Objectives: To investigate whether multimodal intratumour and peritumour ultrasound features correlate with specific breast cancer molecular subtypes. Methods: From March to November 2021, a total of 85 patients with histologically proven breast cancer underwent B-mode, real-time elastography (RTE), colour Doppler flow imaging (CDFI) and contrast-enhanced ultrasound (CEUS). The time intensity curve (TIC) of CEUS was obtained, and the peak and time to peak (TTP) were analysed. Chi-square and binary logistic regression were used to analyse the connection between multimodal ultrasound imaging features and breast cancer molecular subtype. Results: Among 85 breast cancers, the subtypes were as follows: luminal A (36 cases, 42.4%), luminal B (20 cases, 23.5%), human epidermal growth factor receptor-2 positive (HER2+) (16 cases, 18.8%), and triple negative breast cancer (TNBC) (13 cases, 15.3%). Binary logistic regression models showed that RTE (P < 0.001) and CDFI (P = 0.036) were associated with the luminal A cancer subtype (C-index: 0.741), RTE (P = 0.016) and the peak ratio between intratumour and corpus mamma (P = 0.036) were related to the luminal B cancer subtype (C-index: 0.788). The peak ratio between peritumour and intratumour (P = 0.039) was related to the HER2 + cancer subtype (C-index: 0.859), and CDFI (P = 0.002) was associated with the TNBC subtype (C-index: 0.847). Conclusions: Multimodal ultrasound features could be powerful predictors of specific breast cancer molecular subtypes. The intra- and peritumour CEUS features play assignable roles in separating luminal B and HER2 + breast cancer subtypes.

Background: Continuous contrast-enhanced ultrasound (CEUS) video is a challenging direction for radiomics research. We aimed to evaluate machine learning (ML) approaches with radiomics combined with the XGBoost model and a convolutional neural network (CNN) for discriminating between benign and malignant lesions in CEUS videos with a duration of more than 1 min.Methods: We gathered breast CEUS videos of 109 benign and 81 malignant tumors from two centers. Radiomics combined with the XGBoost model and a CNN was used to classify the breast lesions on the CEUS videos. The lesions were manually segmented by one radiologist. Radiomics combined with the XGBoost model was conducted with a variety of data sampling methods. The CNN used pretrained 3D residual network (ResNet) models with 18, 34, 50, and 101 layers. The machine interpretations were compared with prospective interpretations by two radiologists. Breast biopsies or pathological examinations were used as the reference standard. Areas under the receiver operating curves (AUCs) were used to compare the diagnostic performance of the models.Results: The CNN model achieved the best AUC of 0.84 on the test cohort with the 3D-ResNet-50 model. The radiomics model obtained AUCs between 0.65 and 0.75. Radiologists 1 and 2 had AUCs of 0.75 and 0.70, respectively.Conclusions: The 3D-ResNet-50 model was superior to the radiomics combined with the XGBoost model in classifying enhanced lesions as benign or malignant on CEUS videos. The CNN model was superior to the radiologists, and the radiomics model performance was close to the performance of the radiologists.

Objective: To investigate the ability of contrast-enhanced ultrasound (CEUS)-based radiomics combined with machine learning to detect early protein changes after incomplete thermal ablation.Methods: HCT-26 colorectal adenoma cells were engrafted into the livers of 80 mice, which were randomly divided into 4 groups for palliative laser ablation. Changes in heat shock protein (HSP) and apoptosis-related protein expression in the tumors were assessed. SCID mice subjected to CEUS and ultrasonography were divided into training (n=56) and test (n=24) datasets. Then, 102 features from seven feature groups were extracted. We use the least absolute shrinkage and selection operator (LASSO) feature selection method to fit the machine learning classifiers. The feature selection methods and four classifiers were combined to determine the best prediction model.Results: The areas under the receiver-operating characteristic curves (AUCs) of the classifiers in the test dataset ranged from 0.450 to 0.932 (median: 0.721). The best score was obtained from the model in which the omics data of CEUS was analyzed in the arterial phase by random forest (RF) classification.Conclusions: A machine learning model, in which radiomics characteristics are extracted by multimodal ultrasonography, can accurately, rapidly and noninvasively identify protein changes after ablation.

Background: Based on the conventional ultrasound images of thyroid nodules, contrast-enhanced ultrasound (CEUS) videos were analyzed to investigate whether CEUS improves the classification accuracy of benign and malignant thyroid nodules using machine learning (ML) radiomics and compared with radiologists.Materials and methods: The B-mode ultrasound (B-US), real-time elastography (RTE), color doppler flow images (CDFI) and CEUS cines of patients from two centers were retrospectively gathered. Then, the region of interest (ROI) was delineated to extract the radiomics features. Seven ML algorithms combined with four kinds of radiomics data (B-US, B-US + CDFI + RTE, CEUS, and B-US + CDFI + RTE + CEUS) were applied to establish 28 models. The diagnostic performance of ML models was compared with interpretations from expert and nonexpert readers.Results: A total of 181 thyroid nodules from 181 patients of 64 men (mean age, 42 years +/- 12) and 117 women (mean age, 46 years +/- 12) were included. Adaptive boosting (AdaBoost) achieved the highest area under the receiver operating characteristic curve (AUC) of 0.89 in the test set among 28 models when combined with B-US + CDFI + RTE + CEUS data and an AUC of 0.72 and 0.66 when combined with B-US and B-US + CDFI + RTE data. The AUC achieved by senior and junior radiologists was 0.78 versus (vs.) 0.69 (p > 0.05), 0.79 vs. 0.64 (p < 0.05), and 0.88 vs. 0.69 (p < 0.05) combined with B-US, B-US+CDFI+RTE and B-US+CDFI+RTE+CEUS, respectively.Conclusion: With the addition of CEUS, the diagnostic performance was enhanced for all seven classifiers and senior radiologists based on conventional ultrasound images, while no enhancement was observed for junior radiologists. The diagnostic performance of ML models was similar to senior radiologists, but superior to those junior radiologists.

[This corrects the article DOI: 10.3389/fonc.2022.951973.].