Automatic breast tumor detection in ABVS images based on convolutional neural network and superpixel patterns

We investigated the accuracy of Automated Breast Volume Scanner (ABVS) compared to handheld ultrasound (HHUS) for monitoring tumor response to neoadjuvant treatment (NAT) in breast cancer (BC). All the patients submitted to biopsy in our Institution, from January 2017 to May 2017, proven invasive BC and eligible for NAT, were enrolled in this prospective study. The participants underwent ABVS, HHUS, dynamic contrast-enhanced Magnetic Resonance Imaging (DCE-MRI) and mammography at the beginning of NAT and ABVS, HHUS and DCE-MRI at the halfway point of therapy and before the surgery. DCE-MRI was considered the standard of reference. Two breast radiologists (R1, R2), with fifteen and five years of experience in breast imaging, independently assigned a visibility score (ordinal 5-point scale) to ABVS, HHUS, and DCE-MRI. Diagnostic performance of ABVS and HHUS as measured by sensitivity, specificity, positive and negative predictive values (PPV and NPV) was calculated. Correlations between ABVS and MRI, and between HHUS and MRI were analyzed using Pearson's correlation test. A total of 21 patients were enrolled. 189 examinations were performed. The comparison between ABVS and DCE-MRI was similar for the both readers: ABVS had a sensitivity of 63,16%, specificity of 83,58%, PPV of 76,60%, NPV of 72,73%, accuracy of 74,19% (R1) and a sensitivity of 54.54%, specificity of 85.51%, PPV of 75%, NPV of 70,24%, accuracy of 71.77% (R2). The comparison between HHUS and DCE-MRI showed that HHUS had a sensitivity of 63,16 %, specificity of 83,58%, PPV of 76,60%, NPV of 72,73%, accuracy of 74,19% (R1) and a sensitivity of 36.84%, specificity of 85.07%, PPV of 67.74%, NPV of 61.29%, accuracy of 62.90% (R2). The calculated Pearson's correlation coefficient r values were 7.8 for HHUS vs. DCE-MRI and 28.5 for ABVS vs. DCE-MRI (R1) and 7.8 for HHUS vs. DCE-MRI and 22.4 for ABVS vs. DCE-MRI (R2). Statistical significance of ABVS and HHUS was p < 0.0001 and 0.005 < p < 0.01, respectively (R1, R2). DCE-MRI is recommended for the tumor response assessment. ABVS, a product of the biotechnology development, providing reproducible images, in addition to DCE-MRI, can be a potentially useful tool for the monitoring of response to NAT.

BackgroundAutomated whole breast ultrasound scanners of the latest generation have reached a level of comfortable application and high quality volume acquisition. Nevertheless, there is a lack of data concerning this technology. We investigated the diagnostic performance and inter-observer concordance of the Automated Breast Volume Scanner (ABVS) ACUSON S2000™ and questioned its implications in breast cancer diagnostics.MethodsWe collected 100 volume data sets and created a database containing 52 scans with no detectable lesions in conventional ultrasound (BI-RADS®-US 1), 30 scans with benign lesions (BI-RADS®-US 2) and 18 scans with breast cancer (BI-RADS®-US 5).Two independent examiners evaluated the ABVS data on a separate workstation without any prior knowledge of the patients’ histories.ResultsThe inter-rater reliability reached fair agreement (κ=0.36; 95% confidence interval (CI): 0.19-0.53). With respect to the true category, the conditional inter-rater validity coefficient was κ=0.18 (95% CI: 0.00-0.26) for the benign cases and κ=0.80 (95% CI: 0.61-1.00) for the malignant cases.Combining the assessments of examiner 1 and examiner 2, the diagnostic accuracy (AC), sensitivity (SE) and specificity (SP) for the experimental ABVS were AC = 79.0% (95% CI: 67.3-86.1), SE = 83.3% (95% CI: 57.7-95.6) and SP = 78.1% (% CI: 67.3-86.1), respectively.However, after the ABVS examination, there were a high number of requests for second-look ultrasounds in up to 48.8% of the healthy women due to assumed suspicious findings in the volume data.In an exploratory analysis, we estimated that an ABVS examination in addition to mammography alone could detect a relevant number of previously occult breast cancers (about 1 cancer in 300 screened and otherwise healthy women).ConclusionsThe ABVS is a reliable imaging method for the evaluation of the breast with high sensitivity and a fair inter-observer concordance. However, we have to overcome the problem of the high number of false-positive results. Therefore, further prospective studies in larger collectives are necessary to define standard procedures in image acquisition and interpretation. Nevertheless, we consider the ABVS as being suitable for integration into breast diagnostics as a beneficial and reliable imaging method.

Breast cancer has become the primary factor threatening women's health. Automated breast volume scanner (ABVS) is applied for automatic scanning which is meaningful for the rapid and accurate detection of breast tumor. However, accurate segmentation of tumor regions is a huge challenge for clinicians from the ABVS images since it has the large image size and low data quality. Therefore, we propose a novel 3D deep convolutional neural network for automatic breast tumor segmentation from ABVS data. The structure based on 3D U-Net is designed with attention mechanism and transformer layers to optimize the extracted image features. In addition, we integrate the atrous spatial pyramid pooling block and the deep supervision for further performance improvement. The experimental results demonstrate that our model has achieved dice coefficient of 76.36% for 3D segmentation of breast tumor via our self-collected data.

The objective of this study was to evaluate the performance of a computer-aided detection (CAD) system for the detection of breast cancer, based on mammographic appearance and histopathology. From 1000 consecutive screening mammograms from women with biopsy-proven breast carcinoma, 273 mammograms were selected randomly for retrospective evaluation by CAD. The sensitivity of the CAD system for breast cancer was assessed from the proportion of masses and microcalcifications detected. The corresponding tumor histopathologies also were evaluated. Normal mammograms (n = 155 patients) were used to determine the false-positive rate of the system. Of the 273 breast carcinomas, 149 appeared mammographically as masses, and 88 appeared as microcalcifications, including 36 carcinomas that presented as mixed lesions. The CAD system marked 125 of 149 masses correctly (84%), marked 86 of 88 microcalcifications correctly (98%), and marked 32 of 36 of mixed lesions correctly (89%.). The system showed a high sensitivity for the detection of ductal carcinoma in situ (95%; 73 of 77 lesions), invasive lobular carcinoma (95%; 18 of 19 lesions), invasive ductal carcinoma (85%; 125 of 147 lesions), and invasive mammary carcinoma (90%; 27 of 30 lesions). The highest CAD system sensitivity was for all invasive carcinomas that presented as microcalcifications (100%). On normal mammograms, there was an average of 1.3 false-positive CAD marks per image. The CAD system correctly marked a large majority of biopsy-proven breast cancers, with a greater sensitivity for lesions with microcalcifications and without significant impact of performance based on tumor histopathology. CAD was highly effective in detecting invasive lobular carcinoma (sensitivity, 95%) and ductal carcinoma in situ (sensitivity, 95%). CAD represents a useful tool for the detection of breast cancer.

Computer-Aided Mass Detection Based on Ipsilateral Multiview Mammograms

Effect of CAD on Radiologists' Detection of Lung Nodules on Thoracic CT Scans: Analysis of an Observer Performance Study by Nodule Size

Artificial intelligence in breast imaging

To explore the values of automated breast volume scanning (ABVS) combined with shear wave elastography (SWE) in the differential diagnosis of triple-negative breast cancer (TNBC) and human epidermal growth factor receptor 2-positive breast cancers (HER2+BC). In this study, 28 patients with TNBC and 32 patients with HER2+BC were enrolled. The characteristics of ABVS and virtual touch quantification (VTQ) in SWE of all patients were reviewed. The multivariate logistic regression analysis was carried out and the receiver operating characteristic curves of ABVS and ABVS+VTQ were drawn. In ABVS imaging, the microcalcification, posterior echo, internal echo, shape, and edge had significant difference between TNBC and HER2+BC groups (p<0.05). The regular shape was the independent factor for TNBC (p=0.04, odds ratio [OR]=4.479), and the microcalcification in mass was the independent factor for HER2+BC (p=0.01, OR=2.997). In VTQ imaging, the shear wave velocity (SWV)max, SWVmin, and SWVmean in TNBC group were significantly lower than those in HER2+BC group (p<0.001). The sensitivity, specificity, and accuracy of ABVS+VTQ in diagnosing TNBC were higher than those of ABVS alone. ABVS combined with SWE has certain advantages in differentiating TNBC from HER2+BC, which is helpful for the treatment planning and prognosis judgment.

The automated breast volume scanner (ABVS) is used for breast cancer detection and consists of a linear transducer translating over the breast while collecting ultrasound data to reconstruct a breast volume. Although the ABVS shows high sensitivity, specificity remains a limitation resulting in high recall-rates. To improve ABVS specificity, we verified if it is feasible to implement 3-D strain imaging to discriminate between malignant (firmly-bound and stiff) and benign lesions (loosely bound and less stiff) based on lesion-to-surrounding-tissue connectivity (shear strain) and on stiffness (axial strain). Three phantoms containing loosely- (1) and firmly-bound (2) three times stiffer lesions were scanned twice by the ABVS, pre- and post-deformation (0.5 mm). Data was collected by plane-wave imaging to reduce acquisition times below one breath-hold. Displacements were calculated by cross-correlation from which axial and octahedral-shear strains were derived. Shear strains were distributed tightly and globally around the lesion for phantom (1) and (2), respectively. Using axial strain, the lesions could be clearly differentiated from the surrounding tissue in both phantoms (CNRe 30; SNRe 19dB). Therefore, we can conclude that it is feasible to implement 3-D strain imaging in the ABVS, and to improve its specificity by discriminating lesions by elastic properties.

One-stage pulmonary nodule detection using 3-D DCNN with feature fusion and attention mechanism in CT image

Lung cancer is the most common and the deadliest cancer in the world. Lung cancer staging usually was done by radiologist by detecting mediastinal lymph node (LN) enlargement. Mediastinal LN is difficult to be detected visually due to its low contrast to the surrounding tissues, various size and shape, and sparse location. Therefore, computer-aided detection (CADe) system has been developed as a tool for radiologist to detect mediastinal LN automatically. The state of the art mediastinal LN CADe system use complex architectural convolutional neural network (CNN). However, more simple architecture of the CNN is needed to reduce the computational complexity of the CADe system, especially if the system was intended to be used in a regular computer. Therefore, in this experiment we used simple architectural 2D CNN which is converted to fully convolutional network (FCN) to detect mediastinal LN candidate in a stack of CT images. Then, the mediastinal LN candidates were classified using 3D CNN to reduce the false positive (FP). The best performance of this CADe system was 65% of sensitivity at 5 FP/patient.

Development and validation of artificial neural networks model for detection of Barrett’s neoplasia: a multicenter pragmatic nonrandomized trial (with video)

We have developed a computer-aided detection (CAD) system for breast masses on mammograms. In this study, our purpose was to improve the performance of our mass detection system by using a new dual system approach which combines a CAD system optimized with "average" masses with another CAD system optimized with subtle masses. The latter system is trained to provide high sensitivity in detecting subtle masses. For an unknown mammogram, the two systems are used in parallel to detect suspicious objects. A feed-forward backpropagation neural network trained to merge the scores of the two linear discriminant analysis (LDA) classifiers from the two systems makes the final decision in differentiation of true masses from normal tissue. A data set of 86 patients containing 172 mammograms with biopsy-proven masses was partitioned into a training set and an independent test set. This data set is referred to as the average data set. A second data set of 214 prior mammograms was used for training the second CAD system for detection of subtle masses. When the single CAD system trained on the average data set was applied to the test set, the A_z for false positive (FP) classification was 0.81 and the FP rates were 2.1, 1.5 and 1.3 FPs/image at the case-based sensitivities of 95%, 90% and 85%, respectively. With the dual CAD system, the A_z was 0.85 and the FP rates were improved to 1.7, 1.2 and 0.8 FPs/image at the same case-based sensitivities. Our results indicate that the dual CAD system can improve the performance of mass detection on mammograms.

Physicians use sonographic characteristics as a reference for the possible diagnosis of thyroid cancers. The purpose of this study was to investigate whether physicians were more effective in their tentative diagnosis based on the information provided by a computer-aided detection (CAD) system. A computer compared software-defined and physician-adjusted tumor loci. A multicenter, multireader, and multicase (MRMC) study was designed to compare clinician performance without and with the use of CAD. Interobserver variability was also analyzed. Excellent, satisfactory, and poor segmentations were observed in 25.3%, 58.9%, and 15.8% of nodules, respectively. There were 200 patients with 265 nodules in the study set. Nineteen physicians scored the malignancy potential of the nodules. The average area under the curve (AUC) of all readers was 0.728 without CAD and significantly increased to 0.792 with CAD. The average standard deviation of the malignant potential score significantly decreased from 18.97 to 16.29. The mean malignant potential score significantly decreased from 35.01 to 31.24 for benign cases. With the CAD system, an additional 7.6% of malignant nodules would be suggested for further evaluation, and biopsy would not be recommended for an additional 10.8% of benign nodules. The results demonstrated that applying a CAD system would improve clinicians’ interpretations and lessen the variability in diagnosis. However, more studies are needed to explore the use of the CAD system in an actual ultrasound diagnostic situation where much more benign thyroid nodules would be seen.

Removed.