第一部分

上呼吸消化道结外鼻型NK/T细胞淋巴瘤多模态磁共振影像组学预后预测价值

研究目的：构建基于多模态头颈部磁共振的影像组学模型，及影像组学与临床因素相结合的预后预测模型，预测上呼吸消化道结外鼻型NK/T细胞淋巴瘤患者总生存率（overall survival, OS）。

材料与方法：共176例上呼吸消化道结外鼻型NK/T细胞淋巴瘤患者纳入分析，治疗前头颈部MRI图像完整。应用无监督学习谱聚类方法划分影像组学特征及患者分类，比较各组OS。使用Cox比例风险模型和Spearman等级相关性分析明确影像组学特征与OS及临床因素之间的关系。将影像组学结果与nomogram风险指数(nomogram-revised risk index, NRI)进行整合，根据预后因素相应的风险比，构建Nomogram风险指数+多模态磁共振影像组学（nomogram-revised risk index + multi-modal magnetic resonance imaging radiomics signature, NRI-M）模型。对模型进行内部验证，并通过评估模型区分度、校准度和决策曲线与NRI进行比较。

研究结果：影像组学分类的2组患者OS存在显著差异，全组患者5年OS为87.2% vs.67.3%（P = 0.002），早期患者为88.8% vs.69.2%（P = 0.003）。影像组学分类和原发肿瘤浸润之间具有强相关性。在所有患者中，NRI-M的曲线下面积、Harrell’s C指数、绝对净重新分类指数和Brier得分分别为0.748、0.740、13.07%和0.142，在早期患者中分别为0.717、0.729、13.86%和0.140。与影像组学和NRI相比，NRI-M表现出更好的区分度和校准度。而且，NRI-M比NRI提高了预测效能改善度。NRI-M的决策曲线分析也显示了较好的结果。

结论：综合影像组学和临床因素得出的NRI-M模型具有良好的预测上呼吸消化道结外鼻型NK/T细胞淋巴瘤患者预后的效能，可能有助于临床试验设计和改善临床决策。

第二部分

上呼吸消化道外结外鼻型NK/T细胞淋巴瘤

临床特征和预后

研究目的：探讨上呼吸消化道外结外鼻型NK/T细胞淋巴瘤临床特征和预后。

材料与方法：2001年11月至2015年12月中国淋巴瘤协作组(China Lymphoma Collaborative Group, CLCG)数据库中，159例原发上呼吸消化道外结外鼻型NK/T细胞淋巴瘤，分析其临床病理特征、治疗和预后。103例（64.8%）接受单纯化疗，56例接受放疗/手术±化疗。采用Kaplan-Meier法、Logrank检验进行生存分析。

研究结果：上呼吸消化道外结外鼻型NK/T细胞淋巴瘤常原发于皮肤软组织（66.7%）和胃肠道（19.5%）。Ann Arbor III-IV期和乳酸脱氢酶升高患者比例较高，分别为64.2%和47.8%。全组患者的3年总生存率为43.6%，3年无进展生存率为27.9%。原发皮肤软组织和胃肠道患者的3年总生存率分别为41.0%和59.4%（P = 0.281），3年无进展生存率分别为24.8%和48.3%（P = 0.109），差异均无统计学差异。综合治疗较单纯化疗提高了全组患者的3年总生存率（58.4% vs 33.9%，P = 0.001）和3年无进展生存率（40.7% vs 20.7%，P = 0.008）。57例I-II期患者中，放疗/手术±化疗组(n=29)和单纯化疗组(n=28)患者的3年总生存率分别为71.3%和61.3%（P = 0.051），3年无进展生存率分别为46.6%和40.8%（P = 0.102），差异无统计学差异可能和例数少有关。失败模式以远处结外器官复发进展为主。

结论：上呼吸消化道外结外鼻型NK/T细胞淋巴瘤具有独特的临床特征和较差的预后，综合治疗较单纯化疗可能提高上呼吸消化道外ENKTCL疗效。

Part I: Evaluation of prognostic effect of radiomics based on multi-modal magnetic resonance imaging for extranodal nasal-type NK/T-cell lymphoma of the upper aerodigestive tract

Purpose: To develop a radiomics signature based on multi-modal magnetic resonance imaging (MRI) of the head and neck for estimating overall survival (OS) in patients with extranodal nasal-type NK/T-cell lymphoma (ENKTCL) of the upper aerodigestive tract (UADT) and to establish a radiomics-based model that incorporates the radiomics signature and clinical findings.

Patients and Methods: We identified 176 patients with UADT ENKTCL who underwent MRI of the head and neck before treatment. An unsupervised spectral clustering method was used to identify groups of patients and radiomic features. The OS was compared between two groups. Multivariate Cox proportional hazards model and Spearman rank correlation analysis were used to determine the association of the radiomics signature with OS and clinical factors. A combined radiomics-based model (nomogram-revised risk index plus multi-modal magnetic resonance imaging radiomics signature, NRI-M) was developed by incorporating the radiomics signature with nomogram-revised risk index (NRI). The radiomics signature and combined radiomics-based model were internally validated and compared with NRI by examining discrimination, calibration, and decision curve.

Results: Differences in OS were found between 2 groups of the developed radiomics signature. The 5-y OS rates between the two groups were 87.2% vs.67.3% (P = 0.002) in all patients and 88.8% vs. 69.2% (P = 0.003) in early-stage patients. A significant correlation was observed between the radiomics signature and primary tumor invasion. The area under the curve (AUC), Harrell’s C index, absolute net reclassification index and integrated Brier score of NRI-M were 0.748, 0.740, 13.07%, and 0.142, respectively, for all patients, and 0.717, 0.729, 13.86%, and 0.140, respectively, for early-stage patients. Compared with either radiomics signature or NRI, NRI-M showed the best discrimination and calibration. Moreover, NRI-M improved net reclassification index than NRI. Decision curve analysis of the NRI-M also showed a good outcome.

Conclusions: The NRI-M model derived from radiomics signature and clinical factors has good performance for predicting the prognosis in UADT ENKTCL and may help to design clinical trials and improve clinical decision making.

Part II: Clinical features and treatment outcome of extranodal nasal-type NK/T-cell lymphoma of the extra-upper aerodigestive tract

Purpose: To investigate the clinical features and prognosis of extranodal nasal-type NK/T-cell lymphoma (ENKTCL) of the extra-upper aerodigestive tract (extra-UADT ENKTCL).

Patients and Methods: A total of 159 patients with extra-UADT ENKTCL from the China Lymphoma Collaborative Group (CLCG) database between November 2001 and December 2015 were retrospectively analyzed. One hundred and three patients received chemotherapy alone, whereas 56 patients received radiotherapy/surgery with or without chemotherapy. Kaplan-Meier survival analyses and Logrank test were used to calculate actuarial survival.

Results: Extra-UADT ENKTCL commonly occurs in skin and soft tissues (66.7%) and the gastrointestinal tract (19.5%). According to the Ann Arbor staging system, 102 (64.2%) patients had advanced-stage disease. Elevated lactate dehydrogenase (LDH) level was observed in 47.8% of patients. The 3-year overall survival (OS) and progression-free survival (PFS) rates were 43.6% and 27.9%, respectively. The corresponding OS and PFS rates were 41.0% and 24.8% for primary skin/soft tissue sites, and 59.4% (P = 0.281) and 48.3% (P = 0.109) for primary gastrointestinal tract sites, respectively. For all patients, combined modality treatment improved the 3-year OS (58.4% vs 33.9%, P = 0.001) and 3-year PFS (40.7% vs 20.7%, P = 0.008), compared with chemotherapy alone. For the 57 patients with stage I and II diseases, the 3-year OS and PFS rates, respectively, were 69.4% and 43.8% for radiotherapy ± chemotherapy, compared with 63.0% (P = 0.099) and 43.2% (P = 0.277) for chemotherapy alone. No significant differences between two treatment groups may be due to the small sample of patients. The distant extranodal dissemination was the primary failure patterns.

Conclusions: Extra-UADT ENKTCL appears to have distinct clinical characteristics and poor outcomes. Compared with chemotherapy alone, combined modality treatment may improve the prognosis of patients with extra-UADT ENKTCL.