基于影像组学结合机器学习在子宫内膜增生与子宫内膜癌鉴别中的研究进展

Research Progress of Radiomics Combined with Machine Learning in the Differentiation of Endometrial Hyperplasia and Endometrial Cancer

摘要: 子宫内膜癌(Endometrial Cancer, EC)是女性生殖系统当中常见的一种恶性肿瘤，发病率正呈现出持续不断上升的趋势，且发病年龄有年轻化倾向，这严重地威胁到了女性的生命健康。子宫内膜增生(Endometrial Hyperplasia, EH)属于子宫内膜癌的癌前病变，如何精准地区分子宫内膜增生和子宫内膜癌，并且对不同的患者开展个体化的治疗，到现在依然是当前妇科临床所面临的一个关键挑战。对于那些有生育需求或者需要进行个体化管理的患者来说，这一点至关重要。传统的诊断方法主要依赖于影像检查、诊断性刮宫以及宫腔镜检查活检，不过这些诊断手段存在一定的局限性，它们的敏感性、准确性和特异性有限。近年来，影像组学(Radiomics)和机器学习(Machine Learning, ML)技术的融合，为子宫内膜病变的无创、精准鉴别开辟了一条新的路径。影像组学能够从超声、磁共振成像(MRI)、计算机断层扫描(CT)以及正电子发射断层扫描(PET-CT)等医学图像当中，以高通量的方式提取定量特征，从而揭示出人眼无法识别的肿瘤异质性信息；而机器学习则是凭借构建预测模型，去挖掘这些特征和病理结局之间存在的复杂关联。本文系统地对影像组学结合机器学习在EH和EC鉴别诊断方面的研究进展进行了综述，内容囊括了它的基本技术流程、基于不同影像模态的应用现状，文章还分析了当前研究在标准化、可重复性以及临床转化这些方面所面临的挑战，并且对未来的发展方向进行了展望，以期能够为推动子宫内膜病变的精准、无创诊断提供理论依据和技术参考。

Abstract: Endometrial cancer (EC) is a common malignant tumor of the female reproductive system. Its incidence has been continuously rising, with a trend to younger starting age, bringing a big danger to women’s health. Endometrial hyperplasia (EH) is a pre-cancer illness of endometrial cancer. To distinguish accurately between endometrial hyperplasia and endometrial cancer, and to carry out patient-specific treatment for different persons, still is a very important difficult problem in present gynecological clinic work. This is especially vital for patients who have wish to bear children or who need person-targeted management. Traditional diagnosis methods mainly depend on image check, diagnostic scraping of uterus, and hysteroscope biopsy; however, these ways have inside limits, including not-best sensitivity, correctness, and specificity. In recent years, the combination of radiomics and machine learning (ML) has opened a new road for non-invasive and accurate distinction of endometrial lesions. Radiomics can carry out high-volume extraction of quantity features from medical images like ultrasound, magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography-computed tomography (PET-CT), therefore revealing information about tumor heterogeneity that human eyes cannot feel. Machine learning, through building predictive models, finds out the complex connections between these features and pathological results. Therefore, this article systematically reviews the research progress of radiomics combined with machine learning in the differential diagnosis of EH and EC. It includes the basic technical work processes, the current application situation among different image methods, and analyzes the difficulties met in present research about standardization, repeatability, and clinical translation. Hence, future development directions are also discussed, with the aim to provide a theory foundation and technology reference for pushing forward the accurate, non-invasive diagnosis of endometrial lesions.