人工智能辅助乳腺癌HER2 FISH判读的评估及可行性分析

Evaluation and feasibility analysis of artificial intelligence-assisted HER2 FISHinterpretation in breast cancer

  • 摘要:
    目的 评估自动化扫描和摄取系统辅助病理医师进行HER2 FISH判读的准确性和可行性。
    方法 通过FISH检测HER2的基因扩增情况,以“病理医师独立判读结果”为“金标准”,分析CytoVision*系统与人工识别判读结合的“人机对话结果”以及“系统判读结果”与“金标准”的一致性。
    结果 “人机对话结果”与“金标准”的一致率可达91%以上,且可以节省约50%的人工操作时间。每个细胞核HER2拷贝数被“低估”的倾向是造成“人机对话判读”低拷贝数扩增和部分HER2表达异质性病例灵敏度偏低的主要原因。
    结论 全自动FISH图像分析和摄取系统模拟了人工判读的过程,保证了细胞选择的随机性,提高工作效率。通过杂交区域的精准选择和人机对话,有望“代替”病理医师独立判读。

     

    Abstract:
    Objective To evaluate the accuracy and feasibility of an automated scanning and uptake system to assist pathologists with human epidermal growth factor receptor 2 (HER2) FISH interpretation.
    Methods HER2 gene amplification is detected using FISH, and “result interpretation by independent pathologists” is regarded as the “gold standard.” The consistency of “human-machine dialogue results” (use of a CytoVision* system combined with manual interpretation) and “CytoVision*-based automated interpretation” with the “gold standard” was assessed.
    Results Consistency between "human-machine dialogue results" and the "gold standard" can surpass 91%, with the former method saving up to 50% of the manual operation time. The tendency of each cell nucleus's HER2 copy number to be "underestimated" is the main reason for the low sensitivity observed in cases with low copy number amplification and HER2 heterogeneous expression cases in "human-machine dialogue interpretation."
    Conclusions Automatic FISH image analysis and uptake systems simulate the process of manually interpreted cell selection, ensure random cell selection, and improve work efficiency. With its accurate selection of the hybridization region and “human-computer dialogue,” the system is expected to “replace” interpretation by independent pathologists.

     

/

返回文章
返回