Video sequence compared to conventional freeze image documentation: a way to improve the sonographic assessment of breast lesions?

Deep learning-based classification of breast lesions using dynamic ultrasound video

PURPOSE

We intended to develop a deep-learning-based classification model based on breast ultrasound dynamic video, then evaluate its diagnostic performance in comparison with the classic model based on ultrasound static image and that of different radiologists.

METHOD

We collected 1000 breast lesions from 888 patients from May 2020 to December 2021. Each lesion contained two static images and two dynamic videos. We divided these lesions randomly into training, validation, and test sets by the ratio of 7:2:1. Two deep learning (DL) models, namely DL-video and DL-image, were developed based on 3D Resnet-50 and 2D Resnet-50 using 2000 dynamic videos and 2000 static images, respectively. Lesions in the test set were evaluated to compare the diagnostic performance of two models and six radiologists with different seniority.

RESULTS

The area under the curve of the DL-video model was significantly higher than those of the DL-image model (0.969 vs. 0.925, P = 0.0172) and six radiologists (0.969 vs. 0.779-0.912, P < 0.05). All radiologists performed better when evaluating the dynamic videos compared to the static images. Furthermore, radiologists performed better with increased seniority both in reading images and videos.

CONCLUSIONS

The DL-video model can discern more detailed spatial and temporal information for accurate classification of breast lesions than the conventional DL-image model and radiologists, and its clinical application can further improve the diagnosis of breast cancer.

Section Discrepancy and Diagnostic Performance of Breast Lesions in Two-dimensional Ultrasound by Dynamic Videos versus Static Images

Background: Benign or malignant breast lesions with typical ultrasonic characteristics could be easily and correctly diagnosed with two-dimensional ultrasound (2D US). However, diagnosis of atypical lesions remains a challenge. Most atypical lesions have different ultrasonographic features with probe direction variation. Thus, the interpretation of ultrasonographic features based on static images empirically collected by sonographers might be inaccurate. We aimed to investigate the section discrepancy and diagnostic performance of breast lesions in 2D US by dynamic videos versus static images.Methods: Static images and dynamic videos based on two perpendicular planes of 468 breast lesions were collected and evaluated. The Breast Imaging and Reporting Data System (BI-RADS&#x00AE;) US lexicon was used. Category 3 was used as the cut-off point, and section discrepancy was defined as two perpendicular planes showing different BI-RADS categories (3 versus 4A, 4B, 4C, and 5).Results: This retrospective study included 315 benign and 153 malignant lesions. There were 53 and 50 lesions with section discrepancy during static and dynamic observations, respectively. The proportion of benign lesions with section discrepancy was significantly higher than that of malignant lesions (P &#x003C; 0.05) either in dynamic or static observation, and the contingency coefficient was 0.2 between section discrepancy and histopathology. Duct changes were more clearly depicted in dynamic videos than in static images (P &#x003C; 0.05) both in malignant and benign lesions. Calcification and architectural distortion were more sensitively detected by dynamic videos than with static images (P &#x003C; 0.05) in malignant lesions. The interpretation of &#x201C;margin&#x201D; significantly differed in benign lesions between static images and dynamic videos (P &#x003C; 0.05). The areas under the curve of static image-horizontal, static image-sagittal, dynamic video-horizontal, and dynamic video-sagittal were 0.807, 0.820, 0.837, and 0.846, respectively. The specificities of dynamic videos were higher than those of static images (P &#x003C; 0.05).Conclusion: Breast lesions have section discrepancy in 2D US. Observations based on dynamic videos could more accurately reflect lesion features and increase the specificity of US in the differentiation of atypical breast lesions.

Does the radiologist need to rescan the breast lesion to validate the final BI-RADS US assessment made on the static images in the diagnostic setting?

Purpose: To assess whether radiologist needs to rescan the breast lesion to validate the final American College of Radiology (ACR) Breast Imaging Reporting and Data System (BI-RADS) ultrasonography (US) assessment made on the static images in the diagnostic setting.

Patients and methods: Image data on 1,070 patients with 1,070 category 3–5 breast lesions with a pathological diagnosis scanned between January and June 2016 were included. Both real-time and static image assessments were acquired for each lesion. The diagnostic performance was evaluated by receiver operating characteristic (ROC) curves. The positive predictive values (PPVs) of each category in the two groups were calculated according to the ACR BI-RADS manual and compared. Kappas were determined for agreement on two assessment approaches.

Results: The sensitivity, specificity, PPV, and negative predictive value for real-time US were 98.9%, 58.2%, 44.8% and 99.4%, and for static images were 98.9%, 57.1%, 44.1% and 99.3%, respectively. The performance of the two groups was not significantly different (areas under ROCs: 0.786 vs 0.780, P=0.566) if the final assessment was only dichotomized as negative (category 3) and positive (categories 4 and 5). All PPVs of each category for each assessment were within the reference range provided by the ACR in 2013 except subcategory 4B (reference range: >10% and ≤50%) of static image evaluation, which was also significantly higher than that of real-time assessment (54.8% vs 40.7%, P=0.037). The overall agreement of the two approaches was moderate (κ=0.43–0.56 according to different detailed assessment).

Conclusion: Both static image and real-time assessment had similar diagnostic performance if only the treatment recommendations were considered, that is, follow-up or biopsy. However, as for subcategory 4B lesions without obviously benign or malignant US features, real-time scanning by the interpreter is recommended to obtain a more accurate BI-RADS assessment after assessing static images.

Comparison of Inter-Observer Variability and Diagnostic Performance of the Fifth Edition of BI-RADS for Breast Ultrasound of Static versus Video Images

Our aim was to compare the inter-observer variability and diagnostic performance of the Breast Imaging Reporting and Data System (BI-RADS) lexicon for breast ultrasound of static and video images. Ninety-nine breast masses visible on ultrasound examination from 95 women 19-81 y of age at five institutions were enrolled in this study. They were scheduled to undergo biopsy or surgery or had been stable for at least 2 y of ultrasound follow-up after benign biopsy results or typically benign findings. For each mass, representative long- and short-axis static ultrasound images were acquired; real-time long- and short-axis B-mode video images through the mass area were separately saved as cine clips. Each image was reviewed independently by five radiologists who were asked to classify ultrasound features according to the fifth edition of the BI-RADS lexicon. Inter-observer variability was assessed using kappa (κ) statistics. Diagnostic performance on static and video images was compared using the area under the receiver operating characteristic curve. No significant difference was found in κ values between static and video images for all descriptors, although κ values of video images were higher than those of static images for shape, orientation, margin and calcifications. After receiver operating characteristic curve analysis, the video images (0.83, range: 0.77-0.87) had higher areas under the curve than the static images (0.80, range: 0.75-0.83; p = 0.08). Inter-observer variability and diagnostic performance of video images was similar to that of static images on breast ultrasonography according to the new edition of BI-RADS.