Scholarlink

Purpose <p>INI1 (Integrase interactor 1) deficiency (defined as loss of nuclear INI1 expression on immunohistochemistry) identifies an aggressive subtype of sinonasal undifferentiated carcinoma (SNUC) associated with poor prognosis. This study aimed to develop and validate an interpretable clinical-radiomics fusion model for noninvasively prediction of INI1 deficiency and assess its prognostic value.</p> Methods <p>In this two-center retrospective study, 183 SNUC patients (training: <i>n</i> = 140; test: <i>n</i> = 43) were enrolled. Clinical variables were analyzed via univariate and multivariate analyses to identify independent predictors. Radiomic features were extracted from T2-weighted imaging (T2WI), contrast-enhanced T1-weighted imaging (CEI), and apparent diffusion coefficient (ADC) maps, with feature selection (variance thresholding, mRMR, RFE) applied. Using the AutoGluon-Tabular framework, clinical, radiomics, and fusion models were developed, validated, and evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) and calibration. SHAP analysis assessed interpretability.</p> Results <p>Tumor location and nasal cavity involvement were two independent clinical predictors. Feature selection identified two key morphological traits: Flatness on CEI and LeastAxisLength on ADC. The fusion model outperformed the clinical model (<i>p</i> = 0.046) with AUC = 0.993 (sensitivity = 96.97%, specificity = 93.46%) in training and AUC = 0.928 (sensitivity = 100%, specificity = 79.41%) in test cohort. SHAP analysis highlighted CE-T1WI Flatness as the top predictor. High- and low-risk groups showed significantly different 2-year disease-free survival (training: 23.1% vs. 79.2%, <i>p</i> = 0.013; test:10.0% vs.73.9%, <i>p</i> = 0.028).</p> Conclusion <p>The interpretable clinical-radiomics fusion model showed promising performance in noninvasively predicting INI1 deficiency in SNUC and also provided prognostic insights, potentially guiding personalized treatment strategies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Interpretable clinical-radiomics fusion model for noninvasive prediction of INI1 deficiency and prognosis in sinonasal undifferentiated carcinoma (SNUC): a two-center study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月13日

查看全文
  • Naier Lin,
  • Xingcan Li,
  • Min Ye,
  • Yue Geng,
  • Yuxin He,
  • Yan Sha,
  • Zuohua Tang

摘要

Purpose

INI1 (Integrase interactor 1) deficiency (defined as loss of nuclear INI1 expression on immunohistochemistry) identifies an aggressive subtype of sinonasal undifferentiated carcinoma (SNUC) associated with poor prognosis. This study aimed to develop and validate an interpretable clinical-radiomics fusion model for noninvasively prediction of INI1 deficiency and assess its prognostic value.

Methods

In this two-center retrospective study, 183 SNUC patients (training: n = 140; test: n = 43) were enrolled. Clinical variables were analyzed via univariate and multivariate analyses to identify independent predictors. Radiomic features were extracted from T2-weighted imaging (T2WI), contrast-enhanced T1-weighted imaging (CEI), and apparent diffusion coefficient (ADC) maps, with feature selection (variance thresholding, mRMR, RFE) applied. Using the AutoGluon-Tabular framework, clinical, radiomics, and fusion models were developed, validated, and evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) and calibration. SHAP analysis assessed interpretability.

Results

Tumor location and nasal cavity involvement were two independent clinical predictors. Feature selection identified two key morphological traits: Flatness on CEI and LeastAxisLength on ADC. The fusion model outperformed the clinical model (p = 0.046) with AUC = 0.993 (sensitivity = 96.97%, specificity = 93.46%) in training and AUC = 0.928 (sensitivity = 100%, specificity = 79.41%) in test cohort. SHAP analysis highlighted CE-T1WI Flatness as the top predictor. High- and low-risk groups showed significantly different 2-year disease-free survival (training: 23.1% vs. 79.2%, p = 0.013; test:10.0% vs.73.9%, p = 0.028).

Conclusion

The interpretable clinical-radiomics fusion model showed promising performance in noninvasively predicting INI1 deficiency in SNUC and also provided prognostic insights, potentially guiding personalized treatment strategies.