Wong Charles M, Kezlarian Brie E, Lin Oscar
Memorial Sloan-Kettering Cancer Center, New York, NY, USA.
J Pathol Inform. 2023 Apr 1;14:100309. doi: 10.1016/j.jpi.2023.100309. eCollection 2023.
The implementation of Digital Pathology has allowed the development of computational Pathology. Digital image-based applications that have received FDA Breakthrough Device Designation have been primarily focused on tissue specimens. The development of Artificial Intelligence-assisted algorithms using Cytology digital images has been much more limited due to technical challenges and a lack of optimized scanners for Cytology specimens. Despite the challenges in scanning whole slide images of cytology specimens, there have been many studies evaluating CP to create decision-support tools in Cytopathology. Among different Cytology specimens, thyroid fine needle aspiration biopsy (FNAB) specimens have one of the greatest potentials to benefit from machine learning algorithms (MLA) derived from digital images. Several authors have evaluated different machine learning algorithms focused on thyroid cytology in the past few years. The results are promising. The algorithms have mostly shown increased accuracy in the diagnosis and classification of thyroid cytology specimens. They have brought new insights and demonstrated the potential for improving future cytopathology workflow efficiency and accuracy. However, many issues still need to be addressed to further build on and improve current MLA models and their applications. To optimally train and validate MLA for thyroid cytology specimens, larger datasets obtained from multiple institutions are needed. MLAs hold great potential in improving thyroid cancer diagnostic speed and accuracy that will lead to improvements in patient management.
数字病理学的实施推动了计算病理学的发展。获得美国食品药品监督管理局（FDA）突破性设备认定的基于数字图像的应用主要集中在组织标本上。由于技术挑战以及缺乏针对细胞学标本的优化扫描仪，利用细胞学数字图像开发人工智能辅助算法的进展更为有限。尽管在扫描细胞学标本的全玻片图像方面存在挑战，但已有许多研究评估计算病理学以创建细胞病理学中的决策支持工具。在不同的细胞学标本中，甲状腺细针穿刺活检（FNAB）标本最有潜力受益于从数字图像衍生的机器学习算法（MLA）。在过去几年中，几位作者评估了专注于甲状腺细胞学的不同机器学习算法。结果很有前景。这些算法大多在甲状腺细胞学标本的诊断和分类中显示出更高的准确性。它们带来了新的见解，并展示了提高未来细胞病理学工作流程效率和准确性的潜力。然而，仍有许多问题需要解决，以进一步完善和改进当前的MLA模型及其应用。为了对甲状腺细胞学标本的MLA进行最佳训练和验证，需要从多个机构获得更大的数据集。MLA在提高甲状腺癌诊断速度和准确性方面具有巨大潜力，这将改善患者管理。
