In vivo classification of breast masses using features derived from axial-strain and axial-shear images

The Role of a Deep Learning-Based Computer-Aided Diagnosis System and Elastography in Reducing Unnecessary Breast Lesion Biopsies

OBJECTIVES

Ultrasound examination has inter-observer and intra-observer variability and a high false-positive rate. The aim of this study was to evaluate the value of the combined use of a deep learning-based computer-aided diagnosis (CAD) system and ultrasound elastography with conventional ultrasound (US) in increasing specificity and reducing unnecessary breast lesions biopsies.

MATERIALS AND METHODS

Conventional US, CAD system, and strain elastography (SE) were retrospectively performed on 216 breast lesions before biopsy or surgery. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and biopsy rate were compared between conventional US and the combination of conventional US, SE, and CAD system.

RESULTS

Of 216 lesions, 54 were malignant and 162 were benign. The addition of CAD system and SE to conventional US increased the AUC from 0.716 to 0.910 and specificity from 46.9% to 85.8% without a loss in sensitivity while 89.2% (66 of 74) of benign lesions in Breast Imaging Reporting and Data System (BI-RADS) category 4A lesions would avoid unnecessary biopsies.

CONCLUSION

The addition of CAD system and SE to conventional US improved specificity and AUC without loss of sensitivity, and reduced unnecessary biopsies.

Diagnostic Accuracy of Cross-Polarization OCT and OCT-Elastography for Differentiation of Breast Cancer Subtypes: Comparative Study

The possibility to assess molecular-biological and morphological features of particular breast cancer types can improve the precision of resection margin detection and enable accurate determining of the tumor aggressiveness, which is important for treatment selection. To enable reliable differentiation of breast-cancer subtypes and evaluation of resection margin, without performing conventional histological procedures, here we apply cross-polarization optical coherence tomography (CP-OCT) and compare it with a novel variant of compressional optical coherence elastography (C-OCE) in terms of the diagnostic accuracy (Ac) with histological verification. The study used 70 excised breast cancer specimens with different morphological structure and molecular status (Luminal A, Luminal B, Her2/Neo+, non-luminal and triple-negative cancer). Our first aim was to formulate convenient criteria of visual assessment of CP-OCT and C-OCE images intended (i) to differentiate tumorous and non-tumorous tissues and (ii) to enable more precise differentiation among different malignant states. We identified such criteria based on the presence of heterogeneities and characteristics of signal attenuation in CP-OCT images, as well as the presence of inclusions/mosaic structures combined with visually feasible assessment of several stiffness grades in C-OCE images. Secondly, we performed a blinded reader study of the Ac of C-OCE versus CP-OCT, for delineation of tumorous versus non-tumorous tissues followed by identification of breast cancer subtypes. For tumor detection, C-OCE showed higher specificity than CP-OCT (97.5% versus 93.3%) and higher Ac (96.0 versus 92.4%). For the first time, the Ac of C-OCE and CP-OCT were evaluated for differentiation between non-invasive and invasive breast cancer (90.4% and 82.5%, respectively). Furthermore, for invasive cancers, the difference between invasive but low-aggressive and highly-aggressive subtypes can be detected. For differentiation between non-tumorous tissue and low-aggressive breast-cancer subtypes, Ac was 95.7% for C-OCE and 88.1% for CP-OCT. For differentiation between non-tumorous tissue and highly-aggressive breast cancers, Ac was found to be 98.3% for C-OCE and 97.2% for CP-OCT. In all cases C-OCE showed better diagnostic parameters independently of the tumor type. These findings confirm the high potential of OCT-based examinations for rapid and accurate diagnostics during breast conservation surgery.

Diagnostic Accuracy of Quantitative Micro-Elastography for Margin Assessment in Breast-Conserving Surgery

Inadequate margins in breast-conserving surgery (BCS) are associated with an increased likelihood of local recurrence of breast cancer. Currently, approximately 20% of BCS patients require repeat surgery due to inadequate margins at the initial operation. Implementation of an accurate, intraoperative margin assessment tool may reduce this re-excision rate. This study determined, for the first time, the diagnostic accuracy of quantitative micro-elastography (QME), an optical coherence tomography (OCT)-based elastography technique that produces images of tissue microscale elasticity, for detecting tumor within 1 mm of the margins of BCS specimens. Simultaneous OCT and QME were performed on the margins of intact, freshly excised specimens from 83 patients undergoing BCS and on dissected specimens from 7 patients undergoing mastectomy. The resulting three-dimensional images (45 × 45 × 1 mm) were coregistered with postoperative histology to determine tissue types present in each scan. Data from 12 BCS patients and the 7 mastectomy patients served to build a set of images for reader training. One hundred and fifty-four subimages (10 × 10 × 1 mm) from the remaining 71 BCS patients were included in a blinded reader study, which resulted in 69.0% sensitivity and 79.0% specificity using OCT images, versus 92.9% sensitivity and 96.4% specificity using elasticity images. The quantitative nature of QME also facilitated development of an automated reader, which resulted in 100.0% sensitivity and 97.7% specificity. These results demonstrate high accuracy of QME for detecting tumor within 1 mm of the margin and the potential for this technique to improve outcomes in BCS. SIGNIFICANCE: An optical imaging technology probes breast tissue elasticity to provide accurate assessment of tumor margin involvement in breast-conserving surgery.

Modeling Uncertainty of Strain Ratio Measurements in Ultrasound Breast Strain Elastography: A Factorial Experiment

Strain elastography (SE) is a technique in which images of localized tissue strains are used to detect the relative stiffness of tissues. The application of SE in differentiating malignant breast lesions from benign ones has been broadly investigated. The strain ratio (SR) between the background and the breast tumor has been used and its results have been mixed. Due to the complex nature of tissue elasticity and how it relates to the strain fields measured in SE, the exact reason is not known. In this study, we apply a novel design-of-experiments-based metamodeling approach to mechanical simulation of SE in the human breast. To our knowledge, such a study has not been reported in the ultrasound SE literature. More specifically, we first conduct a screening study to identify the biomechanical factors/simulation inputs that most strongly determine SR. We then apply a response surface experimental design to these factors to produce a metamodel of SR as a function of said factors. Results from the screening study suggest that the SR measurements are primarily influenced by three factors: the initial shear modulus of the lesion, the elastic nonlinearity of the lesion, and the precompression applied during acquisition. In order to investigate the implications of these results, stochastic inputs for these three factors associated with the malignant and benign cases were applied to the resulting response surface. The resulting optimal cutoffs, sensitivity, and specificity were generally in line with a majority (>60%) of 19 clinical trials in the literature.

3-D Single Breath-Hold Shear Strain Estimation for Improved Breast Lesion Detection and Classification in Automated Volumetric Ultrasound Scanners

Automated breast volume scanner (ABVS) is an ultrasound imaging modality used in breast cancer screening. It has high sensitivity but limited specificity as it is hard to discriminate between benign and malignant lesions by echogenic properties. Specificity might be improved by shear strain imaging as malignant lesions, firmly bonded to its host tissue, show different shear patterns compared to benign lesions, often loosely bonded. Therefore, 3-D quasi-static elastography was implemented in an ABVS-like system. Plane wave instead of conventional focused transmissions were used to reduce scan times within a single breath hold. A 3-D strain tensor was obtained and shear strains were reconstructed in phantoms containing firmly and loosely bonded lesions. Experiments were also simulated in finite-element models (FEMs). Experimental results, confirmed by FEM-results, indicated that loosely bonded lesions showed increased maximal shear strains (~2.5%) and different shear patterns compared to firmly bonded lesions (~0.9%). To conclude, we successfully implemented 3-D elastography in an ABVS-like system to assess lesion bonding by shear strain imaging.

A Normalized Shear Deformation Indicator for Ultrasound Strain Elastography in Breast Tissues: An In Vivo Feasibility Study

The shear deformation under loads contains useful information for distinguishing benign breast lesions from malignant ones. In this study, we proposed a normalized shear deformation indicator (NSDI) that was derived from the concept of principal strains. Since the NSDI requires both high-quality axial and lateral (parallel and perpendicular to the beam, resp.) displacement estimates, a strategy combining high-quality speckle tracking with signal “denoising” was employed. Both techniques were previously published by our group. Finite element (FE) models were used to identify possible causes for elevated NSDI values in and around breast lesions, followed by an analysis of ultrasound data acquired from 26 biopsy-confirmed in vivo breast lesions. We found that, theoretically, the elevated NSDI values could be attributed to two factors: significantly hardened tissue stiffness and increasing heterogeneity. The analysis of in vivo data showed that the proposed NSDI values were higher (p < 0.05) among malignant cancers as compared to those measured from benign ones. In conclusion, our preliminary results demonstrated that the calculation of NSDI value is feasible and NSDI could add value to breast lesion differentiation with current clinical equipment as a postprocessing tool.

Comparison and Combination of Strain and Shear Wave Elastography of Breast Masses for Differentiation of Benign and Malignant Lesions by Quantitative Assessment: Preliminary Study

OBJECTIVES

To compare the diagnostic performance of strain and shear wave elastography of breast masses for quantitative assessment in differentiating benign and malignant lesions and to evaluate the diagnostic accuracy of combined strain and shear wave elastography.

METHODS

Between January and February 2016, 37 women with 45 breast masses underwent both strain and shear wave ultrasound (US) elastographic examinations. The American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) final assessment on B-mode US imaging was assessed. We calculated strain ratios for strain elastography and the mean elasticity value and elasticity ratio of the lesion to fat for shear wave elastography. Diagnostic performances were compared by using the area under the receiver operating characteristic curve (AUC).

RESULTS

The 37 women had a mean age of 47.4 years (range, 20-79 years). Of the 45 lesions, 20 were malignant, and 25 were benign. The AUCs for elasticity values on strain and shear wave elastography showed no significant differences (strain ratio, 0.929; mean elasticity, 0.898; and elasticity ratio, 0.868; P > .05). After selectively downgrading BI-RADS category 4a lesions based on strain and shear wave elastographic cutoffs, the AUCs for the combined sets of B-mode US and elastography were improved (B-mode + strain, 0.940; B-mode + shear wave; 0.964; and B-mode, 0.724; P < .001). Combined strain and shear wave elastography showed significantly higher diagnostic accuracy than each individual elastographic modality (P = .031).

CONCLUSIONS

These preliminary results showed that strain and shear wave elastography had similar diagnostic performance. The addition of strain and shear wave elastography to B-mode US improved diagnostic performance. The combination of strain and shear wave elastography results in a higher diagnostic yield than each individual elastographic modality.

Ultrasound Strain Elastography for Breast Lesions: Computer-Aided Evaluation With Quantifiable Elastographic Features

OBJECTIVES

To develop and evaluate a set of quantifiable elastographic features based on ultrasound real-time strain elastography (SE) in differentiating between benign and malignant breast lesions.

METHODS

The SE and conventional B-mode ultrasound images of 226 breast lesions (81 malignant, 145 benign) were obtained from 226 consecutive women. By using a computer-aided tool, four elastographic features (elasticity score, lesion stiffness degree, lesion-to-fat strain ratio, and elastography-to-B-mode lesion area ratio) were respectively calculated and evaluated. Histopathologic results were used as the reference standard. B-mode Breast Imaging Reporting and Data System categorization was used to compare the performances between B-mode ultrasound and SE. Sensitivity, specificity, positive and negative predictive values, and receiver operating characteristic curve analyses were performed to evaluate the diagnostic performances for three data sets (conventional B-mode ultrasound alone, SE features alone, combined SE features).

RESULTS

Quantifiable SE features for malignant lesions all showed significantly higher values than those for benign lesions (all P < .001). The evaluation with any individual SE feature significantly improved the specificity in breast lesion differentiation compared with B-mode ultrasound (all P <.001). The logistic regression model combing SE features significantly improved the diagnostic performance compared with B-mode US, with significantly increased specificity (95.2% versus 54.5%; P < .001) and area under the receiver operating characteristic curve (0.988 versus 0.921, P < .001).

CONCLUSIONS

Computer-aided tool with SE provided further elasticity information for breast characterization. Evaluation using quantifiable SE features showed better diagnostic performance than conventional B-mode ultrasound in breast lesion differentiation.

Post-Procedure Evaluation of Microwave Ablations of Hepatocellular Carcinomas Using Electrode Displacement Elastography

Microwave ablation has been used clinically as an alternative to surgical resection. However, lack of real-time imaging to assess treated regions may compromise treatment outcomes. We previously introduced electrode displacement elastography (EDE) for strain imaging and verified its feasibility in vivo on porcine animal models. In this study, we evaluated EDE on 44 patients diagnosed with hepatocellular carcinoma, treated using microwave ablation. The ablated region was identified on EDE images for 40 of the 44 patients. Ablation areas averaged 13.38 ± 4.99 cm² on EDE, compared with 7.61 ± 3.21 cm² on B-mode imaging. Contrast and contrast-to-noise ratios obtained with EDE were 232% and 98%, respectively, significantly higher than values measured on B-mode images (p < 0.001). This study indicates that EDE is feasible in patients and provides improved visualization of the ablation zone compared with B-mode ultrasound.

Semi-automatic breast ultrasound image segmentation based on mean shift and graph cuts

Computerized tumor segmentation on breast ultrasound (BUS) images remains a challenging task. In this paper, we proposed a new method for semi-automatic tumor segmentation on BUS images using Gaussian filtering, histogram equalization, mean shift, and graph cuts. The only interaction required was to select two diagonal points to determine a region of interest (ROI) on an input image. The ROI image was shrunken by a factor of 2 using bicubic interpolation to reduce computation time. The shrunken image was smoothed by a Gaussian filter and then contrast-enhanced by histogram equalization. Next, the enhanced image was filtered by pyramid mean shift to improve homogeneity. The object and background seeds for graph cuts were automatically generated on the filtered image. Using these seeds, the filtered image was then segmented by graph cuts into a binary image containing the object and background. Finally, the binary image was expanded by a factor of 2 using bicubic interpolation, and the expanded image was processed by morphological opening and closing to refine the tumor contour. The method was implemented with OpenCV 2.4.3 and Visual Studio 2010 and tested for 38 BUS images with benign tumors and 31 BUS images with malignant tumors from different ultrasound scanners. Experimental results showed that our method had a true positive rate (TP) of 91.7%, a false positive (FP) rate of 11.9%, and a similarity (SI) rate of 85.6%. The mean run time on Intel Core 2.66 GHz CPU and 4 GB RAM was 0.49 ± 0.36 s. The experimental results indicate that the proposed method may be useful in BUS image segmentation.

In Vivo response to compression of 35 breast lesions observed with a two-dimensional locally regularized strain estimation method

The objective of this study was to assess the in vivo performance of our 2-D locally regularized strain estimation method with 35 breast lesions, mainly cysts, fibroadenomas and carcinomas. The specific 2-D deformation model used, as well as the method's adaptability, led to an algorithm that is able to track tissue motion from radiofrequency ultrasound images acquired in clinical conditions. Particular attention was paid to strain estimation reliability, implying analysis of the mean normalized correlation coefficient maps. For all lesions examined, the results indicated that strain image interpretation, as well as its comparison with B-mode data, should take into account the information provided by the mean normalized correlation coefficient map. Different trends were observed in the tissue response to compression. In particular, carcinomas appeared larger in strain images than in B-mode images, resulting in a mean strain/B-mode lesion area ratio of 2.59 ± 1.36. In comparison, the same ratio was assessed as 1.04 ± 0.26 for fibroadenomas. These results are in agreement with those of previous studies, and confirm the interest of a more thorough consideration of size difference as one parameter discriminating between malignant and benign lesions.