Lesion classification study details.

Breast cancer screening programs using mammography have led to significant mortality reduction in high-income countries. However, many low- and middle-income countries lack resources for mammographic screening. Handheld breast ultrasound (BUS) is a low-cost alternative but requires substantial training. Artificial intelligence (AI) enabled BUS may aid in both the detection and classification of breast cancer, enabling screening use in low-resource contexts. The purpose of this systematic review is to investigate whether AI-enhanced BUS is sufficiently accurate to serve as the primary modality in screening, particularly in resource-limited environments. This review (CRD42023493053) is reported in accordance with the PRISMA guidelines. Evidence synthesis is reported in accordance with the SWiM (Synthesis Without Meta-analysis) guidelines. PubMed and Google Scholar were searched from January 1, 2016 to December 12, 2023. Studies are grouped according to AI task and assessed for quality. Of 763 candidate studies, 314 full texts were reviewed and 34 studies are included. The AI tasks of included studies are as follows: 1 frame selection, 6 lesion detection, 11 segmentation, and 16 classification. 79% of studies were at high or unclear risk of bias. Exemplary classification and segmentation AI systems perform with 0.976 AUROC and 0.838 Dice similarity coefficient. There has been encouraging development of AI for BUS. However, despite studies demonstrating high performance, substantial further research is required to validate reported performance in real-world screening programs. High-quality model validation on geographically external, screening datasets will be key to realizing the potential for AI-enhanced BUS in increasing screening access in resource-limited environments.

在高收入国家，采用乳腺钼靶摄影的乳腺癌筛查项目已显著降低了乳腺癌死亡率。然而，许多中低收入国家缺乏开展乳腺钼靶筛查的相关资源。手持式乳腺超声（Handheld Breast Ultrasound, BUS）是一种低成本替代方案，但需要大量专业培训。人工智能（Artificial Intelligence, AI）赋能的手持式乳腺超声可辅助乳腺癌的检测与分类，从而使其能够在资源匮乏的场景中用于筛查。本系统综述的目的在于探究人工智能赋能的手持式乳腺超声是否具备足够高的准确率，可作为筛查的主要手段，尤其适用于资源受限的环境。本综述（注册号CRD42023493053）的撰写遵循PRISMA报告规范，证据合成部分则遵循SWiM（Synthesis Without Meta-analysis）指南。本研究检索了2016年1月1日至2023年12月12日期间PubMed与Google Scholar数据库中的文献。研究根据其所涉及的人工智能任务进行分组，并对其研究质量进行评估。在763篇候选文献中，共筛选出314篇全文文献进行审阅，最终纳入34项研究。纳入研究涉及的人工智能任务如下：1项为帧选择，6项为病灶检测，11项为图像分割，16项为分类任务。其中79%的研究存在较高或不明确的偏倚风险。表现优异的分类与分割人工智能系统的受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic Curve, AUROC）可达0.976，戴斯相似系数（Dice Similarity Coefficient）可达0.838。手持式乳腺超声相关人工智能技术已取得令人鼓舞的发展。然而，尽管已有研究证明该技术具备较高性能，但仍需开展大量后续研究，以在真实世界的筛查项目中验证其报告的性能表现。在地理异质性的筛查数据集上开展高质量的模型验证，将是实现人工智能赋能的手持式乳腺超声在资源受限环境中提升筛查可及性的关键。