This study aims to build deep learning-based radiomic methods in differentiating vessel invasion from non-vessel invasion in cervical cancer with multi-parametric MRI data. A set of 1,070 dynamic T1 contrast-enhanced (DCE-T1) and 986 T2 weighted imaging (T2WI) MRI images from 167 early-stage cervical cancer patients (January 2014 - August 2018) were used to train and validate deep learning models. Predictive performances were evaluated using receiver operating characteristic (ROC) curve and confusion matrix analysis, with the DCE-T1 showing more discriminative results than T2WI MRI. By adopting an attention ensemble learning strategy that integrates both MRI sequences, the highest average area was obtained under the ROC curve (AUC) of 0.911 (Sensitivity = 0.881 and Specificity = 0.752). The superior performances in this study, when compared to existing radiomic methods, indicate that a wealth of deep learning-based radiomics could be developed to aid radiologists in preoperatively predicting vessel invasion in cervical cancer patients.

This retrospective study aims to evaluate the use of multi-parametric magnetic resonance imaging (MRI) in predicting lymph-vascular space invasion (LVSI) in early-stage cervical cancer using radiomics methods. A total of 163 patients who underwent contrast-enhanced T1-weighted (CE T1W) and T2-weighted (T2W) MRI scans at 3.0T were enrolled between January 2014 and September 2019. Radiomics features were extracted and selected from the tumoral and peritumoral regions at different dilation distances outside the tumor. Mann–Whitney U test, the least absolute shrinkage and selection operator logistic regression, and logistic regression was applied to select the predictive features and develop the radiomics signature. Univariate analysis was performed on the clinical characteristics. The radiomics nomogram was constructed incorporating the radiomics signature and the selected important clinical predictor. Prediction performance of the radiomics signature, clinical model, and nomogram was evaluated with the area under the curve (AUC), specificity, sensitivity, calibration, and decision curve analysis (DCA). A total of 5 features that were selected from the peritumoral regions with 3- and 7-mm dilation distances outside tumors in CE T1W and T2W MRI, respectively, showed optimal discriminative performance. The radiomics signature comprising the selected features was significantly associated with the LVSI status. The radiomics nomogram integrating the radiomics signature and degree of cellular differentiation exhibited the best predictability with AUCs of 0.771 (specificity (SPE) = 0.831 and sensitivity (SEN) = 0.581) in the training cohort and 0.788 (SPE = 0.727, SEN = 0.773) in the validation cohort. DCA confirmed the clinical usefulness of our model. Our results illustrate that the radiomics nomogram based on MRI features from peritumoral regions and the degree of cellular differentiation can be used as a noninvasive tool for predicting LVSI in cervical cancer.

Purpose: Metastasis is one of the most important causes of treatment failure in nasopharyngeal carcinoma (NPC). In T4 or N2-3 patients at high-risk of metastasis, concurrent chemoradiotherapy (CCRT) is inadequate and additional induction chemotherapy (IC) is controversial. There is a critical need to develop a better patient stratification to efficiently identify patients at high-risk of metastasis for additional IC. Recently, Sequestosome 1 (SQSTM1)/p62, an autophagy adaptor protein, was identified as one of the metastasis-related proteins in NPC. However, the mechanism by which SQSTM1 is involved in NPC metastasis was not investigated. Experimental Design: The effect of SQSTM1 on cell migration and invasion was examined in vitro and in vivo. SQSTM1 expression was analyzed in clinical NPC samples using IHC. Luciferase reporter analyses were conducted to identify the effects of SQSTM1 on NF-κB transcriptional activity. A prediction bio-model was constructed by Cox analysis. Retrospective and prospective randomized clinical data were adopted to build and test the model, respectively. Results: SQSTM1 mediated epithelial to mesenchymal transition (EMT) through the NF-κB pathway to promote NPC metastasis. Inhibiting SQSTM1 enhanced sensitivity to cisplatin in NPC cells. In NPC patients, high SQSTM1 expression was associated with increased risk of distant metastasis. Furthermore, we propose a prognostic bio-model based on SQSTM1 and N-stage to predict NPC metastasis. Most importantly, our prospective randomized study suggested that IC is beneficial for NPC patients with high metastasis risk. Conclusions: The prognostic bio-model identifies NPC patients at high-risk of metastasis for additional IC. Clin Cancer Res; 24(3); 648–58. ©2017 AACR.