Huang Hou-Chun, Chung Hsin-Hsiang, Yu Jia-Ying, Chen Bo-Rong, Wang Ming-Yang, Hsu Cheng-Chih
Department of Chemistry, National Taiwan University, Taipei, Taiwan, ROC.
Department of Surgery, National Taiwan University Hospital, Taipei, Taiwan, ROC.
Commun Med (Lond). 2025 Jul 1;5(1):259. doi: 10.1038/s43856-025-00930-7.
Conventional histopathological examination for breast core needle biopsy diagnosis is time-consuming and labor-intensive, leading to delayed medical treatments and increased psychological burden for patients. A rapid and reliable diagnostic method is needed to assist routine pathological diagnosis.
We developed a miniature mass spectrometry platform coupled with paper spray ionization (MiniMaP) for rapid breast cancer diagnosis. This platform enables direct molecular analysis of biopsy samples without sample preparation. A machine learning model was trained to differentiate benign and malignant samples based on molecular profiles. The platform's performance was further evaluated in a 22-month multicenter validation study.
Here we show that the machine learning model trained on molecular profiles achieves 88% accuracy in distinguishing breast cancer from benign samples. The model identifies 60 molecular features as potential biomarkers. Additionally, MiniMaP is implemented for on-site analysis in a hospital setting, enabling breast cancer diagnosis within 5 min. The platform maintains consistent accuracy (84%) across 540 biopsy samples over the 22-month validation period.
Our results demonstrate that the MiniMaP platform enables rapid breast cancer diagnosis and maintains consistent performance in long-term multicenter validation. It holds promise for assisting clinical breast cancer diagnosis by providing instant diagnostic reports to support timely medical decisions and improve medical care.
用于乳腺粗针活检诊断的传统组织病理学检查耗时且费力，导致医疗治疗延迟并增加患者的心理负担。需要一种快速且可靠的诊断方法来辅助常规病理诊断。
我们开发了一种结合纸喷雾电离的微型质谱平台（MiniMaP）用于快速乳腺癌诊断。该平台无需样品制备即可对活检样品进行直接分子分析。训练了一个机器学习模型，以根据分子谱区分良性和恶性样品。在一项为期22个月的多中心验证研究中进一步评估了该平台的性能。
我们在此表明，基于分子谱训练的机器学习模型在区分乳腺癌和良性样品方面的准确率达到88%。该模型识别出60种分子特征作为潜在生物标志物。此外，MiniMaP在医院环境中用于现场分析，能够在5分钟内完成乳腺癌诊断。在22个月的验证期内，该平台在540个活检样品中保持了一致的准确率（84%）。
我们的结果表明，MiniMaP平台能够实现快速乳腺癌诊断，并在长期多中心验证中保持一致的性能。它有望通过提供即时诊断报告来辅助临床乳腺癌诊断，以支持及时的医疗决策并改善医疗护理。
