Nomogram Based on Super-Resolution Ultrasound Images Outperforms in Predicting Benign and Malignant Breast Lesions

Abstract

Objective

To establish a good predictive model using a deep-learning (DL)-based three-dimensional (3D) super-resolution ultrasound images for the diagnosis of benign and malignant breast lesions.

Methods

This retrospective study included 333 patients with histopathologically confirmed breast lesions, randomly split into training (N=266) and testing (N=67) datasets. Eight models, including four deep learning models (ORResNet101, ORMobileNet_v2, SRResNet101, SRMobileNet_v2) and four machine learning models (OR_LR, OR_SVM, SR_LR, SR_SVM), were developed based on original and super-resolution images. The best performing model was SRMobileNet_v2, which was used to construct a nomogram integrating clinical factors. The performance of nomogram was evaluated using receiver operating characteristic (ROC) analysis, decision curve analysis (DCA), and calibration curves.

Results

SRMobileNet_v2, MobileNet_V2 based on super-resolution ultrasound images, had the best predictive performance in four traditional machine learning models and four deep learning models, with AUC improvements of 0.089 and 0.031 in the training and testing sets, relative to the ORMobileNet_v2 model based on original ultrasound images. The deep-learning nomogram was constructed using the SRMobileNet_v2 model score, tumor size, and patient age, resulting in superior predictive efficacy compared to the nomogram without the SRMobileNet_v2 model score. Furthermore, it demonstrated favorable calibration, discrimination, and clinical utility in both cohorts.

Conclusion

The diagnostic prediction model utilizing super-resolution reconstructed ultrasound images outperforms the model based on original images in distinguishing between benign and malignant breast lesions. The nomogram based on super-resolution ultrasound images has the potential to serve as a reliable auxiliary diagnostic tool for clinicians, exhibiting superior predictive performance in distinguishing between benign and malignant breast lesions.

Keywords:

• deep learning
• radiomics
• machine learning
• breast lesion
• super-resolution

Abbreviations

OR, Original images; SR, Super-Resolution.

Data Confidentiality Statement

This study complied with ethical standards, and all patient data was anonymized and properly protected, including encrypted storage of patient information, strict control of access, and timely destruction of unnecessary information.

Data Sharing Statement

The datasets utilized and analyzed in this study are not publicly available due to patient privacy requirements and ethical restriction.

Ethical Approval and Informed Consent Statement

Ethical approvals for the study were obtained from the Institutional Review Boards of the First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital (YXLL-KY-2023(045)). Patient consent was waived due to the retrospective nature of the study and the analysis used anonymous clinical data. The study was conducted according to the guidelines of the Declaration of Helsinki (2013 revision).

Disclosure

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Additional information

Funding

This study was supported by the Provincial Key Research and Development Fund of Shandong Province, China (Grant #:2016GSF201141).