis. Here we propose, a color-coded diagram based on quantitative blood perfusion parameters for contrast-enhanced ultrasound video. The method realizes the static description of the dynamic blood perfusion process in contrast-enhanced ultrasound videos and reveal the blood perfusion characteristics of all regions of the tissue providing assistance to the doctors in their clinical diagnosis. METHODS: For effective illustration of the blood perfusion through tissues, we propose (a) an improved block matching algorithm to eliminate the image distortions caused by breathing; (b) compute the time-grayscale intensity curve for each pixel to obtain four different quantitative blood perfusion parameters; and finally (c) employ the fuzzy C-means clustering algorithm to cluster the blood perfusion parameters, where each parameter is associated with a particular color. Thus based on the correspondence between the pixel and the blood perfusion parameters, all the pixels are color-coded to obtain the color-coded diagram. RESULTS: To the best of our knowledge, the proposed technique is one-of-its-kind to color code the contrast-enhanced ultrasound videos using blood perfusion parameters in order to understand the hemodynamic characteristics of the benign and malignant lesion. In our experiments, various contrast-enhanced ultrasound videos corresponding to several real-world cases were color-coded and the results of the experiments illustrated that the proposed color-coded diagrams are consistent with the diagnosis presented by the physicians. CONCLUSIONS: The experimental results suggested that the proposed method can comprehensively describe the blood perfusion characteristics of tissues during the angiography process thereby effectively assisting the doctors in diagnosis.

An RDAU-NET model for lesion segmentation in breast ultrasound images.

Breast cancer is a common gynecological disease that poses a great threat to women health due to its high malignant rate. Breast cancer screening tests are used to find any warning signs or symptoms for early detection and currently, Ultrasound screening is the preferred method for breast cancer diagnosis. The localization and segmentation of the lesions in breast ultrasound (BUS) images are helpful for clinical diagnosis of the disease. In this paper, an RDAU-NET (Residual-Dilated-Attention-Gate-UNet) model is proposed and employed to segment the tumors in BUS images. The model is based on the conventional U-Net, but the plain neural units are replaced with residual units to enhance the edge information and overcome the network performance degradation problem associated with deep networks. To increase the receptive field and acquire more characteristic information, dilated convolutions were used to process the feature maps obtained from the encoder stages. The traditional cropping and copying between the encoder-decoder pipelines were replaced by the Attention Gate modules which enhanced the learning capabilities through suppression of background information. The model, when tested with BUS images with benign and malignant tumor presented excellent segmentation results as compared to other Deep Networks. A variety of quantitative indicators including Accuracy, Dice coefficient, AUC(Area-Under-Curve), Precision, Sensitivity, Specificity, Recall, F1score and M-IOU (Mean-Intersection-Over-Union) provided performances above 80%. The experimental results illustrate that the proposed RDAU-NET model can accurately segment breast lesions when compared to other deep learning models and thus has a good prospect for clinical diagnosis.

Application of fractal theory and fuzzy enhancement in ultrasound image segmentation.

The manuscript describes an ultrasound image segmentation technique based on the fractional Brownian motion (FBM) model. Here, the ultrasound images are first enhanced using a fuzzy-based technique, and later the FBM model is employed to obtain the fractal features used for segmentation. The novelty lies in combining the fuzzy-enhancement technique and FBM model, and further illustrating that fractal length-based segmentation provides better results than fractal dimension-based segmentation. Experimental results on ultrasound images of carotid artery clearly illustrate that the segmentation outputs obtained from fractal length are superior, and the high qualitative values of DSC, Precision, Recall and F1 score (0.9617, 0.9629, 0.9653 and 0.9641 respectively), together with a low value of APD (1.9316), indicate that the proposed method is comparable to other state-of-the-art segmentation techniques. Graphical abstract Summary of proposed technique - overall design flow.