Abstract:Purpose: To develop and validate an applicable deep learning radiomic model based on breast ultrasound image features for differentiating malignancy within Breast Imaging Reporting and Data System (BI-RADS) 3 and 4A breast lesions, among which the majority are under low risk of malignancy.Method: Ultrasound images captured by two independent institutions between January 1, 2017, and January 1, 2022, were obtained from 955 patients with breast masses reported as the BI-RADS 3 and 4A category, 706 of which were benign and 249 of which were malignant, as confirmed by pathological examination. In total, 6,286 ultrasound images (3,965 benign and 2,321 malignant) were obtained. The data were randomly divided into a training set, a validation set, and a test set in the ratio of 7:2:1. A deep learning model was constructed using ultrasound image features extracted by the Densenet-121 model pre-trained on ImageNet.Results: The deep learning radiomics model exhibited promising accuracy for malignancy prediction. The area under the receiver operating characteristic (ROC) curve (AUC) of the deep learning model was 0.998 (accuracy = 99.98%, precision = 99.98%, F1-score = 99.98%) for the training cohort and 0.869 (accuracy = 88.84%, precision = 88.78%, F1-score = 88.84%) for the test cohort. More importantly, the model overperformed seven radiologists in interpreting BI-RADS 3-4A lesions. Subgroup analyses quantified the predictive performance for the two hospitals.Interpretation: The deep learning radiomic model has considerable predictive value in differentiating malignant versus benign lesions in BI-RADS 3-4A ultrasound images, thus could avoid unnecessary biopsy within this population. The model development methods can be generalized to other radiological modalities to complement early screening for breast cancer.Funding: This research was funded by "the Fundamental Research Funds for the Central Universities N2219001", "Ningbo Science and Technology Bureau (Grant No. 2021Z027)" and "the Fundamental Research Funds for the Central Universities N2224001-10".Declaration of Interest: The authors have declared that no competing interest exists.Ethical Approval: This retrospective study was approved by the ethical review committee of Shengjing Hospital of China Medical University and the First Affiliated Hospital of China Medical University.

Ultrasonic Spectral Information Promotes Specificity and Visualization of Breast Cancer in Deep Learning

Improving the diagnosis of breast cancer based on biophysical ultrasound features utilizing machine learning

Feature analysis of ultrasound breast tumor images

Breast ultrasound lesion classification based on image decomposition and transfer learning

Deep learning and ultrasound feature fusion model predicts the malignancy of complex cystic and solid breast nodules with color Doppler images

Dual-Intended Deep Learning Model for Breast Cancer Diagnosis in Ultrasound Imaging

Breast Ultrasound-Based Deep Learning Radiomics Model for Differentiating Malignancy in Low Malignant Risk Lesions

Deep learning algorithm using bispectrum analysis energy feature maps based on ultrasound radiofrequency signals to detect breast cancer

Breast Tumor Detection In Ultrasound Images Using Deep Learning

Breast Tumor Classification in Ultrasound Images by Fusion of Deep Convolutional Neural Network and Shallow LBP Feature

Automatic Identification of Breast Ultrasound Image Based on Supervised Block-Based Region Segmentation Algorithm and Features Combination Migration Deep Learning Model

Tomographic Ultrasound Imaging in the Diagnosis of Breast Tumors under the Guidance of Deep Learning Algorithms

Spectral analysis enhanced net (SAE-Net) to classify breast lesions with BI-RADS category 4 or higher

Research on Diagnosis of Breast Cancer Based on Ultrasonic Radio Frequency Signals

Multimodal microscopic imaging with deep learning for highly effective diagnosis of breast cancer

A domain knowledge-based interpretable deep learning system for improving clinical breast ultrasound diagnosis

Evaluating the Accuracy of Breast Cancer and Molecular Subtype Diagnosis by Ultrasound Image Deep Learning Model

Domain Knowledge Powered Deep Learning for Breast Cancer Diagnosis Based on Contrast-Enhanced Ultrasound Videos

Diagnostic Performance of Deep Learning in Video-Based Ultrasonography for Breast Cancer: A Retrospective Multicentre Study

The Clinical Application of Artificial Intelligence Assisted Contrast-Enhanced Ultrasound on BI-RADS Category 4 Breast Lesions.

Medical Breast Ultrasound Image Segmentation by Machine Learning