METHODS: This single institutional retrospective review included a total of 143 patients with IP and 75 patients with IP with malignant transformation to squamous cell carcinoma (IP-SCC). All patients underwent surgical pathology and had preoperative magnetic resonance imaging (MRI) and computed tomography (CT) sinus studies between June 2014 and February 2022. Radiomics features were extracted from contrast-enhanced T1-weighted images (CE-T1WI), T2-weighted images (T2WI), and apparent diffusion coefficient (ADC) maps. The least absolute shrinkage and selection operator (LASSO) were performed to select the features extracted from the sequences mentioned above. Independent clinical risk factors were identified by multivariate logistic regression analysis. Radiomics nomogram was constructed by incorporating independent clinical risk factors and radiomics signature. Based on discrimination and calibration, the diagnostic performance of the nomogram was evaluated.
RESULTS: Twelve radiomics features were selected to develop the radiomics model with an area under the curve (AUC) of 0.987 and 0.989, respectively. Epistaxis (p=0.011), T2 equal signal (p=0.003), extranasal invasion (p<0.001), and loss of convoluted cerebriform pattern (p=0.002) were identified as independent clinical predictors. The radiomics nomogram model showed excellent calibration and discrimination (AUC: 0.993, 95% confidence interval [CI]: 0.985-1.00 and 0.990, 95% CI: 0.974-1.00) in the training and validation sets, respectively.
CONCLUSIONS: The nomogram that the combined radiomics signature and clinical risk factors showed a satisfactory ability to predict IP-SCC.
方法:这项单一的机构回顾性研究包括143例IP患者和75例IP患者恶性转化为鳞状细胞癌(IP-SCC)。在2014年6月至2022年2月期间,所有患者均接受了手术病理,并进行了术前磁共振成像(MRI)和计算机断层扫描(CT)鼻窦研究。从对比增强的T1加权图像(CE-T1WI)中提取影像组学特征,T2加权图像(T2WI),和表观扩散系数(ADC)图。执行最小绝对收缩和选择算子(LASSO)以选择从上述序列提取的特征。通过多因素logistic回归分析确定独立的临床危险因素。通过纳入独立的临床风险因素和影像组学特征来构建影像组学列线图。基于辨别和校准,评估了列线图的诊断性能.
结果:选择了十二个影像组学特征来开发曲线下面积(AUC)分别为0.987和0.989的影像组学模型。鼻出血(p=0.011),T2等信号(p=0.003),鼻外浸润(p<0.001),和卷积脑型模式的丧失(p=0.002)被确定为独立的临床预测因子。在训练集和验证集中,放射组学列线图模型显示出出色的校准和区分(AUC:0.993,95%置信区间[CI]:0.985-1.00和0.990,95%CI:0.974-1.00),分别。
结论:联合影像组学特征和临床危险因素的列线图显示出令人满意的预测IP-SCC的能力。